diff --git a/.circleci/config.yml b/.circleci/config.yml index e171759f1c4..6c7bbddb9f1 100644 --- a/.circleci/config.yml +++ b/.circleci/config.yml @@ -112,6 +112,24 @@ jobs: python -m mypy . cd .. no_output_timeout: 10m + + semgrep: + docker: + - image: cimg/python:3.12 + auth: + username: ${DOCKERHUB_USERNAME} + password: ${DOCKERHUB_PASSWORD} + working_directory: ~/project + steps: + - checkout + - setup_google_dns + - run: + name: Install Semgrep + command: pip install semgrep + - run: + name: Run Semgrep (custom rules only) + command: semgrep scan --config .semgrep/rules . --error + local_testing_part1: docker: - image: cimg/python:3.12 @@ -2277,6 +2295,7 @@ jobs: - run: python ./tests/code_coverage_tests/router_code_coverage.py - run: python ./tests/code_coverage_tests/test_chat_completion_imports.py - run: python ./tests/code_coverage_tests/info_log_check.py + - run: python ./tests/code_coverage_tests/check_guardrail_apply_decorator.py - run: python ./tests/code_coverage_tests/test_ban_set_verbose.py - run: python ./tests/code_coverage_tests/code_qa_check_tests.py - run: python ./tests/code_coverage_tests/check_get_model_cost_key_performance.py @@ -3801,7 +3820,6 @@ jobs: - run: name: Get new version command: | - cd litellm-proxy-extras NEW_VERSION=$(python -c "import toml; print(toml.load('pyproject.toml')['tool']['poetry']['version'])") echo "export NEW_VERSION=$NEW_VERSION" >> $BASH_ENV @@ -3826,7 +3844,6 @@ jobs: - run: name: Publish to PyPI command: | - cd litellm-proxy-extras echo -e "[pypi]\nusername = $PYPI_PUBLISH_USERNAME\npassword = $PYPI_PUBLISH_PASSWORD" > ~/.pypirc python -m pip install --upgrade pip build twine setuptools wheel rm -rf build dist @@ -3933,6 +3950,9 @@ jobs: image: ubuntu-2204:2023.10.1 resource_class: xlarge working_directory: ~/project + parameters: + browser: + type: string steps: - checkout - setup_google_dns @@ -3962,7 +3982,7 @@ jobs: echo "Expires at: $EXPIRES_AT" neon branches create \ --project-id $NEON_PROJECT_ID \ - --name preview/commit-${CIRCLE_SHA1:0:7} \ + --name preview/commit-${CIRCLE_SHA1:0:7}-<< parameters.browser >> \ --expires-at $EXPIRES_AT \ --parent br-fancy-paper-ad1olsb3 \ --api-key $NEON_API_KEY || true @@ -3972,7 +3992,7 @@ jobs: E2E_UI_TEST_DATABASE_URL=$(neon connection-string \ --project-id $NEON_PROJECT_ID \ --api-key $NEON_API_KEY \ - --branch preview/commit-${CIRCLE_SHA1:0:7} \ + --branch preview/commit-${CIRCLE_SHA1:0:7}-<< parameters.browser >> \ --database-name yuneng-trial-db \ --role neondb_owner) echo $E2E_UI_TEST_DATABASE_URL @@ -3984,7 +4004,7 @@ jobs: -e UI_USERNAME="admin" \ -e UI_PASSWORD="gm" \ -e LITELLM_LICENSE=$LITELLM_LICENSE \ - --name litellm-docker-database \ + --name litellm-docker-database-<< parameters.browser >> \ -v $(pwd)/litellm/proxy/example_config_yaml/simple_config.yaml:/app/config.yaml \ litellm-docker-database:ci \ --config /app/config.yaml \ @@ -4000,7 +4020,7 @@ jobs: sudo rm dockerize-linux-amd64-v0.6.1.tar.gz - run: name: Start outputting logs - command: docker logs -f litellm-docker-database + command: docker logs -f litellm-docker-database-<< parameters.browser >> background: true - run: name: Wait for app to be ready @@ -4009,6 +4029,7 @@ jobs: name: Run Playwright Tests command: | npx playwright test \ + --project << parameters.browser >> \ --config ui/litellm-dashboard/e2e_tests/playwright.config.ts \ --reporter=html \ --output=test-results @@ -4115,6 +4136,12 @@ workflows: only: - main - /litellm_.*/ + - semgrep: + filters: + branches: + only: + - main + - /litellm_.*/ - local_testing_part1: filters: branches: @@ -4214,6 +4241,20 @@ workflows: - main - /litellm_.*/ - e2e_ui_testing: + name: e2e_ui_testing_chromium + browser: chromium + context: e2e_ui_tests + requires: + - ui_build + - build_docker_database_image + filters: + branches: + only: + - main + - /litellm_.*/ + - e2e_ui_testing: + name: e2e_ui_testing_firefox + browser: firefox context: e2e_ui_tests requires: - ui_build @@ -4493,6 +4534,7 @@ workflows: - publish_to_pypi: requires: - mypy_linting + - semgrep - local_testing_part1 - local_testing_part2 - build_and_test @@ -4525,7 +4567,8 @@ workflows: - litellm_assistants_api_testing - auth_ui_unit_tests - db_migration_disable_update_check - - e2e_ui_testing + - e2e_ui_testing_chromium + - e2e_ui_testing_firefox - litellm_proxy_unit_testing_key_generation - litellm_proxy_unit_testing_part1 - litellm_proxy_unit_testing_part2 diff --git a/.semgrep/rules/README.md b/.semgrep/rules/README.md new file mode 100644 index 00000000000..0dbb77cdd48 --- /dev/null +++ b/.semgrep/rules/README.md @@ -0,0 +1,22 @@ +# Custom Semgrep rules for LiteLLM + +Add custom rule YAML files here. Semgrep loads all `.yml`/`.yaml` files under this directory. + +**Run only custom rules (CI / fail on findings):** + +```bash +semgrep scan --config .semgrep/rules . --error +``` + +**Run with registry + custom rules:** + +```bash +semgrep scan --config auto --config .semgrep/rules . +``` + +**Layout:** + +- `python/` – Python-specific rules (security, patterns) +- Add more subdirs as needed (e.g. `generic/` for language-agnostic rules) + +See [Semgrep rule syntax](https://semgrep.dev/docs/writing-rules/rule-syntax/). diff --git a/.semgrep/rules/python/unbounded-memory.yml b/.semgrep/rules/python/unbounded-memory.yml new file mode 100644 index 00000000000..811ef689344 --- /dev/null +++ b/.semgrep/rules/python/unbounded-memory.yml @@ -0,0 +1,14 @@ +# Unbounded memory growth – data structures without a clear max limit +# Can lead to OOM under load. + +rules: + - id: unbounded-asyncio-queue + message: asyncio.Queue() with no maxsize can grow unbounded. Use asyncio.Queue(maxsize=N) for integrations (e.g. log queues). + severity: ERROR + languages: [python] + pattern-either: + - pattern: asyncio.Queue() + - pattern: asyncio.Queue(maxsize=0) + metadata: + category: correctness + cwe: "CWE-400: Uncontrolled Resource Consumption" \ No newline at end of file diff --git a/ci_cd/.grype.yaml b/ci_cd/.grype.yaml index 642e2dd9d03..b9bc9db58f5 100644 --- a/ci_cd/.grype.yaml +++ b/ci_cd/.grype.yaml @@ -1,3 +1,36 @@ ignore: - vulnerability: CVE-2026-22184 reason: no fixed zlib package is available yet in the Wolfi repositories, so this is ignored temporarily until an upstream release exists + # Wolfi base image: Python 3.13 and Node from apk have no fixed builds in Wolfi yet / not applicable + - vulnerability: CVE-2025-55130 + reason: Node in Wolfi apk; only used for Admin UI build/prisma + - vulnerability: CVE-2025-59465 + reason: Node in Wolfi apk; only used for Admin UI build/prisma + - vulnerability: CVE-2025-55131 + reason: Node in Wolfi apk; only used for Admin UI build/prisma + - vulnerability: CVE-2025-59466 + reason: Node in Wolfi apk; only used for Admin UI build/prisma + - vulnerability: CVE-2026-21637 + reason: Node in Wolfi apk; only used for Admin UI build/prisma + - vulnerability: CVE-2025-55132 + reason: Node in Wolfi apk; only used for Admin UI build/prisma + - vulnerability: GHSA-hx9q-6w63-j58v + reason: orjson dumps recursion; allowlisted + - vulnerability: GHSA-73rr-hh4g-fpgx + reason: diff npm transitive dep; override in package.json, allowlisted + - vulnerability: CVE-2026-0865 + reason: Python 3.13 in Wolfi base; no fixed apk build yet + - vulnerability: CVE-2025-15282 + reason: Python 3.13 in Wolfi base; no fixed apk build yet + - vulnerability: CVE-2026-0672 + reason: Python 3.13 in Wolfi base; no fixed apk build yet + - vulnerability: CVE-2025-15366 + reason: Python 3.13 in Wolfi base; no fixed apk build yet + - vulnerability: CVE-2025-15367 + reason: Python 3.13 in Wolfi base; no fixed apk build yet + - vulnerability: CVE-2025-11468 + reason: Python 3.13 in Wolfi base; no fixed apk build yet + - vulnerability: CVE-2025-12781 + reason: Python 3.13 in Wolfi base; no fixed apk build yet + - vulnerability: CVE-2026-1299 + reason: Python 3.13 in Wolfi base; no fixed apk build yet diff --git a/ci_cd/security_scans.sh b/ci_cd/security_scans.sh index 3ffa13c444f..2db72ae5c69 100755 --- a/ci_cd/security_scans.sh +++ b/ci_cd/security_scans.sh @@ -140,12 +140,14 @@ run_grype_scans() { "GHSA-34x7-hfp2-rc4v" # node-tar hardlink path traversal - not applicable, tar CLI not exposed in application code "GHSA-r6q2-hw4h-h46w" # node-tar not used by application runtime, Linux-only container, not affect by macOS APFS-specific exploit "GHSA-8rrh-rw8j-w5fx" # wheel is from chainguard and will be handled by then TODO: Remove this after Chainguard updates the wheel - "CVE-2025-59465" # We do not use Node in application runtime, only used for building Admin UI - "CVE-2025-55131" # We do not use Node in application runtime, only used for building Admin UI - "CVE-2025-59466" # We do not use Node in application runtime, only used for building Admin UI - "CVE-2025-55130" # We do not use Node in application runtime, only used for building Admin UI - "CVE-2025-59467" # We do not use Node in application runtime, only used for building Admin UI - "CVE-2026-21637" # We do not use Node in application runtime, only used for building Admin UI + "CVE-2025-59465" # Node only used for Admin UI build/prisma + "CVE-2025-55131" # Node only used for Admin UI build/prisma + "CVE-2025-59466" # Node only used for Admin UI build/prisma + "CVE-2025-55130" # Node only used for Admin UI build/prisma + "CVE-2025-59467" # Node only used for Admin UI build/prisma + "CVE-2026-21637" # Node only used for Admin UI build/prisma + "CVE-2025-55132" # Node only used for Admin UI build/prisma + "GHSA-hx9q-6w63-j58v" # orjson dumps recursion; allowlisted "CVE-2025-15281" # No fix available yet "CVE-2026-0865" # No fix available yet "CVE-2025-15282" # No fix available yet diff --git a/cookbook/nova_sonic_realtime.py b/cookbook/nova_sonic_realtime.py index 0ea0badfb01..c7a73c1d00f 100644 --- a/cookbook/nova_sonic_realtime.py +++ b/cookbook/nova_sonic_realtime.py @@ -16,10 +16,14 @@ Usage: import asyncio import base64 import json +import os import pyaudio import websockets from typing import Optional +# Bounded queue size for audio chunks (configurable via env to avoid unbounded memory) +AUDIO_QUEUE_MAXSIZE = int(os.getenv("LITELLM_ASYNCIO_QUEUE_MAXSIZE", 10_000)) + # Audio configuration (matching Nova Sonic requirements) INPUT_SAMPLE_RATE = 16000 # Nova Sonic expects 16kHz input OUTPUT_SAMPLE_RATE = 24000 # Nova Sonic outputs 24kHz @@ -40,7 +44,7 @@ class RealtimeClient: self.api_key = api_key self.ws: Optional[websockets.WebSocketClientProtocol] = None self.is_active = False - self.audio_queue = asyncio.Queue() + self.audio_queue = asyncio.Queue(maxsize=AUDIO_QUEUE_MAXSIZE) self.pyaudio = pyaudio.PyAudio() self.input_stream = None self.output_stream = None diff --git a/docs/my-website/blog/model_cost_map_incident/index.md b/docs/my-website/blog/model_cost_map_incident/index.md new file mode 100644 index 00000000000..b9ff20e4128 --- /dev/null +++ b/docs/my-website/blog/model_cost_map_incident/index.md @@ -0,0 +1,95 @@ +--- +slug: model-cost-map-incident +title: "Incident Report: Invalid model cost map on main" +date: 2026-02-10T10:00:00 +authors: + - name: Ishaan Jaffer + title: "CTO, LiteLLM" + url: https://www.linkedin.com/in/ishaanjaffer/ + image_url: https://pbs.twimg.com/profile_images/1613813310264340481/lz54oEiB_400x400.jpg +tags: [incident-report, stability] +hide_table_of_contents: false +--- + +**Date:** January 27, 2026 +**Duration:** ~20 minutes +**Severity:** Low +**Status:** Resolved + +## Summary + +A malformed JSON entry in `model_prices_and_context_window.json` was merged to `main` ([`562f0a0`](https://github.com/BerriAI/litellm/commit/562f0a028251750e3d75386bee0e630d9796d0df)). This caused LiteLLM to silently fall back to a stale local copy of the model cost map. Users on older package versions lost cost tracking for newer models only (e.g. `azure/gpt-5.2`). No LLM calls were blocked. + +- **LLM calls and proxy routing:** No impact. +- **Cost tracking:** Impacted for newer models not present in the local backup. Older models were unaffected. The incident lasted ~20 minutes until the commit was reverted. + +{/* truncate */} + +--- + +## Background + +The model cost map is not in the request path. It is used after the LLM response comes back, inside a try/catch, to calculate spend. A missing entry never blocks a call. + +```mermaid +flowchart TD + A["1. litellm.completion() receives request + litellm/main.py"] --> B["2. Route to provider + litellm/litellm_core_utils/get_llm_provider_logic.py"] + B --> C["3. LLM returns response + litellm/main.py"] + C --> D["4. Post-call: look up model in cost map + litellm/cost_calculator.py"] + D -->|"found"| E["5a. Attach cost to response"] + D -->|"not found (try/catch)"| F["5b. Log warning, set cost=0"] + E --> G["6. Return response to caller"] + F --> G + + style D fill:#fff3cd,stroke:#ffc107 + style F fill:#fff3cd,stroke:#ffc107 + style E fill:#d4edda,stroke:#28a745 + style G fill:#d4edda,stroke:#28a745 +``` + +Both paths return a response to the caller. When the cost map lookup fails, the only difference is `cost=0` on that request. + +--- + +## Root cause + +LiteLLM fetches the model cost map from GitHub `main` at import time. If the fetch fails, it falls back to a local backup bundled with the package. Before this incident, the fallback was completely silent -- no warning was logged. + +A contributor PR introduced an extra `{` bracket, producing invalid JSON. The remote fetch failed with `JSONDecodeError`, triggering the silent fallback. Users on older package versions had backup files missing newer models. + +**Timeline:** + +1. Malformed JSON merged to `main` +2. LiteLLM installations fall back to local backup on next import +3. Users report `"This model isn't mapped yet"` for newer models +4. Bad commit identified and reverted (~20 minutes) + +--- + +## Remediation + +| # | Action | Status | Code | +|---|---|---|---| +| 1 | CI validation on `model_prices_and_context_window.json` | ✅ Done | [`test-model-map.yaml`](https://github.com/BerriAI/litellm/blob/main/.github/workflows/test-model-map.yaml) | +| 2 | Warning log on fallback to local backup | ✅ Done | [`get_model_cost_map.py#L57-L68`](https://github.com/BerriAI/litellm/blob/main/litellm/litellm_core_utils/get_model_cost_map.py#L57-L68) | +| 3 | `GetModelCostMap` class with integrity validation helpers | ✅ Done | [`get_model_cost_map.py#L24-L149`](https://github.com/BerriAI/litellm/blob/main/litellm/litellm_core_utils/get_model_cost_map.py#L24-L149) | +| 4 | Resilience test suite (bad hosted map, fallback, completion) | ✅ Done | [`test_model_cost_map_resilience.py#L150-L291`](https://github.com/BerriAI/litellm/blob/main/tests/llm_translation/test_model_cost_map_resilience.py#L150-L291) | +| 5 | Test that backup model cost map always exists and contains common models | ✅ Done | [`test_model_cost_map_resilience.py#L213-L228`](https://github.com/BerriAI/litellm/blob/main/tests/llm_translation/test_model_cost_map_resilience.py#L213-L228) | + +Enterprises that require zero external dependencies at import time can set `LITELLM_LOCAL_MODEL_COST_MAP=True` to skip the GitHub fetch entirely. + +--- + +## Other dependencies on external resources + +| Dependency | Impact if unavailable | Fallback | +|---|---|---| +| Model cost map (GitHub) | Cost tracking for newer models | Local backup (now with warning) | +| JWT public keys (IDP/SSO) | Auth fails | None | +| OIDC UserInfo (IDP/SSO) | Auth fails | None | +| HuggingFace model API | HF provider calls fail | None | +| Ollama tags (localhost) | Ollama model list stale | Static list | diff --git a/docs/my-website/docs/mcp_oauth.md b/docs/my-website/docs/mcp_oauth.md index ed69408196f..9cd7b1e77be 100644 --- a/docs/my-website/docs/mcp_oauth.md +++ b/docs/my-website/docs/mcp_oauth.md @@ -1,6 +1,3 @@ -import Tabs from '@theme/Tabs'; -import TabItem from '@theme/TabItem'; - # MCP OAuth LiteLLM supports two OAuth 2.0 flows for MCP servers: @@ -98,8 +95,71 @@ LiteLLM automatically fetches, caches, and refreshes OAuth2 tokens using the `cl ### Setup - - +You can configure M2M OAuth via the LiteLLM UI or `config.yaml`. + +### UI Setup + +Navigate to the **MCP Servers** page and click **+ Add New MCP Server**. + +![](https://colony-recorder.s3.amazonaws.com/files/2026-02-10/d1f1e89c-a789-4975-8846-b15d9821984a/ascreenshot_630800e00a2e4b598baabfc25efbabd3_text_export.jpeg) + +Enter a name for your server and select **HTTP** as the transport type. + +![](https://colony-recorder.s3.amazonaws.com/files/2026-02-10/2008c9d6-6093-4121-beab-1e52c71376aa/ascreenshot_516ffd6c7b524465a253a56048c3d228_text_export.jpeg) + +Paste the MCP server URL. + +![](https://colony-recorder.s3.amazonaws.com/files/2026-02-10/b0ee8b7d-6de8-492b-8962-287987feec29/ascreenshot_b3efca82078a4c6bb1453c58161909f9_text_export.jpeg) + +Under **Authentication**, select **OAuth**. + +![](https://colony-recorder.s3.amazonaws.com/files/2026-02-10/e1597814-ff8e-40b9-9d7b-864dcdbe0910/ascreenshot_2097612712264d8f9e553f7ca9175fb0_text_export.jpeg) + +![](https://colony-recorder.s3.amazonaws.com/files/2026-02-10/f6ea5694-f28a-4bc3-9c9a-bb79f199bd65/ascreenshot_9be839f55b1b4f96bfe24030ba2c7f8d_text_export.jpeg) + +Choose **Machine-to-Machine (M2M)** as the OAuth flow type. This is for server-to-server authentication using the `client_credentials` grant — no browser interaction required. + +![](https://colony-recorder.s3.amazonaws.com/files/2026-02-10/9853310c-1d86-4628-bad1-7a391eca0e4d/ascreenshot_f302a286fa264fdd8d56db53b8f9395c_text_export.jpeg) + +![](https://colony-recorder.s3.amazonaws.com/files/2026-02-10/df64dc65-ef86-475d-adaf-12e227d5e873/ascreenshot_9e2f41d43a76435f918a00b52ffcc639_text_export.jpeg) + +Fill in the **Client ID** and **Client Secret** provided by your OAuth provider. + +![](https://colony-recorder.s3.amazonaws.com/files/2026-02-10/0de5a7bd-9898-4fc7-8843-b23dd5aac47f/ascreenshot_b9087aaa81a14b5b9c199929efc4a563_text_export.jpeg) + +Enter the **Token URL** — this is the endpoint LiteLLM will call to fetch access tokens using `client_credentials`. + +![](https://colony-recorder.s3.amazonaws.com/files/2026-02-10/0aea70f1-558c-4dca-91bc-1175fe1ddc89/ascreenshot_b3fcf8a1287e4e2d9a3d67c4a29f7bff_text_export.jpeg) + +![](https://colony-recorder.s3.amazonaws.com/files/2026-02-10/e842ef09-1fd7-47a6-909b-252d389f0abc/ascreenshot_2a87dad3624847e7ac370591d1d1aedd_text_export.jpeg) + +Scroll down and review the server URL and all fields, then click **Create MCP Server**. + +![](https://colony-recorder.s3.amazonaws.com/files/2026-02-10/0857712b-4b53-40f8-8c1f-a4c72edaa644/ascreenshot_47be3fcd5de64ed391f70c1fb74a8bfc_text_export.jpeg) + +![](https://colony-recorder.s3.amazonaws.com/files/2026-02-10/9d961765-955f-4905-a3dc-1a446aa3b2cc/ascreenshot_43fd39d014224564bc6b35aced1fb6d3_text_export.jpeg) + +![](https://colony-recorder.s3.amazonaws.com/files/2026-02-10/3825d5fa-8fd1-4e71-b090-77ff0259c3f6/ascreenshot_2509a7ebd9bf421eb0e82f2553566745_text_export.jpeg) + +Once created, open the server and navigate to the **MCP Tools** tab to verify that LiteLLM can connect and list available tools. + +![](https://colony-recorder.s3.amazonaws.com/files/2026-02-10/8107e27b-5072-4675-8fd6-89b47692b1bd/ascreenshot_f774bc76138f430d808fb4482ebfcdca_text_export.jpeg) + +![](https://colony-recorder.s3.amazonaws.com/files/2026-02-10/ce94bb7b-c81b-4396-9939-178efb2cdfce/ascreenshot_28b838ab6ae34c76858454555c4c1d79_text_export.jpeg) + +Select a tool (e.g. **echo**) to test it. Fill in the required parameters and click **Call Tool**. + +![](https://colony-recorder.s3.amazonaws.com/files/2026-02-10/c459c1d3-ec29-4211-9c28-37fbe7783bbc/ascreenshot_e9b138b3c2cc4440bb1a6f42ac7ae861_text_export.jpeg) + +![](https://colony-recorder.s3.amazonaws.com/files/2026-02-10/5438ac60-e0ac-4a79-bf6f-5594f160d3b5/ascreenshot_9133a17d26204c46bce497e74685c483_text_export.jpeg) + +![](https://colony-recorder.s3.amazonaws.com/files/2026-02-10/a8f6821b-3982-4b4d-9b25-70c8aff5ac31/ascreenshot_28d474d0e62545a482cff6128527883a_text_export.jpeg) + +LiteLLM automatically fetches an OAuth token behind the scenes and calls the tool. The result confirms the M2M OAuth flow is working end-to-end. + +![](https://colony-recorder.s3.amazonaws.com/files/2026-02-10/c3924549-a949-48d1-ac67-ab4c30475859/ascreenshot_8f6eca9d717f45478d50a881bd244bb3_text_export.jpeg) + +### Config.yaml Setup ```yaml title="config.yaml" showLineNumbers mcp_servers: @@ -112,14 +172,6 @@ mcp_servers: scopes: ["mcp:read", "mcp:write"] # optional ``` - - - -Navigate to **MCP Servers → Add Server → Authentication → OAuth**, then fill in `client_id`, `client_secret`, and `token_url`. - - - - ### How It Works 1. On first MCP request, LiteLLM POSTs to `token_url` with `grant_type=client_credentials` diff --git a/docs/my-website/docs/pass_through/bedrock.md b/docs/my-website/docs/pass_through/bedrock.md index b8d20d77da0..65c5d8caadc 100644 --- a/docs/my-website/docs/pass_through/bedrock.md +++ b/docs/my-website/docs/pass_through/bedrock.md @@ -556,3 +556,147 @@ for event in response.get("completion"): print(completion) ``` + +## Using LangChain AWS SDK with LiteLLM + +You can use the [LangChain AWS SDK](https://python.langchain.com/docs/integrations/chat/bedrock/) with LiteLLM Proxy to get cost tracking, load balancing, and other LiteLLM features. + +### Quick Start + +**1. Install LangChain AWS**: + +```bash showLineNumbers +pip install langchain-aws +``` + +**2. Setup LiteLLM Proxy**: + +Create a `config.yaml`: + +```yaml showLineNumbers +model_list: + - model_name: claude-sonnet + litellm_params: + model: bedrock/us.anthropic.claude-3-7-sonnet-20250219-v1:0 + aws_region_name: us-east-1 + custom_llm_provider: bedrock +``` + +Start the proxy: + +```bash showLineNumbers +export AWS_ACCESS_KEY_ID="your-access-key" +export AWS_SECRET_ACCESS_KEY="your-secret-key" + +litellm --config config.yaml + +# RUNNING on http://0.0.0.0:4000 +``` + +**3. Use LangChain with LiteLLM**: + +```python showLineNumbers +from langchain_aws import ChatBedrockConverse +from langchain_core.messages import HumanMessage + +# Your LiteLLM API key +API_KEY = "Bearer sk-1234" + +# Initialize ChatBedrockConverse pointing to LiteLLM proxy +llm = ChatBedrockConverse( + model_id="us.anthropic.claude-3-7-sonnet-20250219-v1:0", + endpoint_url="http://localhost:4000/bedrock", + region_name="us-east-1", + aws_access_key_id=API_KEY, + aws_secret_access_key="bedrock" # Any non-empty value works +) + +# Invoke the model +messages = [HumanMessage(content="Hello, how are you?")] +response = llm.invoke(messages) + +print(response.content) +``` + +### Advanced Example: PDF Document Processing with Citations + +LangChain AWS SDK supports Bedrock's document processing features. Here's how to use it with LiteLLM: + +```python showLineNumbers +import os +import json +from langchain_aws import ChatBedrockConverse +from langchain_core.messages import HumanMessage + +# Your LiteLLM API key +API_KEY = "Bearer sk-1234" + +def get_llm() -> ChatBedrockConverse: + """Initialize LLM pointing to LiteLLM proxy""" + llm = ChatBedrockConverse( + model_id="us.anthropic.claude-3-7-sonnet-20250219-v1:0", + base_model_id="anthropic.claude-3-7-sonnet-20250219-v1:0", + endpoint_url="http://localhost:4000/bedrock", + region_name="us-east-1", + aws_access_key_id=API_KEY, + aws_secret_access_key="bedrock" + ) + return llm + +if __name__ == "__main__": + # Initialize the LLM + llm = get_llm() + + # Read PDF file as bytes (Converse API requires raw bytes) + with open("your-document.pdf", "rb") as file: + file_bytes = file.read() + + # Prepare messages with document attachment + messages = [ + HumanMessage(content=[ + {"text": "What is the policy number in this document?"}, + { + "document": { + "format": "pdf", + "name": "PolicyDocument", + "source": {"bytes": file_bytes}, + "citations": {"enabled": True} + } + } + ]) + ] + + # Invoke the LLM + response = llm.invoke(messages) + + # Print response with citations + print(json.dumps(response.content, indent=4)) +``` + +### Supported LangChain Features + +All LangChain AWS features work with LiteLLM: + +| Feature | Supported | Notes | +|---------|-----------|-------| +| Text Generation | ✅ | Full support | +| Streaming | ✅ | Use `stream()` method | +| Document Processing | ✅ | PDF, images, etc. | +| Citations | ✅ | Enable in document config | +| Tool Use | ✅ | Function calling support | +| Multi-modal | ✅ | Text + images + documents | + +### Troubleshooting + +**Issue**: `UnknownOperationException` error + +**Solution**: Make sure you're using the correct endpoint URL format: +- ✅ Correct: `http://localhost:4000/bedrock` +- ❌ Wrong: `http://localhost:4000/bedrock/v2` + +**Issue**: Authentication errors + +**Solution**: Ensure your API key is in the correct format: +```python +aws_access_key_id="Bearer sk-1234" # Include "Bearer " prefix +``` diff --git a/docs/my-website/docs/providers/perplexity.md b/docs/my-website/docs/providers/perplexity.md index 2fcb49c60fa..68adf9939c6 100644 --- a/docs/my-website/docs/providers/perplexity.md +++ b/docs/my-website/docs/providers/perplexity.md @@ -120,6 +120,293 @@ All models listed here https://docs.perplexity.ai/docs/model-cards are supported +## Agentic Research API (Responses API) + +Requires v1.72.6+ + + +### Using Presets + +Presets provide optimized defaults for specific use cases. Start with a preset for quick setup: + + + + +```python +from litellm import responses +import os + +os.environ['PERPLEXITY_API_KEY'] = "" + +# Using the pro-search preset +response = responses( + model="perplexity/preset/pro-search", + input="What are the latest developments in AI?", + custom_llm_provider="perplexity", +) + +print(response.output) +``` + + + + +1. Setup config.yaml + +```yaml +model_list: + - model_name: perplexity-pro-search + litellm_params: + model: perplexity/preset/pro-search + api_key: os.environ/PERPLEXITY_API_KEY +``` + +2. Start proxy + +```bash +litellm --config /path/to/config.yaml +``` + +3. Test it! + +```bash +curl http://0.0.0.0:4000/v1/responses \ + -H "Content-Type: application/json" \ + -H "Authorization: Bearer anything" \ + -d '{ + "model": "perplexity-pro-search", + "input": "What are the latest developments in AI?" + }' +``` + + + + +### Using Third-Party Models + +Access models from OpenAI, Anthropic, Google, xAI, and other providers through Perplexity's unified API: + + + + +```python +from litellm import responses +import os + +os.environ['PERPLEXITY_API_KEY'] = "" + +response = responses( + model="perplexity/openai/gpt-4o", + input="Explain quantum computing in simple terms", + custom_llm_provider="perplexity", + max_output_tokens=500, +) + +print(response.output) +``` + + + + +```python +from litellm import responses +import os + +os.environ['PERPLEXITY_API_KEY'] = "" + +response = responses( + model="perplexity/anthropic/claude-3-5-sonnet-20241022", + input="Write a short story about a robot learning to paint", + custom_llm_provider="perplexity", + max_output_tokens=500, +) + +print(response.output) +``` + + + + +```python +from litellm import responses +import os + +os.environ['PERPLEXITY_API_KEY'] = "" + +response = responses( + model="perplexity/google/gemini-2.0-flash-exp", + input="Explain the concept of neural networks", + custom_llm_provider="perplexity", + max_output_tokens=500, +) + +print(response.output) +``` + + + + +```python +from litellm import responses +import os + +os.environ['PERPLEXITY_API_KEY'] = "" + +response = responses( + model="perplexity/xai/grok-2-1212", + input="What makes a good AI assistant?", + custom_llm_provider="perplexity", + max_output_tokens=500, +) + +print(response.output) +``` + + + + +### Web Search Tool + +Enable web search capabilities to access real-time information: + +```python +from litellm import responses +import os + +os.environ['PERPLEXITY_API_KEY'] = "" + +response = responses( + model="perplexity/openai/gpt-4o", + input="What's the weather in San Francisco today?", + custom_llm_provider="perplexity", + tools=[{"type": "web_search"}], + instructions="You have access to a web_search tool. Use it for questions about current events.", +) + +print(response.output) +``` + + +### Reasoning Effort (Responses API) + +Control the reasoning effort level for reasoning-capable models: + +```python +from litellm import responses +import os + +os.environ['PERPLEXITY_API_KEY'] = "" + +response = responses( + model="perplexity/openai/gpt-5.2", + input="Solve this complex problem step by step", + custom_llm_provider="perplexity", + reasoning={"effort": "high"}, # Options: low, medium, high + max_output_tokens=1000, +) + +print(response.output) +``` + +### Multi-Turn Conversations + +Use message arrays for multi-turn conversations with context: + +```python +from litellm import responses +import os + +os.environ['PERPLEXITY_API_KEY'] = "" + +response = responses( + model="perplexity/anthropic/claude-3-5-sonnet-20241022", + input=[ + {"type": "message", "role": "system", "content": "You are a helpful assistant."}, + {"type": "message", "role": "user", "content": "What are the latest AI developments?"}, + ], + custom_llm_provider="perplexity", + instructions="Provide detailed, well-researched answers.", + max_output_tokens=800, +) + +print(response.output) +``` + +### Streaming Responses + +Stream responses for real-time output: + +```python +from litellm import responses +import os + +os.environ['PERPLEXITY_API_KEY'] = "" + +response = responses( + model="perplexity/openai/gpt-4o", + input="Tell me a story about space exploration", + custom_llm_provider="perplexity", + stream=True, + max_output_tokens=500, +) + +for chunk in response: + if hasattr(chunk, 'type'): + if chunk.type == "response.output_text.delta": + print(chunk.delta, end="", flush=True) +``` + +### Supported Third-Party Models + +| Provider | Model Name | Function Call | +|----------|------------|---------------| +| OpenAI | gpt-4o | `responses(model="perplexity/openai/gpt-4o", ...)` | +| OpenAI | gpt-4o-mini | `responses(model="perplexity/openai/gpt-4o-mini", ...)` | +| OpenAI | gpt-5.2 | `responses(model="perplexity/openai/gpt-5.2", ...)` | +| Anthropic | claude-3-5-sonnet-20241022 | `responses(model="perplexity/anthropic/claude-3-5-sonnet-20241022", ...)` | +| Anthropic | claude-3-5-haiku-20241022 | `responses(model="perplexity/anthropic/claude-3-5-haiku-20241022", ...)` | +| Google | gemini-2.0-flash-exp | `responses(model="perplexity/google/gemini-2.0-flash-exp", ...)` | +| Google | gemini-2.0-flash-thinking-exp | `responses(model="perplexity/google/gemini-2.0-flash-thinking-exp", ...)` | +| xAI | grok-2-1212 | `responses(model="perplexity/xai/grok-2-1212", ...)` | +| xAI | grok-2-vision-1212 | `responses(model="perplexity/xai/grok-2-vision-1212", ...)` | + +### Available Presets + +| Preset Name | Function Call | +|----------------|--------------------------------------------------------| +| fast-search | `responses(model="perplexity/preset/fast-search", ...)`| +| pro-search | `responses(model="perplexity/preset/pro-search", ...)` | +| deep-research | `responses(model="perplexity/preset/deep-research", ...)`| + +### Complete Example + +```python +from litellm import responses +import os + +os.environ['PERPLEXITY_API_KEY'] = "" + +# Comprehensive example with multiple features +response = responses( + model="perplexity/openai/gpt-4o", + input="Research the latest developments in quantum computing and provide sources", + custom_llm_provider="perplexity", + tools=[ + {"type": "web_search"}, + {"type": "fetch_url"} + ], + instructions="Use web_search to find relevant information and fetch_url to retrieve detailed content from sources. Provide citations for all claims.", + max_output_tokens=1000, + temperature=0.7, +) + +print(f"Response ID: {response.id}") +print(f"Model: {response.model}") +print(f"Status: {response.status}") +print(f"Output: {response.output}") +print(f"Usage: {response.usage}") +``` + :::info For more information about passing provider-specific parameters, [go here](../completion/provider_specific_params.md) diff --git a/docs/my-website/docs/proxy/admin_ui_sso.md b/docs/my-website/docs/proxy/admin_ui_sso.md index 37e45b50284..f88d3480446 100644 --- a/docs/my-website/docs/proxy/admin_ui_sso.md +++ b/docs/my-website/docs/proxy/admin_ui_sso.md @@ -223,6 +223,7 @@ GENERIC_USER_FIRST_NAME_ATTRIBUTE = "first_name" GENERIC_USER_LAST_NAME_ATTRIBUTE = "last_name" GENERIC_USER_ROLE_ATTRIBUTE = "given_role" GENERIC_USER_PROVIDER_ATTRIBUTE = "provider" +GENERIC_USER_EXTRA_ATTRIBUTES = "department,employee_id,manager" # comma-separated list of additional fields to extract from SSO response GENERIC_CLIENT_STATE = "some-state" # if the provider needs a state parameter GENERIC_INCLUDE_CLIENT_ID = "false" # some providers enforce that the client_id is not in the body GENERIC_SCOPE = "openid profile email" # default scope openid is sometimes not enough to retrieve basic user info like first_name and last_name located in profile scope @@ -239,6 +240,40 @@ Use `GENERIC_USER_ROLE_ATTRIBUTE` to specify which attribute in the SSO token co Nested attribute paths are supported (e.g., `claims.role` or `attributes.litellm_role`). +**Capturing Additional SSO Fields** + +Use `GENERIC_USER_EXTRA_ATTRIBUTES` to extract additional fields from the SSO provider response beyond the standard user attributes (id, email, name, etc.). This is useful when you need to access custom organization-specific data (e.g., department, employee ID, groups) in your [custom SSO handler](./custom_sso.md). + +```shell +# Comma-separated list of field names to extract +GENERIC_USER_EXTRA_ATTRIBUTES="department,employee_id,manager,groups" +``` + +**Accessing Extra Fields in Custom SSO Handler:** + +```python +from litellm.proxy.management_endpoints.types import CustomOpenID + +async def custom_sso_handler(userIDPInfo: CustomOpenID): + # Access the extra fields + extra_fields = getattr(userIDPInfo, 'extra_fields', None) or {} + + user_department = extra_fields.get("department") + employee_id = extra_fields.get("employee_id") + user_groups = extra_fields.get("groups", []) + + # Use these fields for custom logic (e.g., team assignment, access control) + # ... +``` + +**Nested Field Paths:** + +Dot notation is supported for nested fields: + +```shell +GENERIC_USER_EXTRA_ATTRIBUTES="org_info.department,org_info.cost_center,metadata.employee_type" +``` + - Set Redirect URI, if your provider requires it - Set a redirect url = `/sso/callback` ```shell diff --git a/docs/my-website/docs/proxy/config_settings.md b/docs/my-website/docs/proxy/config_settings.md index 5cdae51f448..52504faede8 100644 --- a/docs/my-website/docs/proxy/config_settings.md +++ b/docs/my-website/docs/proxy/config_settings.md @@ -548,6 +548,10 @@ router_settings: | DEFAULT_MCP_SEMANTIC_FILTER_EMBEDDING_MODEL | Default embedding model for MCP semantic tool filtering. Default is "text-embedding-3-small" | DEFAULT_MCP_SEMANTIC_FILTER_SIMILARITY_THRESHOLD | Default similarity threshold for MCP semantic tool filtering. Default is 0.3 | DEFAULT_MCP_SEMANTIC_FILTER_TOP_K | Default number of top results to return for MCP semantic tool filtering. Default is 10 +| MCP_OAUTH2_TOKEN_CACHE_DEFAULT_TTL | Default TTL in seconds for MCP OAuth2 token cache. Default is 3600 +| MCP_OAUTH2_TOKEN_CACHE_MAX_SIZE | Maximum number of entries in MCP OAuth2 token cache. Default is 200 +| MCP_OAUTH2_TOKEN_CACHE_MIN_TTL | Minimum TTL in seconds for MCP OAuth2 token cache. Default is 10 +| MCP_OAUTH2_TOKEN_EXPIRY_BUFFER_SECONDS | Seconds to subtract from token expiry when computing cache TTL. Default is 60 | DEFAULT_MOCK_RESPONSE_COMPLETION_TOKEN_COUNT | Default token count for mock response completions. Default is 20 | DEFAULT_MOCK_RESPONSE_PROMPT_TOKEN_COUNT | Default token count for mock response prompts. Default is 10 | DEFAULT_MODEL_CREATED_AT_TIME | Default creation timestamp for models. Default is 1677610602 @@ -640,6 +644,7 @@ router_settings: | GENERIC_TOKEN_ENDPOINT | Token endpoint for generic OAuth providers | GENERIC_USER_DISPLAY_NAME_ATTRIBUTE | Attribute for user's display name in generic auth | GENERIC_USER_EMAIL_ATTRIBUTE | Attribute for user's email in generic auth +| GENERIC_USER_EXTRA_ATTRIBUTES | Comma-separated list of additional fields to extract from generic SSO provider response (e.g., "department,employee_id,groups"). Accessible via `CustomOpenID.extra_fields` in custom SSO handlers. Supports dot notation for nested fields | GENERIC_USER_FIRST_NAME_ATTRIBUTE | Attribute for user's first name in generic auth | GENERIC_USER_ID_ATTRIBUTE | Attribute for user ID in generic auth | GENERIC_USER_LAST_NAME_ATTRIBUTE | Attribute for user's last name in generic auth @@ -779,6 +784,7 @@ router_settings: | LITELLM_USER_AGENT | Custom user agent string for LiteLLM API requests. Used for partner telemetry attribution | LITELLM_PRINT_STANDARD_LOGGING_PAYLOAD | If true, prints the standard logging payload to the console - useful for debugging | LITELM_ENVIRONMENT | Environment for LiteLLM Instance. This is currently only logged to DeepEval to determine the environment for DeepEval integration. +| LITELLM_ASYNCIO_QUEUE_MAXSIZE | Maximum size for asyncio queues (e.g. log queues, spend update queues, and cookbook examples such as realtime audio in `nova_sonic_realtime.py`). Bounds in-memory growth to prevent OOM. Default is 1000. | LOGFIRE_TOKEN | Token for Logfire logging service | LOGFIRE_BASE_URL | Base URL for Logfire logging service (useful for self hosted deployments) | LOGGING_WORKER_CONCURRENCY | Maximum number of concurrent coroutine slots for the logging worker on the asyncio event loop. Default is 100. Setting too high will flood the event loop with logging tasks which will lower the overall latency of the requests. diff --git a/docs/my-website/docs/proxy/custom_sso.md b/docs/my-website/docs/proxy/custom_sso.md index bbd7f41bee1..8b7adeb0c5a 100644 --- a/docs/my-website/docs/proxy/custom_sso.md +++ b/docs/my-website/docs/proxy/custom_sso.md @@ -142,6 +142,18 @@ async def custom_sso_handler(userIDPInfo: OpenID) -> SSOUserDefinedValues: f"No ID found for user. userIDPInfo.id is None {userIDPInfo}" ) + ################################################# + # Access extra fields from SSO provider (requires GENERIC_USER_EXTRA_ATTRIBUTES env var) + # Example: Set GENERIC_USER_EXTRA_ATTRIBUTES="department,employee_id,groups" + extra_fields = getattr(userIDPInfo, 'extra_fields', None) or {} + user_department = extra_fields.get("department") + employee_id = extra_fields.get("employee_id") + user_groups = extra_fields.get("groups", []) + + print(f"User department: {user_department}") # noqa + print(f"Employee ID: {employee_id}") # noqa + print(f"User groups: {user_groups}") # noqa + ################################################# ################################################# # Run your custom code / logic here diff --git a/docs/my-website/sidebars.js b/docs/my-website/sidebars.js index 539068fbcb1..579b5699101 100644 --- a/docs/my-website/sidebars.js +++ b/docs/my-website/sidebars.js @@ -1108,6 +1108,17 @@ const sidebars = { "troubleshoot/max_callbacks", ], }, + { + type: "category", + label: "Blog", + items: [ + { + type: "link", + label: "Incident: Broken Model Cost Map", + href: "/blog/model-cost-map-incident", + }, + ], + }, ], }; diff --git a/enterprise/litellm_enterprise/proxy/hooks/managed_files.py b/enterprise/litellm_enterprise/proxy/hooks/managed_files.py index 569ea17f6d8..a41b3f3bf6f 100644 --- a/enterprise/litellm_enterprise/proxy/hooks/managed_files.py +++ b/enterprise/litellm_enterprise/proxy/hooks/managed_files.py @@ -899,49 +899,49 @@ class _PROXY_LiteLLMManagedFiles(CustomLogger, BaseFileEndpoints): batch_id=response.id, model_id=model_id ) - if ( - response.output_file_id and model_id - ): # return a file id with the model_id and output_file_id - original_output_file_id = response.output_file_id - response.output_file_id = self.get_unified_output_file_id( - output_file_id=response.output_file_id, - model_id=model_id, - model_name=model_name, - ) - - # Fetch the actual file object for the output file - file_object = None - try: - # Use litellm to retrieve the file object from the provider - from litellm import afile_retrieve - file_object = await afile_retrieve( - custom_llm_provider=model_name.split("/")[0] if model_name and "/" in model_name else "openai", - file_id=original_output_file_id + # Handle both output_file_id and error_file_id + for file_attr in ["output_file_id", "error_file_id"]: + file_id_value = getattr(response, file_attr, None) + if file_id_value and model_id: + original_file_id = file_id_value + unified_file_id = self.get_unified_output_file_id( + output_file_id=original_file_id, + model_id=model_id, + model_name=model_name, ) - verbose_logger.debug( - f"Successfully retrieved file object for output_file_id={original_output_file_id}" + setattr(response, file_attr, unified_file_id) + + # Fetch the actual file object from the provider + file_object = None + try: + # Use litellm to retrieve the file object from the provider + from litellm import afile_retrieve + file_object = await afile_retrieve( + custom_llm_provider=model_name.split("/")[0] if model_name and "/" in model_name else "openai", + file_id=original_file_id + ) + verbose_logger.debug( + f"Successfully retrieved file object for {file_attr}={original_file_id}" + ) + except Exception as e: + verbose_logger.warning( + f"Failed to retrieve file object for {file_attr}={original_file_id}: {str(e)}. Storing with None and will fetch on-demand." + ) + + await self.store_unified_file_id( + file_id=unified_file_id, + file_object=file_object, + litellm_parent_otel_span=user_api_key_dict.parent_otel_span, + model_mappings={model_id: original_file_id}, + user_api_key_dict=user_api_key_dict, ) - except Exception as e: - verbose_logger.warning( - f"Failed to retrieve file object for output_file_id={original_output_file_id}: {str(e)}. Storing with None and will fetch on-demand." - ) - - await self.store_unified_file_id( - file_id=response.output_file_id, - file_object=file_object, - litellm_parent_otel_span=user_api_key_dict.parent_otel_span, - model_mappings={model_id: original_output_file_id}, - user_api_key_dict=user_api_key_dict, - ) - asyncio.create_task( - self.store_unified_object_id( - unified_object_id=response.id, - file_object=response, - litellm_parent_otel_span=user_api_key_dict.parent_otel_span, - model_object_id=original_response_id, - file_purpose="batch", - user_api_key_dict=user_api_key_dict, - ) + await self.store_unified_object_id( + unified_object_id=response.id, + file_object=response, + litellm_parent_otel_span=user_api_key_dict.parent_otel_span, + model_object_id=original_response_id, + file_purpose="batch", + user_api_key_dict=user_api_key_dict, ) elif isinstance(response, LiteLLMFineTuningJob): ## Check if unified_file_id is in the response @@ -958,15 +958,13 @@ class _PROXY_LiteLLMManagedFiles(CustomLogger, BaseFileEndpoints): response.id = self.get_unified_generic_response_id( model_id=model_id, generic_response_id=response.id ) - asyncio.create_task( - self.store_unified_object_id( - unified_object_id=response.id, - file_object=response, - litellm_parent_otel_span=user_api_key_dict.parent_otel_span, - model_object_id=original_response_id, - file_purpose="fine-tune", - user_api_key_dict=user_api_key_dict, - ) + await self.store_unified_object_id( + unified_object_id=response.id, + file_object=response, + litellm_parent_otel_span=user_api_key_dict.parent_otel_span, + model_object_id=original_response_id, + file_purpose="fine-tune", + user_api_key_dict=user_api_key_dict, ) elif isinstance(response, AsyncCursorPage): """ diff --git a/litellm-proxy-extras/dist/litellm_proxy_extras-0.4.34-py3-none-any.whl b/litellm-proxy-extras/dist/litellm_proxy_extras-0.4.34-py3-none-any.whl new file mode 100644 index 00000000000..175d84543ec Binary files /dev/null and b/litellm-proxy-extras/dist/litellm_proxy_extras-0.4.34-py3-none-any.whl differ diff --git a/litellm-proxy-extras/dist/litellm_proxy_extras-0.4.34.tar.gz b/litellm-proxy-extras/dist/litellm_proxy_extras-0.4.34.tar.gz new file mode 100644 index 00000000000..e1fcc0c603f Binary files /dev/null and b/litellm-proxy-extras/dist/litellm_proxy_extras-0.4.34.tar.gz differ diff --git a/litellm-proxy-extras/pyproject.toml b/litellm-proxy-extras/pyproject.toml index 8937858bbd9..e0a769a5edf 100644 --- a/litellm-proxy-extras/pyproject.toml +++ b/litellm-proxy-extras/pyproject.toml @@ -1,6 +1,6 @@ [tool.poetry] name = "litellm-proxy-extras" -version = "0.4.33" +version = "0.4.34" description = "Additional files for the LiteLLM Proxy. Reduces the size of the main litellm package." authors = ["BerriAI"] readme = "README.md" @@ -22,7 +22,7 @@ requires = ["poetry-core"] build-backend = "poetry.core.masonry.api" [tool.commitizen] -version = "0.4.33" +version = "0.4.34" version_files = [ "pyproject.toml:version", "../requirements.txt:litellm-proxy-extras==", diff --git a/litellm/__init__.py b/litellm/__init__.py index 8174b9d2655..dacb928e8a7 100644 --- a/litellm/__init__.py +++ b/litellm/__init__.py @@ -1393,6 +1393,7 @@ if TYPE_CHECKING: from .llms.litellm_proxy.responses.transformation import LiteLLMProxyResponsesAPIConfig as LiteLLMProxyResponsesAPIConfig from .llms.volcengine.responses.transformation import VolcEngineResponsesAPIConfig as VolcEngineResponsesAPIConfig from .llms.manus.responses.transformation import ManusResponsesAPIConfig as ManusResponsesAPIConfig + from .llms.perplexity.responses.transformation import PerplexityResponsesConfig as PerplexityResponsesConfig from .llms.gemini.interactions.transformation import GoogleAIStudioInteractionsConfig as GoogleAIStudioInteractionsConfig from .llms.openai.chat.o_series_transformation import OpenAIOSeriesConfig as OpenAIOSeriesConfig, OpenAIOSeriesConfig as OpenAIO1Config from .llms.anthropic.skills.transformation import AnthropicSkillsConfig as AnthropicSkillsConfig diff --git a/litellm/_lazy_imports_registry.py b/litellm/_lazy_imports_registry.py index a01fe9c11db..ebe9af9d85c 100644 --- a/litellm/_lazy_imports_registry.py +++ b/litellm/_lazy_imports_registry.py @@ -226,6 +226,7 @@ LLM_CONFIG_NAMES = ( "XAIResponsesAPIConfig", "LiteLLMProxyResponsesAPIConfig", "VolcEngineResponsesAPIConfig", + "PerplexityResponsesConfig", "GoogleAIStudioInteractionsConfig", "OpenAIOSeriesConfig", "AnthropicSkillsConfig", @@ -274,6 +275,7 @@ LLM_CONFIG_NAMES = ( "LmStudioEmbeddingConfig", "NscaleConfig", "PerplexityChatConfig", + "PerplexityResponsesConfig", "AzureOpenAIO1Config", "IBMWatsonXAIConfig", "IBMWatsonXChatConfig", @@ -901,6 +903,10 @@ _LLM_CONFIGS_IMPORT_MAP = { ".llms.manus.responses.transformation", "ManusResponsesAPIConfig", ), + "PerplexityResponsesConfig": ( + ".llms.perplexity.responses.transformation", + "PerplexityResponsesConfig", + ), "GoogleAIStudioInteractionsConfig": ( ".llms.gemini.interactions.transformation", "GoogleAIStudioInteractionsConfig", diff --git a/litellm/_logging.py b/litellm/_logging.py index e222627e76c..fd833f7056a 100644 --- a/litellm/_logging.py +++ b/litellm/_logging.py @@ -1,9 +1,13 @@ -import json +import ast import logging import os import sys from datetime import datetime from logging import Formatter +from typing import Any, Dict, Optional + +from litellm.litellm_core_utils.safe_json_dumps import safe_dumps +from litellm.litellm_core_utils.safe_json_loads import safe_json_loads set_verbose = False @@ -19,6 +23,67 @@ handler = logging.StreamHandler() handler.setLevel(numeric_level) +def _try_parse_json_message(message: str) -> Optional[Dict[str, Any]]: + """ + Try to parse a log message as JSON. Returns parsed dict if valid, else None. + Handles messages that are entirely valid JSON (e.g. json.dumps output). + Uses shared safe_json_loads for consistent error handling. + """ + if not message or not isinstance(message, str): + return None + msg_stripped = message.strip() + if not (msg_stripped.startswith("{") or msg_stripped.startswith("[")): + return None + parsed = safe_json_loads(message, default=None) + if parsed is None or not isinstance(parsed, dict): + return None + return parsed + + +def _try_parse_embedded_python_dict(message: str) -> Optional[Dict[str, Any]]: + """ + Try to find and parse a Python dict repr (e.g. str(d) or repr(d)) embedded in + the message. Handles patterns like: + "get_available_deployment for model: X, Selected deployment: {'model_name': '...', ...} for model: X" + Uses ast.literal_eval for safe parsing. Returns the parsed dict or None. + """ + if not message or not isinstance(message, str) or "{" not in message: + return None + i = 0 + while i < len(message): + start = message.find("{", i) + if start == -1: + break + depth = 0 + for j in range(start, len(message)): + c = message[j] + if c == "{": + depth += 1 + elif c == "}": + depth -= 1 + if depth == 0: + substr = message[start : j + 1] + try: + result = ast.literal_eval(substr) + if isinstance(result, dict) and len(result) > 0: + return result + except (ValueError, SyntaxError, TypeError): + pass + break + i = start + 1 + return None + + +# Standard LogRecord attribute names - used to identify 'extra' fields. +# Derived at runtime so we automatically include version-specific attrs (e.g. taskName). +def _get_standard_record_attrs() -> frozenset: + """Standard LogRecord attribute names - excludes extra keys from logger.debug(..., extra={...}).""" + return frozenset(logging.LogRecord("", 0, "", 0, "", (), None).__dict__.keys()) + + +_STANDARD_RECORD_ATTRS = _get_standard_record_attrs() + + class JsonFormatter(Formatter): def __init__(self): super(JsonFormatter, self).__init__() @@ -29,16 +94,31 @@ class JsonFormatter(Formatter): return dt.isoformat() def format(self, record): - json_record = { - "message": record.getMessage(), + message_str = record.getMessage() + json_record: Dict[str, Any] = { + "message": message_str, "level": record.levelname, "timestamp": self.formatTime(record), } + # Parse embedded JSON or Python dict repr in message so sub-fields become first-class properties + parsed = _try_parse_json_message(message_str) + if parsed is None: + parsed = _try_parse_embedded_python_dict(message_str) + if parsed is not None: + for key, value in parsed.items(): + if key not in json_record: + json_record[key] = value + + # Include extra attributes passed via logger.debug("msg", extra={...}) + for key, value in record.__dict__.items(): + if key not in _STANDARD_RECORD_ATTRS and key not in json_record: + json_record[key] = value + if record.exc_info: json_record["stacktrace"] = self.formatException(record.exc_info) - return json.dumps(json_record) + return safe_dumps(json_record) # Function to set up exception handlers for JSON logging @@ -169,15 +249,15 @@ def _initialize_loggers_with_handler(handler: logging.Handler): def _get_uvicorn_json_log_config(): """ Generate a uvicorn log_config dictionary that applies JSON formatting to all loggers. - + This ensures that uvicorn's access logs, error logs, and all application logs are formatted as JSON when json_logs is enabled. """ json_formatter_class = "litellm._logging.JsonFormatter" - + # Use the module-level log_level variable for consistency uvicorn_log_level = log_level.upper() - + log_config = { "version": 1, "disable_existing_loggers": False, @@ -222,7 +302,7 @@ def _get_uvicorn_json_log_config(): }, }, } - + return log_config diff --git a/litellm/constants.py b/litellm/constants.py index f288b11f2b5..88c57d3ce4c 100644 --- a/litellm/constants.py +++ b/litellm/constants.py @@ -48,6 +48,14 @@ DEFAULT_REPLICATE_POLLING_DELAY_SECONDS = int( os.getenv("DEFAULT_REPLICATE_POLLING_DELAY_SECONDS", 1) ) DEFAULT_IMAGE_TOKEN_COUNT = int(os.getenv("DEFAULT_IMAGE_TOKEN_COUNT", 250)) + +# Model cost map validation constants +MODEL_COST_MAP_MIN_MODEL_COUNT = int( + os.getenv("MODEL_COST_MAP_MIN_MODEL_COUNT", 50) +) # Minimum number of models a fetched cost map must contain to be considered valid +MODEL_COST_MAP_MAX_SHRINK_RATIO = float( + os.getenv("MODEL_COST_MAP_MAX_SHRINK_RATIO", 0.5) +) # Maximum allowed shrinkage ratio vs local backup (0.5 = reject if fetched map is <50% of backup) DEFAULT_IMAGE_WIDTH = int(os.getenv("DEFAULT_IMAGE_WIDTH", 300)) DEFAULT_IMAGE_HEIGHT = int(os.getenv("DEFAULT_IMAGE_HEIGHT", 300)) # Maximum size for image URL downloads in MB (default 50MB, set to 0 to disable limit) @@ -205,6 +213,10 @@ REDIS_DAILY_AGENT_SPEND_UPDATE_BUFFER_KEY = "litellm_daily_agent_spend_update_bu REDIS_DAILY_TAG_SPEND_UPDATE_BUFFER_KEY = "litellm_daily_tag_spend_update_buffer" MAX_REDIS_BUFFER_DEQUEUE_COUNT = int(os.getenv("MAX_REDIS_BUFFER_DEQUEUE_COUNT", 100)) MAX_SIZE_IN_MEMORY_QUEUE = int(os.getenv("MAX_SIZE_IN_MEMORY_QUEUE", 2000)) +# Bounds asyncio.Queue() instances (log queues, spend update queues, etc.) to prevent unbounded memory growth +LITELLM_ASYNCIO_QUEUE_MAXSIZE = int( + os.getenv("LITELLM_ASYNCIO_QUEUE_MAXSIZE", 1000) +) MAX_IN_MEMORY_QUEUE_FLUSH_COUNT = int( os.getenv("MAX_IN_MEMORY_QUEUE_FLUSH_COUNT", 1000) ) diff --git a/litellm/integrations/cloudzero/transform.py b/litellm/integrations/cloudzero/transform.py index e06b944a419..c36833a6dbf 100644 --- a/litellm/integrations/cloudzero/transform.py +++ b/litellm/integrations/cloudzero/transform.py @@ -103,10 +103,15 @@ class CBFTransformer: # Use team_alias if available, otherwise team_id, otherwise fallback to 'unknown' entity_id = str(team_alias) if team_alias else (str(team_id) if team_id else 'unknown') + # Get alias fields if they exist + api_key_alias = row.get('api_key_alias') + organization_alias = row.get('organization_alias') + project_alias = row.get('project_alias') + user_alias = row.get('user_alias') + dimensions = { 'entity_type': CZEntityType.TEAM.value, 'entity_id': entity_id, - 'team_id': str(team_id) if team_id else 'unknown', 'team_alias': str(team_alias) if team_alias else 'unknown', 'model': model, 'model_group': str(row.get('model_group', '')), @@ -119,28 +124,37 @@ class CBFTransformer: 'failed_requests': str(row.get('failed_requests', 0)), 'cache_creation_tokens': str(row.get('cache_creation_input_tokens', 0)), 'cache_read_tokens': str(row.get('cache_read_input_tokens', 0)), + 'organization_alias': str(organization_alias) if organization_alias else '', + 'project_alias': str(project_alias) if project_alias else '', + 'user_alias': str(user_alias) if user_alias else '', } # Extract CZRN components to populate corresponding CBF columns czrn_components = self.czrn_generator.extract_components(resource_id) service_type, provider, region, owner_account_id, resource_type, cloud_local_id = czrn_components + # Build resource/account as concat of api_key_alias and api_key_prefix + resource_account = f"{api_key_alias}|{api_key_hash}" if api_key_alias else api_key_hash + # CloudZero CBF format with proper column names cbf_record = { # Required CBF fields 'time/usage_start': usage_date.isoformat() if usage_date else None, # Required: ISO-formatted UTC datetime 'cost/cost': float(row.get('spend', 0.0)), # Required: billed cost - 'resource/id': resource_id, # Required when resource tags are present + 'resource/id': model, # Send model name # Usage metrics for token consumption 'usage/amount': total_tokens, # Numeric value of tokens consumed 'usage/units': 'tokens', # Description of token units - # CBF fields that correspond to CZRN components - 'resource/service': service_type, # Maps to CZRN service-type (litellm) - 'resource/account': owner_account_id, # Maps to CZRN owner-account-id (entity_id) + # CBF fields - updated per LIT-1907 + 'resource/service': str(row.get('model_group', '')), # Send model_group + 'resource/account': resource_account, # Send api_key_alias|api_key_prefix 'resource/region': region, # Maps to CZRN region (cross-region) - 'resource/usage_family': resource_type, # Maps to CZRN resource-type (llm-usage) + 'resource/usage_family': str(row.get('custom_llm_provider', '')), # Send provider + + # Action field + 'action/operation': str(team_id) if team_id else '', # Send team_id # Line item details 'lineitem/type': 'Usage', # Standard usage line item @@ -155,13 +169,11 @@ class CBFTransformer: if value and value != 'N/A' and value != 'unknown': # Only add meaningful tags cbf_record[f'resource/tag:{key}'] = str(value) - # Add token breakdown as resource tags for analysis + # Add token breakdown as resource tags for analysis (excluding total_tokens per LIT-1907) if prompt_tokens > 0: cbf_record['resource/tag:prompt_tokens'] = str(prompt_tokens) if completion_tokens > 0: cbf_record['resource/tag:completion_tokens'] = str(completion_tokens) - if total_tokens > 0: - cbf_record['resource/tag:total_tokens'] = str(total_tokens) return CBFRecord(cbf_record) diff --git a/litellm/integrations/custom_guardrail.py b/litellm/integrations/custom_guardrail.py index bbd55a59bce..407bc581f71 100644 --- a/litellm/integrations/custom_guardrail.py +++ b/litellm/integrations/custom_guardrail.py @@ -616,6 +616,7 @@ class CustomGuardrail(CustomLogger): end_time: Optional[float] = None, duration: Optional[float] = None, event_type: Optional[GuardrailEventHooks] = None, + original_inputs: Optional[Dict] = None, ): """ Add StandardLoggingGuardrailInformation to the request data @@ -625,6 +626,17 @@ class CustomGuardrail(CustomLogger): # Convert None to empty dict to satisfy type requirements guardrail_response = {} if response is None else response + # For apply_guardrail functions in custom_code_guardrail scenario, + # simplify the logged response to "allow", "deny", or "mask" + if original_inputs is not None and isinstance(response, dict): + # Check if inputs were modified by comparing them + if self._inputs_were_modified(original_inputs, response): + guardrail_response = "mask" + else: + guardrail_response = "allow" + + verbose_logger.debug(f"Guardrail response: {response}") + self.add_standard_logging_guardrail_information_to_request_data( guardrail_json_response=guardrail_response, request_data=request_data, @@ -650,8 +662,14 @@ class CustomGuardrail(CustomLogger): This gets logged on downsteam Langfuse, DataDog, etc. """ + # For custom_code_guardrail scenario, log as "deny" instead of full exception + # Check if this is from custom_code_guardrail by checking the class name + guardrail_response: Union[Exception, str] = e + if "CustomCodeGuardrail" in self.__class__.__name__: + guardrail_response = "deny" + self.add_standard_logging_guardrail_information_to_request_data( - guardrail_json_response=e, + guardrail_json_response=guardrail_response, request_data=request_data, guardrail_status="guardrail_failed_to_respond", duration=duration, @@ -661,6 +679,25 @@ class CustomGuardrail(CustomLogger): ) raise e + def _inputs_were_modified(self, original_inputs: Dict, response: Dict) -> bool: + """ + Compare original inputs with response to determine if content was modified. + + Returns True if the inputs were modified (mask scenario), False otherwise (allow scenario). + """ + # Get all keys from both dictionaries + all_keys = set(original_inputs.keys()) | set(response.keys()) + + # Compare each key's value + for key in all_keys: + original_value = original_inputs.get(key) + response_value = response.get(key) + if original_value != response_value: + return True + + # No modifications detected + return False + def mask_content_in_string( self, content_string: str, @@ -768,6 +805,12 @@ def log_guardrail_information(func): self: CustomGuardrail = args[0] request_data: dict = kwargs.get("data") or kwargs.get("request_data") or {} event_type = _infer_event_type_from_function_name(func.__name__) + + # Store original inputs for comparison (for apply_guardrail functions) + original_inputs = None + if func.__name__ == "apply_guardrail" and "inputs" in kwargs: + original_inputs = kwargs.get("inputs") + try: response = await func(*args, **kwargs) return self._process_response( @@ -777,6 +820,7 @@ def log_guardrail_information(func): end_time=datetime.now().timestamp(), duration=(datetime.now() - start_time).total_seconds(), event_type=event_type, + original_inputs=original_inputs, ) except Exception as e: return self._process_error( @@ -794,6 +838,12 @@ def log_guardrail_information(func): self: CustomGuardrail = args[0] request_data: dict = kwargs.get("data") or kwargs.get("request_data") or {} event_type = _infer_event_type_from_function_name(func.__name__) + + # Store original inputs for comparison (for apply_guardrail functions) + original_inputs = None + if func.__name__ == "apply_guardrail" and "inputs" in kwargs: + original_inputs = kwargs.get("inputs") + try: response = func(*args, **kwargs) return self._process_response( @@ -801,6 +851,7 @@ def log_guardrail_information(func): request_data=request_data, duration=(datetime.now() - start_time).total_seconds(), event_type=event_type, + original_inputs=original_inputs, ) except Exception as e: return self._process_error( diff --git a/litellm/integrations/gcs_bucket/gcs_bucket.py b/litellm/integrations/gcs_bucket/gcs_bucket.py index 3cb62905531..0f1ba4a4093 100644 --- a/litellm/integrations/gcs_bucket/gcs_bucket.py +++ b/litellm/integrations/gcs_bucket/gcs_bucket.py @@ -9,6 +9,7 @@ from typing import TYPE_CHECKING, Any, Dict, List, Optional, Tuple from urllib.parse import quote from litellm._logging import verbose_logger +from litellm.constants import LITELLM_ASYNCIO_QUEUE_MAXSIZE from litellm.integrations.additional_logging_utils import AdditionalLoggingUtils from litellm.integrations.gcs_bucket.gcs_bucket_base import GCSBucketBase from litellm.proxy._types import CommonProxyErrors @@ -41,7 +42,9 @@ class GCSBucketLogger(GCSBucketBase, AdditionalLoggingUtils): batch_size=self.batch_size, flush_interval=self.flush_interval, ) - self.log_queue: asyncio.Queue[GCSLogQueueItem] = asyncio.Queue() # type: ignore[assignment] + self.log_queue: asyncio.Queue[GCSLogQueueItem] = asyncio.Queue( # type: ignore[assignment] + maxsize=LITELLM_ASYNCIO_QUEUE_MAXSIZE + ) asyncio.create_task(self.periodic_flush()) AdditionalLoggingUtils.__init__(self) @@ -69,6 +72,9 @@ class GCSBucketLogger(GCSBucketBase, AdditionalLoggingUtils): ) if logging_payload is None: raise ValueError("standard_logging_object not found in kwargs") + # When queue is at maxsize, flush immediately to make room (no blocking, no data dropped) + if self.log_queue.full(): + await self.flush_queue() await self.log_queue.put( GCSLogQueueItem( payload=logging_payload, kwargs=kwargs, response_obj=response_obj @@ -91,9 +97,9 @@ class GCSBucketLogger(GCSBucketBase, AdditionalLoggingUtils): ) if logging_payload is None: raise ValueError("standard_logging_object not found in kwargs") - # Add to logging queue - this will be flushed periodically - # Use asyncio.Queue.put() for thread-safe concurrent access - # If queue is full, this will block until space is available (backpressure) + # When queue is at maxsize, flush immediately to make room (no blocking, no data dropped) + if self.log_queue.full(): + await self.flush_queue() await self.log_queue.put( GCSLogQueueItem( payload=logging_payload, kwargs=kwargs, response_obj=response_obj diff --git a/litellm/integrations/opentelemetry.py b/litellm/integrations/opentelemetry.py index 296a88f9a0b..b847180174a 100644 --- a/litellm/integrations/opentelemetry.py +++ b/litellm/integrations/opentelemetry.py @@ -5,6 +5,10 @@ from typing import TYPE_CHECKING, Any, Dict, List, Optional, Union, cast import litellm from litellm._logging import verbose_logger +from litellm.integrations._types.open_inference import ( + OpenInferenceSpanKindValues, + SpanAttributes, +) from litellm.integrations.custom_logger import CustomLogger from litellm.litellm_core_utils.safe_json_dumps import safe_dumps from litellm.secret_managers.main import get_secret_bool @@ -17,10 +21,6 @@ from litellm.types.utils import ( StandardCallbackDynamicParams, StandardLoggingPayload, ) -from litellm.integrations._types.open_inference import ( - OpenInferenceSpanKindValues, - SpanAttributes, -) # OpenTelemetry imports moved to individual functions to avoid import errors when not installed @@ -40,7 +40,9 @@ if TYPE_CHECKING: Context = Union[_Context, Any] SpanExporter = Union[_SpanExporter, Any] UserAPIKeyAuth = Union[_UserAPIKeyAuth, Any] - ManagementEndpointLoggingPayload = Union[_ManagementEndpointLoggingPayload, Any] + ManagementEndpointLoggingPayload = Union[ + _ManagementEndpointLoggingPayload, Any + ] else: Span = Any Tracer = Any @@ -70,6 +72,13 @@ class OpenTelemetryConfig: model_id: Optional[str] = None def __post_init__(self) -> None: + # If endpoint is specified but exporter is still the default "console", + # automatically infer "otlp_http" to send traces to the endpoint. + # This fixes an issue where UI-configured OTEL settings would default + # to console output instead of sending traces to the configured endpoint. + if self.endpoint and isinstance(self.exporter, str) and self.exporter == "console": + self.exporter = "otlp_http" + if not self.service_name: self.service_name = os.getenv("OTEL_SERVICE_NAME", "litellm") if not self.deployment_environment: @@ -95,12 +104,16 @@ class OpenTelemetryConfig: exporter = os.getenv( "OTEL_EXPORTER_OTLP_PROTOCOL", os.getenv("OTEL_EXPORTER", "console") ) - endpoint = os.getenv("OTEL_EXPORTER_OTLP_ENDPOINT", os.getenv("OTEL_ENDPOINT")) + endpoint = os.getenv( + "OTEL_EXPORTER_OTLP_ENDPOINT", os.getenv("OTEL_ENDPOINT") + ) headers = os.getenv( "OTEL_EXPORTER_OTLP_HEADERS", os.getenv("OTEL_HEADERS") ) # example: OTEL_HEADERS=x-honeycomb-team=B85YgLm96***" enable_metrics: bool = ( - os.getenv("LITELLM_OTEL_INTEGRATION_ENABLE_METRICS", "false").lower() + os.getenv( + "LITELLM_OTEL_INTEGRATION_ENABLE_METRICS", "false" + ).lower() == "true" ) enable_events: bool = ( @@ -108,7 +121,9 @@ class OpenTelemetryConfig: == "true" ) service_name = os.getenv("OTEL_SERVICE_NAME", "litellm") - deployment_environment = os.getenv("OTEL_ENVIRONMENT_NAME", "production") + deployment_environment = os.getenv( + "OTEL_ENVIRONMENT_NAME", "production" + ) model_id = os.getenv("OTEL_MODEL_ID", service_name) if exporter == "in_memory": @@ -157,7 +172,9 @@ class OpenTelemetry(CustomLogger): logging.getLogger(__name__) # Enable OpenTelemetry logging - otel_exporter_logger = logging.getLogger("opentelemetry.sdk.trace.export") + otel_exporter_logger = logging.getLogger( + "opentelemetry.sdk.trace.export" + ) otel_exporter_logger.setLevel(logging.DEBUG) # init CustomLogger params @@ -253,7 +270,9 @@ class OpenTelemetry(CustomLogger): # Don't call set_provider to preserve existing context else: # Default proxy provider or unknown type, create our own - verbose_logger.debug("OpenTelemetry: Creating new %s", provider_name) + verbose_logger.debug( + "OpenTelemetry: Creating new %s", provider_name + ) provider = create_new_provider_fn() set_provider_fn(provider) except Exception as e: @@ -274,7 +293,9 @@ class OpenTelemetry(CustomLogger): from opentelemetry.trace import SpanKind def create_tracer_provider(): - provider = TracerProvider(resource=self._get_litellm_resource(self.config)) + provider = TracerProvider( + resource=self._get_litellm_resource(self.config) + ) provider.add_span_processor(self._get_span_processor()) return provider @@ -388,10 +409,14 @@ class OpenTelemetry(CustomLogger): def log_failure_event(self, kwargs, response_obj, start_time, end_time): self._handle_failure(kwargs, response_obj, start_time, end_time) - async def async_log_success_event(self, kwargs, response_obj, start_time, end_time): + async def async_log_success_event( + self, kwargs, response_obj, start_time, end_time + ): self._handle_success(kwargs, response_obj, start_time, end_time) - async def async_log_failure_event(self, kwargs, response_obj, start_time, end_time): + async def async_log_failure_event( + self, kwargs, response_obj, start_time, end_time + ): self._handle_failure(kwargs, response_obj, start_time, end_time) async def async_service_success_hook( @@ -588,7 +613,9 @@ class OpenTelemetry(CustomLogger): if dynamic_headers is not None: # Create spans using a temporary tracer with dynamic headers - tracer_to_use = self._get_tracer_with_dynamic_headers(dynamic_headers) + tracer_to_use = self._get_tracer_with_dynamic_headers( + dynamic_headers + ) verbose_logger.debug( "Using dynamic headers for this request: %s", dynamic_headers ) @@ -624,7 +651,9 @@ class OpenTelemetry(CustomLogger): ) # Create a temporary tracer provider with dynamic headers - temp_provider = TracerProvider(resource=self._get_litellm_resource(self.config)) + temp_provider = TracerProvider( + resource=self._get_litellm_resource(self.config) + ) temp_provider.add_span_processor( self._get_span_processor(dynamic_headers=dynamic_headers) ) @@ -755,7 +784,9 @@ class OpenTelemetry(CustomLogger): metadata = litellm_params.get("metadata") or {} generation_name = metadata.get("generation_name") - raw_span_name = generation_name if generation_name else RAW_REQUEST_SPAN_NAME + raw_span_name = ( + generation_name if generation_name else RAW_REQUEST_SPAN_NAME + ) otel_tracer: Tracer = self.get_tracer_to_use_for_request(kwargs) raw_span = otel_tracer.start_span( @@ -780,7 +811,9 @@ class OpenTelemetry(CustomLogger): } std_log = kwargs.get("standard_logging_object") - md = getattr(std_log, "metadata", None) or (std_log or {}).get("metadata", {}) + md = getattr(std_log, "metadata", None) or (std_log or {}).get( + "metadata", {} + ) for key in [ "user_api_key_hash", "user_api_key_alias", @@ -802,9 +835,9 @@ class OpenTelemetry(CustomLogger): common_attrs[f"metadata.{key}"] = str(md[key]) # get hidden params - hidden_params = getattr(std_log, "hidden_params", None) or (std_log or {}).get( - "hidden_params", {} - ) + hidden_params = getattr(std_log, "hidden_params", None) or ( + std_log or {} + ).get("hidden_params", {}) if hidden_params: common_attrs["hidden_params"] = safe_dumps(hidden_params) @@ -838,7 +871,9 @@ class OpenTelemetry(CustomLogger): self._record_response_duration_metric(kwargs, end_time, common_attrs) @staticmethod - def _to_timestamp(val: Optional[Union[datetime, float, str]]) -> Optional[float]: + def _to_timestamp( + val: Optional[Union[datetime, float, str]], + ) -> Optional[float]: """Convert datetime/float/string to timestamp.""" if val is None: return None @@ -855,7 +890,9 @@ class OpenTelemetry(CustomLogger): except ValueError: return None - def _record_time_to_first_token_metric(self, kwargs: dict, common_attrs: dict): + def _record_time_to_first_token_metric( + self, kwargs: dict, common_attrs: dict + ): """Record Time to First Token (TTFT) metric for streaming requests.""" optional_params = kwargs.get("optional_params", {}) is_streaming = optional_params.get("stream", False) @@ -868,7 +905,10 @@ class OpenTelemetry(CustomLogger): api_call_start_time = kwargs.get("api_call_start_time", None) completion_start_time = kwargs.get("completion_start_time", None) - if api_call_start_time is not None and completion_start_time is not None: + if ( + api_call_start_time is not None + and completion_start_time is not None + ): # Convert to timestamps if needed (handles datetime, float, and string) api_call_start_ts = self._to_timestamp(api_call_start_time) completion_start_ts = self._to_timestamp(completion_start_time) @@ -876,7 +916,9 @@ class OpenTelemetry(CustomLogger): if api_call_start_ts is None or completion_start_ts is None: return # Skip recording if conversion failed - time_to_first_token_seconds = completion_start_ts - api_call_start_ts + time_to_first_token_seconds = ( + completion_start_ts - api_call_start_ts + ) self._time_to_first_token_histogram.record( time_to_first_token_seconds, attributes=common_attrs ) @@ -946,7 +988,9 @@ class OpenTelemetry(CustomLogger): generation_time_seconds = duration_s if generation_time_seconds > 0: - time_per_output_token_seconds = generation_time_seconds / completion_tokens + time_per_output_token_seconds = ( + generation_time_seconds / completion_tokens + ) self._time_per_output_token_histogram.record( time_per_output_token_seconds, attributes=common_attrs ) @@ -1007,21 +1051,26 @@ class OpenTelemetry(CustomLogger): # See: https://github.com/open-telemetry/opentelemetry-python/pull/4676 # TODO: Refactor to use the proper OTEL Logs API instead of directly creating SDK LogRecords - from opentelemetry._logs import SeverityNumber, get_logger, get_logger_provider + from opentelemetry._logs import ( + SeverityNumber, + get_logger, + ) - try: - from opentelemetry.sdk._logs import LogRecord as SdkLogRecord # type: ignore[attr-defined] # OTEL < 1.39.0 - except ImportError: - from opentelemetry.sdk._logs._internal import LogRecord as SdkLogRecord # type: ignore[attr-defined, no-redef] # OTEL >= 1.39.0 + # MyPy evaluates both branches of try/except imports and can fail when + # newer OTEL stubs remove/relocate symbols. Gate the typing import so + # only the canonical location is type-checked. + if TYPE_CHECKING: + from opentelemetry.sdk._logs._internal import LogRecord as SdkLogRecord + else: + try: + from opentelemetry.sdk._logs import ( + LogRecord as SdkLogRecord, # type: ignore[attr-defined] + ) + except ImportError: + from opentelemetry.sdk._logs._internal import LogRecord as SdkLogRecord otel_logger = get_logger(LITELLM_LOGGER_NAME) - # Get the resource from the logger provider - logger_provider = get_logger_provider() - resource = getattr( - logger_provider, "_resource", None - ) or self._get_litellm_resource(self.config) - parent_ctx = span.get_span_context() provider = (kwargs.get("litellm_params") or {}).get( "custom_llm_provider", "Unknown" @@ -1030,7 +1079,10 @@ class OpenTelemetry(CustomLogger): # per-message events for msg in kwargs.get("messages", []): role = msg.get("role", "user") - attrs = {"event_name": "gen_ai.content.prompt", "gen_ai.system": provider} + attrs = { + "event_name": "gen_ai.content.prompt", + "gen_ai.system": provider, + } if role == "tool" and msg.get("id"): attrs["id"] = msg["id"] if self.message_logging and msg.get("content"): @@ -1044,7 +1096,6 @@ class OpenTelemetry(CustomLogger): severity_number=SeverityNumber.INFO, severity_text="INFO", body=msg.copy(), - resource=resource, attributes=attrs, ) otel_logger.emit(log_record) @@ -1076,7 +1127,6 @@ class OpenTelemetry(CustomLogger): severity_number=SeverityNumber.INFO, severity_text="INFO", body=body, - resource=resource, attributes=attrs, ) otel_logger.emit(log_record) @@ -1146,7 +1196,9 @@ class OpenTelemetry(CustomLogger): value=guardrail_information.get("guardrail_mode"), ) - masked_entity_count = guardrail_information.get("masked_entity_count") + masked_entity_count = guardrail_information.get( + "masked_entity_count" + ) if masked_entity_count is not None: guardrail_span.set_attribute( "masked_entity_count", safe_dumps(masked_entity_count) @@ -1173,8 +1225,9 @@ class OpenTelemetry(CustomLogger): # Decide whether to create a primary span # Always create if no parent span exists (backward compatibility) # OR if USE_OTEL_LITELLM_REQUEST_SPAN is explicitly enabled - should_create_primary_span = parent_otel_span is None or get_secret_bool( - "USE_OTEL_LITELLM_REQUEST_SPAN" + should_create_primary_span = ( + parent_otel_span is None + or get_secret_bool("USE_OTEL_LITELLM_REQUEST_SPAN") ) if should_create_primary_span: @@ -1200,7 +1253,9 @@ class OpenTelemetry(CustomLogger): if parent_otel_span.is_recording(): parent_otel_span.set_status(Status(StatusCode.ERROR)) self.set_attributes(parent_otel_span, kwargs, response_obj) - self._record_exception_on_span(span=parent_otel_span, kwargs=kwargs) + self._record_exception_on_span( + span=parent_otel_span, kwargs=kwargs + ) # Create span for guardrail information self._create_guardrail_span(kwargs=kwargs, context=_parent_context) @@ -1223,7 +1278,9 @@ class OpenTelemetry(CustomLogger): 2. Sets structured error attributes from StandardLoggingPayloadErrorInformation """ try: - from litellm.integrations._types.open_inference import ErrorAttributes + from litellm.integrations._types.open_inference import ( + ErrorAttributes, + ) # Get the exception object if available exception = kwargs.get("exception") @@ -1233,15 +1290,17 @@ class OpenTelemetry(CustomLogger): span.record_exception(exception) # Get StandardLoggingPayload for structured error information - standard_logging_payload: Optional[StandardLoggingPayload] = kwargs.get( - "standard_logging_object" + standard_logging_payload: Optional[StandardLoggingPayload] = ( + kwargs.get("standard_logging_object") ) if standard_logging_payload is None: return # Extract error_information from StandardLoggingPayload - error_information = standard_logging_payload.get("error_information") + error_information = standard_logging_payload.get( + "error_information" + ) if error_information is None: # Fallback to error_str if error_information is not available @@ -1331,7 +1390,9 @@ class OpenTelemetry(CustomLogger): ) pass - def cast_as_primitive_value_type(self, value) -> Union[str, bool, int, float]: + def cast_as_primitive_value_type( + self, value + ) -> Union[str, bool, int, float]: """ Casts the value to a primitive OTEL type if it is not already a primitive type. @@ -1401,8 +1462,8 @@ class OpenTelemetry(CustomLogger): optional_params = kwargs.get("optional_params", {}) litellm_params = kwargs.get("litellm_params", {}) or {} - standard_logging_payload: Optional[StandardLoggingPayload] = kwargs.get( - "standard_logging_object" + standard_logging_payload: Optional[StandardLoggingPayload] = ( + kwargs.get("standard_logging_object") ) if standard_logging_payload is None: raise ValueError("standard_logging_object not found in kwargs") @@ -1424,11 +1485,13 @@ class OpenTelemetry(CustomLogger): ) or (standard_logging_payload or {}).get("hidden_params", {}) if hidden_params: self.safe_set_attribute( - span=span, key="hidden_params", value=safe_dumps(hidden_params) + span=span, + key="hidden_params", + value=safe_dumps(hidden_params), ) # Cost breakdown tracking - cost_breakdown: Optional[CostBreakdown] = standard_logging_payload.get( - "cost_breakdown" + cost_breakdown: Optional[CostBreakdown] = ( + standard_logging_payload.get("cost_breakdown") ) if cost_breakdown: for key, value in cost_breakdown.items(): @@ -1504,7 +1567,9 @@ class OpenTelemetry(CustomLogger): # The unique identifier for the completion. if response_obj and response_obj.get("id"): self.safe_set_attribute( - span=span, key="gen_ai.response.id", value=response_obj.get("id") + span=span, + key="gen_ai.response.id", + value=response_obj.get("id"), ) # The model used to generate the response. @@ -1639,7 +1704,9 @@ class OpenTelemetry(CustomLogger): "OpenTelemetry logging error in set_attributes %s", str(e) ) - def _cast_as_primitive_value_type(self, value) -> Union[str, bool, int, float]: + def _cast_as_primitive_value_type( + self, value + ) -> Union[str, bool, int, float]: """ Casts the value to a primitive OTEL type if it is not already a primitive type. @@ -1673,7 +1740,10 @@ class OpenTelemetry(CustomLogger): if isinstance(messages, str): # Handle system_instructions passed as a string return [ - {"role": "system", "parts": [{"type": "text", "content": messages}]} + { + "role": "system", + "parts": [{"type": "text", "content": messages}], + } ] transformed = [] @@ -1714,9 +1784,11 @@ class OpenTelemetry(CustomLogger): message = choice.get("message") or {} finish_reason = choice.get("finish_reason") - transformed_msg = self._transform_messages_to_otel_semantic_conventions( - [message] - )[0] + transformed_msg = ( + self._transform_messages_to_otel_semantic_conventions( + [message] + )[0] + ) if finish_reason: transformed_msg["finish_reason"] = finish_reason @@ -1728,7 +1800,9 @@ class OpenTelemetry(CustomLogger): self.set_attributes(span, kwargs, response_obj) kwargs.get("optional_params", {}) litellm_params = kwargs.get("litellm_params", {}) or {} - custom_llm_provider = litellm_params.get("custom_llm_provider", "Unknown") + custom_llm_provider = litellm_params.get( + "custom_llm_provider", "Unknown" + ) _raw_response = kwargs.get("original_response") _additional_args = kwargs.get("additional_args", {}) or {} @@ -1741,7 +1815,9 @@ class OpenTelemetry(CustomLogger): if complete_input_dict and isinstance(complete_input_dict, dict): for param, val in complete_input_dict.items(): self.safe_set_attribute( - span=span, key=f"llm.{custom_llm_provider}.{param}", value=val + span=span, + key=f"llm.{custom_llm_provider}.{param}", + value=val, ) ############################################# @@ -1773,7 +1849,8 @@ class OpenTelemetry(CustomLogger): ) except Exception as e: verbose_logger.exception( - "OpenTelemetry logging error in set_raw_request_attributes %s", str(e) + "OpenTelemetry logging error in set_raw_request_attributes %s", + str(e), ) def _to_ns(self, dt): @@ -1813,7 +1890,9 @@ class OpenTelemetry(CustomLogger): ) litellm_params = kwargs.get("litellm_params", {}) or {} - proxy_server_request = litellm_params.get("proxy_server_request", {}) or {} + proxy_server_request = ( + litellm_params.get("proxy_server_request", {}) or {} + ) headers = proxy_server_request.get("headers", {}) or {} traceparent = headers.get("traceparent", None) _metadata = litellm_params.get("metadata", {}) or {} @@ -1832,7 +1911,10 @@ class OpenTelemetry(CustomLogger): "OpenTelemetry: Using traceparent header for context propagation" ) carrier = {"traceparent": traceparent} - return TraceContextTextMapPropagator().extract(carrier=carrier), None + return ( + TraceContextTextMapPropagator().extract(carrier=carrier), + None, + ) # Priority 3: Active span from global context (auto-detection) try: @@ -1960,10 +2042,14 @@ class OpenTelemetry(CustomLogger): self.OTEL_HEADERS, ) - _split_otel_headers = OpenTelemetry._get_headers_dictionary(self.OTEL_HEADERS) + _split_otel_headers = OpenTelemetry._get_headers_dictionary( + self.OTEL_HEADERS + ) # Normalize endpoint for logs - ensure it points to /v1/logs instead of /v1/traces - normalized_endpoint = self._normalize_otel_endpoint(self.OTEL_ENDPOINT, "logs") + normalized_endpoint = self._normalize_otel_endpoint( + self.OTEL_ENDPOINT, "logs" + ) verbose_logger.debug( "OpenTelemetry: Log endpoint normalized from %s to %s", @@ -2051,14 +2137,18 @@ class OpenTelemetry(CustomLogger): self.OTEL_HEADERS, ) - _split_otel_headers = OpenTelemetry._get_headers_dictionary(self.OTEL_HEADERS) + _split_otel_headers = OpenTelemetry._get_headers_dictionary( + self.OTEL_HEADERS + ) normalized_endpoint = self._normalize_otel_endpoint( self.OTEL_ENDPOINT, "metrics" ) if self.OTEL_EXPORTER == "console": exporter = ConsoleMetricExporter() - return PeriodicExportingMetricReader(exporter, export_interval_millis=5000) + return PeriodicExportingMetricReader( + exporter, export_interval_millis=5000 + ) elif ( self.OTEL_EXPORTER == "otlp_http" @@ -2074,7 +2164,9 @@ class OpenTelemetry(CustomLogger): headers=_split_otel_headers, preferred_temporality={Histogram: AggregationTemporality.DELTA}, ) - return PeriodicExportingMetricReader(exporter, export_interval_millis=5000) + return PeriodicExportingMetricReader( + exporter, export_interval_millis=5000 + ) elif self.OTEL_EXPORTER == "otlp_grpc" or self.OTEL_EXPORTER == "grpc": try: @@ -2092,7 +2184,9 @@ class OpenTelemetry(CustomLogger): headers=_split_otel_headers, preferred_temporality={Histogram: AggregationTemporality.DELTA}, ) - return PeriodicExportingMetricReader(exporter, export_interval_millis=5000) + return PeriodicExportingMetricReader( + exporter, export_interval_millis=5000 + ) else: verbose_logger.warning( @@ -2100,7 +2194,9 @@ class OpenTelemetry(CustomLogger): self.OTEL_EXPORTER, ) exporter = ConsoleMetricExporter() - return PeriodicExportingMetricReader(exporter, export_interval_millis=5000) + return PeriodicExportingMetricReader( + exporter, export_interval_millis=5000 + ) def _normalize_otel_endpoint( self, endpoint: Optional[str], signal_type: str @@ -2171,7 +2267,9 @@ class OpenTelemetry(CustomLogger): return endpoint @staticmethod - def _get_headers_dictionary(headers: Optional[Union[str, dict]]) -> Dict[str, str]: + def _get_headers_dictionary( + headers: Optional[Union[str, dict]], + ) -> Dict[str, str]: """ Convert a string or dictionary of headers into a dictionary of headers. """ diff --git a/litellm/litellm_core_utils/get_model_cost_map.py b/litellm/litellm_core_utils/get_model_cost_map.py index 9b86f4ca2f0..e622a317454 100644 --- a/litellm/litellm_core_utils/get_model_cost_map.py +++ b/litellm/litellm_core_utils/get_model_cost_map.py @@ -8,40 +8,187 @@ export LITELLM_LOCAL_MODEL_COST_MAP=True ``` """ +import json import os +from importlib.resources import files import httpx +from litellm import verbose_logger +from litellm.constants import ( + MODEL_COST_MAP_MAX_SHRINK_RATIO, + MODEL_COST_MAP_MIN_MODEL_COUNT, +) + + +class GetModelCostMap: + """ + Handles fetching, validating, and loading the model cost map. + + Only the backup model *count* is cached (a single int). The full + backup dict is never held in memory — it is only parsed when it + needs to be *returned* as a fallback. + """ + + _backup_model_count: int = -1 # -1 = not yet loaded + + @staticmethod + def load_local_model_cost_map() -> dict: + """Load the local backup model cost map bundled with the package.""" + content = json.loads( + files("litellm") + .joinpath("model_prices_and_context_window_backup.json") + .read_text(encoding="utf-8") + ) + return content + + @classmethod + def _get_backup_model_count(cls) -> int: + """Return the number of models in the local backup (cached int).""" + if cls._backup_model_count < 0: + backup = cls.load_local_model_cost_map() + cls._backup_model_count = len(backup) + return cls._backup_model_count + + @staticmethod + def _check_is_valid_dict(fetched_map: dict) -> bool: + """Check 1: fetched map is a non-empty dict.""" + if not isinstance(fetched_map, dict): + verbose_logger.warning( + "LiteLLM: Fetched model cost map is not a dict (type=%s). " + "Falling back to local backup.", + type(fetched_map).__name__, + ) + return False + + if len(fetched_map) == 0: + verbose_logger.warning( + "LiteLLM: Fetched model cost map is empty. " + "Falling back to local backup.", + ) + return False + + return True + + @classmethod + def _check_model_count_not_reduced( + cls, + fetched_map: dict, + backup_model_count: int, + min_model_count: int = MODEL_COST_MAP_MIN_MODEL_COUNT, + max_shrink_ratio: float = MODEL_COST_MAP_MAX_SHRINK_RATIO, + ) -> bool: + """Check 2: model count has not reduced significantly vs backup.""" + fetched_count = len(fetched_map) + + if fetched_count < min_model_count: + verbose_logger.warning( + "LiteLLM: Fetched model cost map has only %d models (minimum=%d). " + "This may indicate a corrupted upstream file. " + "Falling back to local backup.", + fetched_count, + min_model_count, + ) + return False + + if backup_model_count > 0 and fetched_count < backup_model_count * max_shrink_ratio: + verbose_logger.warning( + "LiteLLM: Fetched model cost map shrank significantly " + "(fetched=%d, backup=%d, threshold=%.0f%%). " + "This may indicate a corrupted upstream file. " + "Falling back to local backup.", + fetched_count, + backup_model_count, + max_shrink_ratio * 100, + ) + return False + + return True + + @classmethod + def validate_model_cost_map( + cls, + fetched_map: dict, + backup_model_count: int, + min_model_count: int = MODEL_COST_MAP_MIN_MODEL_COUNT, + max_shrink_ratio: float = MODEL_COST_MAP_MAX_SHRINK_RATIO, + ) -> bool: + """ + Validate the integrity of a fetched model cost map. + + Runs each check in order and returns False on the first failure. + + Checks: + 1. ``_check_is_valid_dict`` -- fetched map is a non-empty dict. + 2. ``_check_model_count_not_reduced`` -- model count meets minimum + and has not shrunk >``max_shrink_ratio`` vs backup. + + Returns True if all checks pass, False otherwise. + """ + if not cls._check_is_valid_dict(fetched_map): + return False + + if not cls._check_model_count_not_reduced( + fetched_map=fetched_map, + backup_model_count=backup_model_count, + min_model_count=min_model_count, + max_shrink_ratio=max_shrink_ratio, + ): + return False + + return True + + @staticmethod + def fetch_remote_model_cost_map(url: str, timeout: int = 5) -> dict: + """ + Fetch the model cost map from a remote URL. + + Returns the parsed JSON dict. Raises on network/parse errors + (caller is expected to handle). + """ + response = httpx.get(url, timeout=timeout) + response.raise_for_status() + return response.json() + def get_model_cost_map(url: str) -> dict: - if ( - os.getenv("LITELLM_LOCAL_MODEL_COST_MAP", False) - or os.getenv("LITELLM_LOCAL_MODEL_COST_MAP", False) == "True" - ): - from importlib.resources import files - import json + """ + Public entry point — returns the model cost map dict. - content = json.loads( - files("litellm") - .joinpath("model_prices_and_context_window_backup.json") - .read_text(encoding="utf-8") - ) - return content + 1. If ``LITELLM_LOCAL_MODEL_COST_MAP`` is set, uses the local backup only. + 2. Otherwise fetches from ``url``, validates integrity, and falls back + to the local backup on any failure. + + Only the backup model count is cached (a single int) for validation. + The full backup dict is only parsed when it must be *returned* as a + fallback — it is never held in memory long-term. + """ + # Note: can't use get_secret_bool here — this runs during litellm.__init__ + # before litellm._key_management_settings is set. + if os.getenv("LITELLM_LOCAL_MODEL_COST_MAP", "").lower() == "true": + return GetModelCostMap.load_local_model_cost_map() try: - response = httpx.get( - url, timeout=5 - ) # set a 5 second timeout for the get request - response.raise_for_status() # Raise an exception if the request is unsuccessful - content = response.json() - return content - except Exception: - from importlib.resources import files - import json - - content = json.loads( - files("litellm") - .joinpath("model_prices_and_context_window_backup.json") - .read_text(encoding="utf-8") + content = GetModelCostMap.fetch_remote_model_cost_map(url) + except Exception as e: + verbose_logger.warning( + "LiteLLM: Failed to fetch remote model cost map from %s: %s. " + "Falling back to local backup.", + url, + str(e), ) - return content + return GetModelCostMap.load_local_model_cost_map() + + # Validate using cached count (cheap int comparison, no file I/O) + if not GetModelCostMap.validate_model_cost_map( + fetched_map=content, + backup_model_count=GetModelCostMap._get_backup_model_count(), + ): + verbose_logger.warning( + "LiteLLM: Fetched model cost map failed integrity check. " + "Using local backup instead. url=%s", + url, + ) + return GetModelCostMap.load_local_model_cost_map() + + return content diff --git a/litellm/litellm_core_utils/prompt_templates/common_utils.py b/litellm/litellm_core_utils/prompt_templates/common_utils.py index b1c2d0a52f5..cdddee4e54e 100644 --- a/litellm/litellm_core_utils/prompt_templates/common_utils.py +++ b/litellm/litellm_core_utils/prompt_templates/common_utils.py @@ -1272,3 +1272,59 @@ def parse_tool_call_arguments( ) raise ValueError(error_message) from e + + +def split_concatenated_json_objects(raw: str) -> List[Dict[str, Any]]: + """ + Split a string that contains one or more concatenated JSON objects into + a list of parsed dicts. + + LLM providers (notably Bedrock Claude Sonnet 4.5) sometimes return + multiple tool-call argument objects concatenated in a single + ``arguments`` string, e.g.:: + + '{"command":["curl",...]}{"command":["curl",...]}{"command":["curl",...]}' + + ``json.loads()`` fails on this with ``JSONDecodeError: Extra data``. + This helper uses ``json.JSONDecoder.raw_decode()`` to walk the string + and extract each JSON object individually. + + Returns + ------- + list[dict] + A list of parsed dicts – one per JSON object found. If *raw* is + empty or whitespace-only, an empty list is returned. + + Raises + ------ + json.JSONDecodeError + If the string contains text that cannot be parsed as JSON at all. + """ + import json + + raw = raw.strip() + if not raw: + return [] + + decoder = json.JSONDecoder() + results: List[Dict[str, Any]] = [] + idx = 0 + length = len(raw) + + while idx < length: + # Skip whitespace between objects + while idx < length and raw[idx] in " \t\n\r": + idx += 1 + if idx >= length: + break + + obj, end_idx = decoder.raw_decode(raw, idx) + if isinstance(obj, dict): + results.append(obj) + else: + # Non-dict JSON value – wrap in empty dict (Bedrock requires + # toolUse.input to be an object). + results.append({}) + idx = end_idx + + return results diff --git a/litellm/litellm_core_utils/prompt_templates/factory.py b/litellm/litellm_core_utils/prompt_templates/factory.py index f9ecd78ff1c..c907ed32b95 100644 --- a/litellm/litellm_core_utils/prompt_templates/factory.py +++ b/litellm/litellm_core_utils/prompt_templates/factory.py @@ -3287,25 +3287,68 @@ def _convert_to_bedrock_tool_call_invoke( - extract name - extract id """ + from litellm.litellm_core_utils.prompt_templates.common_utils import ( + split_concatenated_json_objects, + ) try: _parts_list: List[BedrockContentBlock] = [] for tool in tool_calls: if "function" in tool: - id = tool["id"] + tool_id = tool["id"] name = tool["function"].get("name", "") arguments = tool["function"].get("arguments", "") - arguments_dict = json.loads(arguments) if arguments else {} - # Ensure arguments_dict is always a dict (Bedrock requires toolUse.input to be an object) - # When some providers return arguments: '""' (JSON-encoded empty string), json.loads returns "" - if not isinstance(arguments_dict, dict): - arguments_dict = {} + if not arguments or not arguments.strip(): arguments_dict = {} else: - arguments_dict = json.loads(arguments) + try: + arguments_dict = json.loads(arguments) + # Ensure arguments_dict is always a dict + # (Bedrock requires toolUse.input to be an object). + # Some providers return arguments: '""' which + # json.loads decodes to a bare string. + if not isinstance(arguments_dict, dict): + arguments_dict = {} + except json.JSONDecodeError: + # The model may return multiple JSON objects + # concatenated in a single arguments string, e.g. + # '{"cmd":"a"}{"cmd":"b"}{"cmd":"c"}' + # Split them and emit one toolUse block per object. + # Fixes: https://github.com/BerriAI/litellm/issues/20543 + parsed_objects = split_concatenated_json_objects( + arguments + ) + if parsed_objects: + # First object keeps the original tool id. + for obj_idx, obj in enumerate(parsed_objects): + block_id = ( + tool_id + if obj_idx == 0 + else f"{tool_id}_{obj_idx}" + ) + bedrock_tool = BedrockToolUseBlock( + input=obj, name=name, toolUseId=block_id + ) + _parts_list.append( + BedrockContentBlock(toolUse=bedrock_tool) + ) + # cache_control applies to the whole original + # tool call; attach after the last split block. + if tool.get("cache_control", None) is not None: + _parts_list.append( + BedrockContentBlock( + cachePoint=CachePointBlock( + type="default" + ) + ) + ) + continue + # Fallback: no objects extracted — use empty dict. + arguments_dict = {} + bedrock_tool = BedrockToolUseBlock( - input=arguments_dict, name=name, toolUseId=id + input=arguments_dict, name=name, toolUseId=tool_id ) bedrock_content_block = BedrockContentBlock(toolUse=bedrock_tool) _parts_list.append(bedrock_content_block) diff --git a/litellm/llms/anthropic/experimental_pass_through/adapters/handler.py b/litellm/llms/anthropic/experimental_pass_through/adapters/handler.py index a17eba75b3b..c6caaddf98b 100644 --- a/litellm/llms/anthropic/experimental_pass_through/adapters/handler.py +++ b/litellm/llms/anthropic/experimental_pass_through/adapters/handler.py @@ -30,6 +30,58 @@ ANTHROPIC_ADAPTER = AnthropicAdapter() class LiteLLMMessagesToCompletionTransformationHandler: + @staticmethod + def _route_openai_thinking_to_responses_api_if_needed( + completion_kwargs: Dict[str, Any], + *, + thinking: Optional[Dict[str, Any]], + ) -> None: + """ + When users call `litellm.anthropic.messages.*` with a non-Anthropic model and + `thinking={"type": "enabled", ...}`, LiteLLM converts this into OpenAI + `reasoning_effort`. + + For OpenAI models, Chat Completions typically does not return reasoning text + (only token accounting). To return a thinking-like content block in the + Anthropic response format, we route the request through OpenAI's Responses API + and request a reasoning summary. + """ + custom_llm_provider = completion_kwargs.get("custom_llm_provider") + if custom_llm_provider is None: + try: + _, inferred_provider, _, _ = litellm.utils.get_llm_provider( + model=cast(str, completion_kwargs.get("model")) + ) + custom_llm_provider = inferred_provider + except Exception: + custom_llm_provider = None + + if custom_llm_provider != "openai": + return + + if not isinstance(thinking, dict) or thinking.get("type") != "enabled": + return + + model = completion_kwargs.get("model") + if isinstance(model, str) and model and not model.startswith("responses/"): + # Prefix model with "responses/" to route to OpenAI Responses API + completion_kwargs["model"] = f"responses/{model}" + + reasoning_effort = completion_kwargs.get("reasoning_effort") + if isinstance(reasoning_effort, str) and reasoning_effort: + completion_kwargs["reasoning_effort"] = { + "effort": reasoning_effort, + "summary": "detailed", + } + elif isinstance(reasoning_effort, dict): + if ( + "summary" not in reasoning_effort + and "generate_summary" not in reasoning_effort + ): + updated_reasoning_effort = dict(reasoning_effort) + updated_reasoning_effort["summary"] = "detailed" + completion_kwargs["reasoning_effort"] = updated_reasoning_effort + @staticmethod def _prepare_completion_kwargs( *, @@ -123,6 +175,11 @@ class LiteLLMMessagesToCompletionTransformationHandler: ): completion_kwargs[key] = value + LiteLLMMessagesToCompletionTransformationHandler._route_openai_thinking_to_responses_api_if_needed( + completion_kwargs, + thinking=thinking, + ) + return completion_kwargs, tool_name_mapping @staticmethod diff --git a/litellm/llms/custom_httpx/aiohttp_transport.py b/litellm/llms/custom_httpx/aiohttp_transport.py index a7b83d8c802..1b03ec47643 100644 --- a/litellm/llms/custom_httpx/aiohttp_transport.py +++ b/litellm/llms/custom_httpx/aiohttp_transport.py @@ -1,6 +1,7 @@ import asyncio import contextlib import os +import ssl import typing import urllib.request from typing import Callable, Dict, Optional, Union @@ -139,8 +140,13 @@ class LiteLLMAiohttpTransport(AiohttpTransport): Credit to: https://github.com/karpetrosyan/httpx-aiohttp for this implementation """ - def __init__(self, client: Union[ClientSession, Callable[[], ClientSession]]): + def __init__( + self, + client: Union[ClientSession, Callable[[], ClientSession]], + ssl_verify: Optional[Union[bool, ssl.SSLContext]] = None, + ): self.client = client + self._ssl_verify = ssl_verify # Store for per-request SSL override super().__init__(client=client) # Store the client factory for recreating sessions when needed if callable(client): @@ -214,6 +220,7 @@ class LiteLLMAiohttpTransport(AiohttpTransport): timeout: dict, proxy: Optional[str], sni_hostname: Optional[str], + ssl_verify: Optional[Union[bool, ssl.SSLContext]] = None, ) -> ClientResponse: """ Helper function to make an aiohttp request with the given parameters. @@ -224,6 +231,7 @@ class LiteLLMAiohttpTransport(AiohttpTransport): timeout: Timeout settings dict with 'connect', 'read', 'pool' keys proxy: Optional proxy URL sni_hostname: Optional SNI hostname for SSL + ssl_verify: Optional SSL verification setting (False to disable, SSLContext for custom) Returns: ClientResponse from aiohttp @@ -237,6 +245,13 @@ class LiteLLMAiohttpTransport(AiohttpTransport): data = request.stream # type: ignore request.headers.pop("transfer-encoding", None) # handled by aiohttp + # Only pass ssl kwarg when explicitly configured, to avoid + # overriding the session/connector defaults with None (which is + # not a valid value for aiohttp's ssl parameter). + ssl_kwargs: Dict[str, Union[bool, ssl.SSLContext]] = {} + if ssl_verify is not None: + ssl_kwargs["ssl"] = ssl_verify + response = await client_session.request( method=request.method, url=YarlURL(str(request.url), encoded=True), @@ -251,6 +266,7 @@ class LiteLLMAiohttpTransport(AiohttpTransport): ), proxy=proxy, server_hostname=sni_hostname, + **ssl_kwargs, ).__aenter__() return response @@ -268,6 +284,9 @@ class LiteLLMAiohttpTransport(AiohttpTransport): # Resolve proxy settings from environment variables proxy = await self._get_proxy_settings(request) + # Use stored SSL configuration for per-request override + ssl_config = self._ssl_verify + try: with map_aiohttp_exceptions(): response = await self._make_aiohttp_request( @@ -276,6 +295,7 @@ class LiteLLMAiohttpTransport(AiohttpTransport): timeout=timeout, proxy=proxy, sni_hostname=sni_hostname, + ssl_verify=ssl_config, ) except RuntimeError as e: # Handle the case where session was closed between our check and actual use @@ -296,6 +316,7 @@ class LiteLLMAiohttpTransport(AiohttpTransport): timeout=timeout, proxy=proxy, sni_hostname=sni_hostname, + ssl_verify=ssl_config, ) else: # Re-raise if it's a different RuntimeError diff --git a/litellm/llms/custom_httpx/http_handler.py b/litellm/llms/custom_httpx/http_handler.py index ac9dd5998e2..95f411c397c 100644 --- a/litellm/llms/custom_httpx/http_handler.py +++ b/litellm/llms/custom_httpx/http_handler.py @@ -846,6 +846,16 @@ class AsyncHTTPHandler: if str_to_bool(os.getenv("AIOHTTP_TRUST_ENV", "False")) is True: trust_env = True + ######################################################### + # Determine SSL config to pass to transport for per-request override + # This ensures ssl_verify works even with shared sessions + ######################################################### + ssl_for_transport: Optional[Union[bool, ssl.SSLContext]] = None + if ssl_context is not None: + ssl_for_transport = ssl_context + elif ssl_verify is False: + ssl_for_transport = False + verbose_logger.debug("Creating AiohttpTransport...") # Use shared session if provided and valid @@ -853,7 +863,10 @@ class AsyncHTTPHandler: verbose_logger.debug( f"SHARED SESSION: Reusing existing ClientSession (ID: {id(shared_session)})" ) - return LiteLLMAiohttpTransport(client=shared_session) + return LiteLLMAiohttpTransport( + client=shared_session, + ssl_verify=ssl_for_transport, + ) # Create new session only if none provided or existing one is invalid verbose_logger.debug( @@ -877,6 +890,7 @@ class AsyncHTTPHandler: connector=TCPConnector(**transport_connector_kwargs), trust_env=trust_env, ), + ssl_verify=ssl_for_transport, ) @staticmethod diff --git a/litellm/llms/databricks/chat/transformation.py b/litellm/llms/databricks/chat/transformation.py index e9ae94307d4..7c2a9569c58 100644 --- a/litellm/llms/databricks/chat/transformation.py +++ b/litellm/llms/databricks/chat/transformation.py @@ -60,6 +60,38 @@ from ...anthropic.chat.transformation import AnthropicConfig from ...openai_like.chat.transformation import OpenAILikeChatConfig from ..common_utils import DatabricksBase, DatabricksException +def _sanitize_empty_content(message_dict: dict[str, Any]) -> None: + """ + Remove or filter content so empty text blocks are not sent. + Databricks Model Serving uses Anthropic Messages API spec and rejects empty text blocks. + """ + content = message_dict.get("content") + if content is None: + message_dict.pop("content", None) + return + if isinstance(content, str): + if not content.strip(): + message_dict.pop("content") + return + if isinstance(content, list): + if not content: + message_dict.pop("content") + return + filtered = [ + block + for block in content + if not ( + isinstance(block, dict) + and block.get("type") == "text" + and not (block.get("text") or "").strip() + ) + ] + if not filtered: + message_dict.pop("content") + else: + message_dict["content"] = filtered + + if TYPE_CHECKING: from litellm.litellm_core_utils.litellm_logging import Logging as _LiteLLMLoggingObj @@ -350,6 +382,7 @@ class DatabricksConfig(DatabricksBase, OpenAILikeChatConfig, AnthropicConfig): # Move message-level cache_control into a content block when content is a string. if "cache_control" in _message and isinstance(_message.get("content"), str): _message = self._move_cache_control_into_string_content_block(_message) + _sanitize_empty_content(cast(dict[str, Any], _message)) new_messages.append(_message) if is_async: diff --git a/litellm/llms/ollama/chat/transformation.py b/litellm/llms/ollama/chat/transformation.py index 8c98cc54050..bc5aa654aad 100644 --- a/litellm/llms/ollama/chat/transformation.py +++ b/litellm/llms/ollama/chat/transformation.py @@ -502,13 +502,12 @@ class OllamaChatCompletionResponseIterator(BaseModelResponseIterator): reasoning_content: Optional[str] = None content: Optional[str] = None if chunk["message"].get("thinking") is not None: - if self.started_reasoning_content is False: - reasoning_content = chunk["message"].get("thinking") - self.started_reasoning_content = True - elif self.finished_reasoning_content is False: - reasoning_content = chunk["message"].get("thinking") - self.finished_reasoning_content = True + reasoning_content = chunk["message"].get("thinking") + self.started_reasoning_content = True elif chunk["message"].get("content") is not None: + if self.started_reasoning_content and not self.finished_reasoning_content: + self.finished_reasoning_content = True + message_content = chunk["message"].get("content") if "" in message_content: message_content = message_content.replace("", "") diff --git a/litellm/llms/perplexity/responses/__init__.py b/litellm/llms/perplexity/responses/__init__.py new file mode 100644 index 00000000000..9bdf810e839 --- /dev/null +++ b/litellm/llms/perplexity/responses/__init__.py @@ -0,0 +1,7 @@ +""" +Perplexity Agentic Research API (Responses API) module +""" + +from .transformation import PerplexityResponsesConfig + +__all__ = ["PerplexityResponsesConfig"] diff --git a/litellm/llms/perplexity/responses/transformation.py b/litellm/llms/perplexity/responses/transformation.py new file mode 100644 index 00000000000..178e76ea970 --- /dev/null +++ b/litellm/llms/perplexity/responses/transformation.py @@ -0,0 +1,409 @@ +""" +Transformation logic for Perplexity Agentic Research API (Responses API) + +This module handles the translation between OpenAI's Responses API format +and Perplexity's Responses API format, which supports: +- Third-party model access (OpenAI, Anthropic, Google, xAI, etc.) +- Presets for optimized configurations +- Web search and URL fetching tools +- Reasoning effort control +- Instructions parameter for system-level guidance +""" + +from typing import Any, Dict, List, Optional, Union + +import httpx + +from litellm._logging import verbose_logger +from litellm.litellm_core_utils.litellm_logging import Logging as LiteLLMLoggingObj +from litellm.llms.base_llm.chat.transformation import BaseLLMException +from litellm.llms.openai.responses.transformation import OpenAIResponsesAPIConfig +from litellm.secret_managers.main import get_secret_str +from litellm.types.llms.openai import ( + ResponseAPIUsage, + ResponseInputParam, + ResponsesAPIOptionalRequestParams, + ResponsesAPIResponse, + ResponsesAPIStreamingResponse, +) +from litellm.types.router import GenericLiteLLMParams +from litellm.types.utils import LlmProviders + + +class PerplexityResponsesConfig(OpenAIResponsesAPIConfig): + """ + Configuration for Perplexity Agentic Research API (Responses API) + + + Reference: https://docs.perplexity.ai/agentic-research/quickstart + """ + + @property + def custom_llm_provider(self) -> LlmProviders: + return LlmProviders.PERPLEXITY + + def get_supported_openai_params(self, model: str) -> list: + """ + Perplexity Responses API supports a different set of parameters + + Ref: https://docs.perplexity.ai/api-reference/responses-post + """ + return [ + "max_output_tokens", + "stream", + "temperature", + "top_p", + "tools", + "reasoning", + "preset", + "instructions", + "models", # Model fallback support + ] + + def validate_environment( + self, headers: dict, model: str, litellm_params: Optional[GenericLiteLLMParams] + ) -> dict: + """Validate environment and set up headers""" + # Get API key from environment + api_key = ( + get_secret_str("PERPLEXITYAI_API_KEY") + or get_secret_str("PERPLEXITY_API_KEY") + ) + + if api_key: + headers["Authorization"] = f"Bearer {api_key}" + + headers["Content-Type"] = "application/json" + + return headers + + def get_complete_url( + self, + api_base: Optional[str], + litellm_params: dict, + ) -> str: + """Get the complete URL for the Perplexity Responses API""" + if api_base is None: + api_base = get_secret_str("PERPLEXITY_API_BASE") or "https://api.perplexity.ai" + + # Ensure api_base doesn't end with a slash + api_base = api_base.rstrip("/") + + # Add the responses endpoint + return f"{api_base}/v1/responses" + + def map_openai_params( + self, + response_api_optional_params: ResponsesAPIOptionalRequestParams, + model: str, + drop_params: bool, + ) -> Dict: + """ + Map OpenAI Responses API parameters to Perplexity format + + Key differences: + - Supports 'preset' parameter for predefined configurations + - Supports 'instructions' parameter for system-level guidance + - Tools are specified differently (web_search, fetch_url) + """ + mapped_params: Dict[str, Any] = {} + + # Map standard parameters + if response_api_optional_params.get("max_output_tokens"): + mapped_params["max_output_tokens"] = response_api_optional_params["max_output_tokens"] + + if response_api_optional_params.get("temperature"): + mapped_params["temperature"] = response_api_optional_params["temperature"] + + if response_api_optional_params.get("top_p"): + mapped_params["top_p"] = response_api_optional_params["top_p"] + + if response_api_optional_params.get("stream"): + mapped_params["stream"] = response_api_optional_params["stream"] + + if response_api_optional_params.get("stream_options"): + mapped_params["stream_options"] = response_api_optional_params["stream_options"] + + # Map Perplexity-specific parameters (using .get() with Any dict access) + preset = response_api_optional_params.get("preset") # type: ignore + if preset: + mapped_params["preset"] = preset + + instructions = response_api_optional_params.get("instructions") # type: ignore + if instructions: + mapped_params["instructions"] = instructions + + if response_api_optional_params.get("reasoning"): + mapped_params["reasoning"] = response_api_optional_params["reasoning"] + + tools = response_api_optional_params.get("tools") + if tools: + # Convert tools to list of dicts for transformation + tools_list = [dict(tool) if hasattr(tool, '__dict__') else tool for tool in tools] # type: ignore + mapped_params["tools"] = self._transform_tools(tools_list) # type: ignore + + return mapped_params + + def _transform_tools(self, tools: List[Dict[str, Any]]) -> List[Dict[str, Any]]: + """ + Transform tools to Perplexity format + + Perplexity supports: + - web_search: Performs web searches + - fetch_url: Fetches content from URLs + """ + perplexity_tools = [] + + for tool in tools: + if isinstance(tool, dict): + tool_type = tool.get("type") + + # Direct Perplexity tool format + if tool_type in ["web_search", "fetch_url"]: + perplexity_tools.append(tool) + + # OpenAI function format - try to map to Perplexity tools + elif tool_type == "function": + function = tool.get("function", {}) + function_name = function.get("name", "") + + if function_name == "web_search" or "search" in function_name.lower(): + perplexity_tools.append({"type": "web_search"}) + elif function_name == "fetch_url" or "fetch" in function_name.lower(): + perplexity_tools.append({"type": "fetch_url"}) + + return perplexity_tools + + def transform_responses_api_request( + self, + model: str, + input: Union[str, ResponseInputParam], + response_api_optional_request_params: Dict, + litellm_params: GenericLiteLLMParams, + headers: dict, + ) -> Dict: + """ + Transform request to Perplexity Responses API format + """ + # Check if the model is a preset (format: preset/preset-name) + if model.startswith("preset/"): + preset_name = model.replace("preset/", "") + data = { + "preset": preset_name, + "input": self._format_input(input), + } + # Check if preset is explicitly provided in params + elif response_api_optional_request_params.get("preset"): + data = { + "preset": response_api_optional_request_params.pop("preset"), + "input": self._format_input(input), + } + else: + # Full request format for third-party models + data = { + "model": model, + "input": self._format_input(input), + } + + # Add all optional parameters + for key, value in response_api_optional_request_params.items(): + data[key] = value + + return data + + def _format_input(self, input: Union[str, ResponseInputParam]) -> Union[str, List[Dict[str, Any]]]: + """ + Format input for Perplexity Responses API + + The API accepts either: + - A simple string for single-turn queries + - An array of message objects for multi-turn conversations + """ + if isinstance(input, str): + return input + + # Handle ResponseInputParam format + if isinstance(input, list): + formatted_messages = [] + for item in input: + if isinstance(item, dict): + formatted_message = { + "type": "message", + "role": item.get("role"), + "content": item.get("content", ""), + } + formatted_messages.append(formatted_message) + return formatted_messages + + return str(input) + + def transform_response_api_response( + self, + model: str, + raw_response: httpx.Response, + logging_obj: LiteLLMLoggingObj, + ) -> ResponsesAPIResponse: + """ + Transform Perplexity Responses API response to OpenAI Responses API format + """ + try: + raw_response_json = raw_response.json() + except Exception as e: + raise BaseLLMException( + status_code=raw_response.status_code, + message=f"Failed to parse response: {str(e)}", + ) + + # Check for error status + status = raw_response_json.get("status") + if status == "failed": + error = raw_response_json.get("error", {}) + error_message = error.get("message", "Unknown error") + raise BaseLLMException( + status_code=raw_response.status_code, + message=error_message, + ) + + # Transform usage to handle Perplexity's cost structure + usage_data = raw_response_json.get("usage", {}) + transformed_usage_dict = self._transform_usage(usage_data) + + # Convert usage dict to ResponseAPIUsage object + usage_obj = ResponseAPIUsage(**transformed_usage_dict) if transformed_usage_dict else None + + # Map Perplexity response to OpenAI Responses API format + response = ResponsesAPIResponse( + id=raw_response_json.get("id", ""), + object="response", + created_at=raw_response_json.get("created_at", 0), + status=raw_response_json.get("status", "completed"), + model=raw_response_json.get("model", model), + output=raw_response_json.get("output", []), + usage=usage_obj, + ) + + return response + + def _transform_usage(self, usage_data: Dict[str, Any]) -> Dict[str, Any]: + """ + Transform Perplexity usage data to OpenAI format + + Perplexity returns: + { + "input_tokens": 100, + "output_tokens": 200, + "total_tokens": 300, + "cost": { + "currency": "USD", + "input_cost": 0.0001, + "output_cost": 0.0002, + "total_cost": 0.0003 + } + } + + OpenAI expects: + { + "input_tokens": 100, + "output_tokens": 200, + "total_tokens": 300, + "cost": 0.0003 + } + """ + transformed = { + "input_tokens": usage_data.get("input_tokens", 0), + "output_tokens": usage_data.get("output_tokens", 0), + "total_tokens": usage_data.get("total_tokens", 0), + } + + # Transform cost from Perplexity format (dict) to OpenAI format (float) + cost_obj = usage_data.get("cost") + if isinstance(cost_obj, dict) and "total_cost" in cost_obj: + transformed["cost"] = cost_obj["total_cost"] + verbose_logger.debug( + "Transformed Perplexity cost object to float: %s -> %s", + cost_obj, + cost_obj["total_cost"] + ) + elif cost_obj is not None: + # If cost is already a float/number, use it as-is + transformed["cost"] = cost_obj + + # Add input_tokens_details if present + if "input_tokens_details" in usage_data: + transformed["input_tokens_details"] = usage_data["input_tokens_details"] + + # Add output_tokens_details if present + if "output_tokens_details" in usage_data: + transformed["output_tokens_details"] = usage_data["output_tokens_details"] + + return transformed + + def transform_streaming_response( + self, + model: str, + parsed_chunk: dict, + logging_obj: LiteLLMLoggingObj, + ) -> ResponsesAPIStreamingResponse: + """ + Transform a parsed streaming response chunk into a ResponsesAPIStreamingResponse + """ + # Get the event type from the chunk + verbose_logger.debug("Raw Perplexity Chunk=%s", parsed_chunk) + event_type = str(parsed_chunk.get("type")) + event_pydantic_model = PerplexityResponsesConfig.get_event_model_class( + event_type=event_type + ) + + # Transform Perplexity-specific fields to OpenAI format + parsed_chunk = self._transform_perplexity_chunk(parsed_chunk) + + # Defensive: Handle error.code being null (similar to OpenAI implementation) + try: + error_obj = parsed_chunk.get("error") + if isinstance(error_obj, dict) and error_obj.get("code") is None: + # Preserve other fields, but ensure `code` is a non-null string + parsed_chunk = dict(parsed_chunk) + parsed_chunk["error"] = dict(error_obj) + parsed_chunk["error"]["code"] = "unknown_error" + except Exception: + # If anything unexpected happens here, fall back to attempting + # instantiation and let higher-level handlers manage errors. + verbose_logger.debug("Failed to coalesce error.code in parsed_chunk") + + return event_pydantic_model(**parsed_chunk) + + def _transform_perplexity_chunk(self, chunk: dict) -> dict: + """ + Transform Perplexity-specific fields in a streaming chunk to OpenAI format. + + This handles: + - Converting Perplexity's cost object to a simple float + """ + # Make a copy to avoid modifying the original + chunk = dict(chunk) + + # Transform usage.cost from Perplexity format to OpenAI format + # Perplexity: {"currency": "USD", "input_cost": 0.0001, "output_cost": 0.0002, "total_cost": 0.0003} + # OpenAI: 0.0003 (just the total_cost as a float) + try: + response_obj = chunk.get("response") + if isinstance(response_obj, dict): + usage_obj = response_obj.get("usage") + if isinstance(usage_obj, dict): + cost_obj = usage_obj.get("cost") + if isinstance(cost_obj, dict) and "total_cost" in cost_obj: + # Replace the cost object with just the total_cost value + chunk = dict(chunk) + chunk["response"] = dict(response_obj) + chunk["response"]["usage"] = dict(usage_obj) + chunk["response"]["usage"]["cost"] = cost_obj["total_cost"] + verbose_logger.debug( + "Transformed Perplexity cost object to float: %s -> %s", + cost_obj, + cost_obj["total_cost"] + ) + except Exception as e: + # If transformation fails, log and continue with original chunk + verbose_logger.debug("Failed to transform Perplexity cost object: %s", e) + + return chunk diff --git a/litellm/llms/vertex_ai/common_utils.py b/litellm/llms/vertex_ai/common_utils.py index a0e2ddf5e98..02b69b94d94 100644 --- a/litellm/llms/vertex_ai/common_utils.py +++ b/litellm/llms/vertex_ai/common_utils.py @@ -1,4 +1,5 @@ import re +from copy import deepcopy from enum import Enum from typing import Any, Dict, List, Literal, Optional, Set, Tuple, Union, get_type_hints @@ -684,7 +685,7 @@ def convert_anyof_null_to_nullable(schema, depth=0): if anyof is not None: contains_null = False for atype in anyof: - if atype == {"type": "null"}: + if isinstance(atype, dict) and atype.get("type") == "null": # remove null type anyof.remove(atype) contains_null = True @@ -801,8 +802,38 @@ def _convert_schema_types(schema, depth=0): if "type" in schema: type_val = schema["type"] if isinstance(type_val, list) and len(type_val) > 1: - # Convert ["string", "number"] -> {"anyOf": [{"type": "STRING"}, {"type": "NUMBER"}]} - schema["anyOf"] = [{"type": t} for t in type_val if isinstance(t, str)] + # Convert type arrays to anyOf format + # Fields that are specific to object/array types and should move into anyOf + type_specific_fields = {"properties", "required", "additionalProperties", "items", "minItems", "maxItems", "minProperties", "maxProperties"} + + any_of: List[Dict[str, Any]] = [] + for t in type_val: + if not isinstance(t, str): + continue + if t == "null": + # Keep null entry minimal so we can strip it later. + any_of.append({"type": "null"}) + continue + + # For object/array types, include type-specific fields + if t in ("object", "array"): + item_schema = {"type": t} + # Move type-specific fields into this anyOf item + for field in type_specific_fields: + if field in schema: + item_schema[field] = deepcopy(schema[field]) + any_of.append(item_schema) + else: + # For primitive types, only include the type + any_of.append({"type": t}) + + # Remove type-specific fields from parent if we moved them into anyOf + has_object_or_array = any(t in ("object", "array") for t in type_val if isinstance(t, str)) + if has_object_or_array: + for field in type_specific_fields: + schema.pop(field, None) + + schema["anyOf"] = any_of schema.pop("type") elif isinstance(type_val, list) and len(type_val) == 1: schema["type"] = type_val[0] diff --git a/litellm/llms/vertex_ai/gemini/transformation.py b/litellm/llms/vertex_ai/gemini/transformation.py index 3004f39b973..00b461dcda0 100644 --- a/litellm/llms/vertex_ai/gemini/transformation.py +++ b/litellm/llms/vertex_ai/gemini/transformation.py @@ -437,6 +437,27 @@ def _gemini_convert_messages_with_history( # noqa: PLR0915 else: assistant_content.append(PartType(text=assistant_text)) # type: ignore + ## HANDLE ASSISTANT IMAGES FIELD + # Process images field if present (for generated images from assistant) + assistant_images = assistant_msg.get("images") + if assistant_images is not None and isinstance(assistant_images, list): + for image_item in assistant_images: + if isinstance(image_item, dict): + image_url_obj = image_item.get("image_url") + if isinstance(image_url_obj, dict): + assistant_image_url = image_url_obj.get("url") + format = image_url_obj.get("format") + detail = image_url_obj.get("detail") + media_resolution_enum = _convert_detail_to_media_resolution_enum(detail) + if assistant_image_url: + _part = _process_gemini_media( + image_url=assistant_image_url, + format=format, + media_resolution_enum=media_resolution_enum, + model=model, + ) + assistant_content.append(_part) + ## HANDLE ASSISTANT FUNCTION CALL if ( assistant_msg.get("tool_calls", []) is not None diff --git a/litellm/llms/vertex_ai/vertex_ai_partner_models/count_tokens/handler.py b/litellm/llms/vertex_ai/vertex_ai_partner_models/count_tokens/handler.py index 3842159fd7b..c6914ac3d6b 100644 --- a/litellm/llms/vertex_ai/vertex_ai_partner_models/count_tokens/handler.py +++ b/litellm/llms/vertex_ai/vertex_ai_partner_models/count_tokens/handler.py @@ -107,6 +107,11 @@ class VertexAIPartnerModelsTokenCounter(VertexBase): vertex_project = self.get_vertex_ai_project(litellm_params) vertex_location = self.get_vertex_ai_location(litellm_params) + # Map empty location/cluade models to a supported region for count-tokens endpoint + # https://docs.cloud.google.com/vertex-ai/generative-ai/docs/partner-models/claude/count-tokens + if not vertex_location or "claude" in model.lower(): + vertex_location = "us-central1" + # Get access token and resolved project ID access_token, project_id = await self._ensure_access_token_async( credentials=vertex_credentials, @@ -118,7 +123,7 @@ class VertexAIPartnerModelsTokenCounter(VertexBase): endpoint_url = self._build_count_tokens_endpoint( model=model, project_id=project_id, - vertex_location=vertex_location or "us-central1", + vertex_location=vertex_location, api_base=litellm_params.get("api_base"), ) diff --git a/litellm/model_prices_and_context_window_backup.json b/litellm/model_prices_and_context_window_backup.json index 815d29c7964..d794aa50d2e 100644 --- a/litellm/model_prices_and_context_window_backup.json +++ b/litellm/model_prices_and_context_window_backup.json @@ -5848,6 +5848,19 @@ "output_cost_per_token": 7e-07, "supports_tool_choice": true }, + "azure_ai/kimi-k2.5": { + "input_cost_per_token": 6e-07, + "litellm_provider": "azure_ai", + "max_input_tokens": 262144, + "max_output_tokens": 262144, + "max_tokens": 262144, + "mode": "chat", + "output_cost_per_token": 3e-06, + "source": "https://techcommunity.microsoft.com/blog/azure-ai-foundry-blog/kimi-k2-5-now-in-microsoft-foundry/4492321", + "supports_function_calling": true, + "supports_tool_choice": true, + "supports_vision": true + }, "azure_ai/ministral-3b": { "input_cost_per_token": 4e-08, "litellm_provider": "azure_ai", @@ -6091,6 +6104,28 @@ "output_cost_per_token": 2.4e-05, "supports_tool_choice": true }, + "bedrock/ap-northeast-1/moonshotai.kimi-k2-thinking": { + "input_cost_per_token": 7.3e-07, + "litellm_provider": "bedrock", + "max_input_tokens": 262144, + "max_output_tokens": 262144, + "max_tokens": 262144, + "mode": "chat", + "output_cost_per_token": 3.03e-06, + "supports_function_calling": true, + "supports_reasoning": true + }, + "bedrock/moonshotai.kimi-k2.5": { + "input_cost_per_token": 7.3e-07, + "litellm_provider": "bedrock", + "max_input_tokens": 262144, + "max_output_tokens": 262144, + "max_tokens": 262144, + "mode": "chat", + "output_cost_per_token": 3.03e-06, + "supports_function_calling": true, + "supports_reasoning": true + }, "bedrock/ap-south-1/meta.llama3-70b-instruct-v1:0": { "input_cost_per_token": 3.18e-06, "litellm_provider": "bedrock", @@ -6109,6 +6144,17 @@ "mode": "chat", "output_cost_per_token": 7.2e-07 }, + "bedrock/ap-south-1/moonshotai.kimi-k2-thinking": { + "input_cost_per_token": 7.1e-07, + "litellm_provider": "bedrock", + "max_input_tokens": 262144, + "max_output_tokens": 262144, + "max_tokens": 262144, + "mode": "chat", + "output_cost_per_token": 2.94e-06, + "supports_function_calling": true, + "supports_reasoning": true + }, "bedrock/ca-central-1/meta.llama3-70b-instruct-v1:0": { "input_cost_per_token": 3.05e-06, "litellm_provider": "bedrock", @@ -6314,6 +6360,17 @@ "mode": "chat", "output_cost_per_token": 1.01e-06 }, + "bedrock/sa-east-1/moonshotai.kimi-k2-thinking": { + "input_cost_per_token": 7.3e-07, + "litellm_provider": "bedrock", + "max_input_tokens": 262144, + "max_output_tokens": 262144, + "max_tokens": 262144, + "mode": "chat", + "output_cost_per_token": 3.03e-06, + "supports_function_calling": true, + "supports_reasoning": true + }, "bedrock/us-east-1/1-month-commitment/anthropic.claude-instant-v1": { "input_cost_per_second": 0.011, "litellm_provider": "bedrock", @@ -6450,6 +6507,28 @@ "output_cost_per_token": 7e-07, "supports_tool_choice": true }, + "bedrock/us-east-1/moonshotai.kimi-k2-thinking": { + "input_cost_per_token": 6e-07, + "litellm_provider": "bedrock", + "max_input_tokens": 262144, + "max_output_tokens": 262144, + "max_tokens": 262144, + "mode": "chat", + "output_cost_per_token": 2.5e-06, + "supports_function_calling": true, + "supports_reasoning": true + }, + "bedrock/us-east-2/moonshotai.kimi-k2-thinking": { + "input_cost_per_token": 6e-07, + "litellm_provider": "bedrock", + "max_input_tokens": 262144, + "max_output_tokens": 262144, + "max_tokens": 262144, + "mode": "chat", + "output_cost_per_token": 2.5e-06, + "supports_function_calling": true, + "supports_reasoning": true + }, "bedrock/us-gov-east-1/amazon.nova-pro-v1:0": { "input_cost_per_token": 9.6e-07, "litellm_provider": "bedrock", @@ -6856,6 +6935,17 @@ "output_cost_per_token": 7e-07, "supports_tool_choice": true }, + "bedrock/us-west-2/moonshotai.kimi-k2-thinking": { + "input_cost_per_token": 6e-07, + "litellm_provider": "bedrock", + "max_input_tokens": 262144, + "max_output_tokens": 262144, + "max_tokens": 262144, + "mode": "chat", + "output_cost_per_token": 2.5e-06, + "supports_function_calling": true, + "supports_reasoning": true + }, "bedrock/us.anthropic.claude-3-5-haiku-20241022-v1:0": { "cache_creation_input_token_cost": 1e-06, "cache_read_input_token_cost": 8e-08, @@ -25614,6 +25704,66 @@ "supports_function_calling": true, "supports_tool_choice": true }, + "perplexity/preset/pro-search": { + "litellm_provider": "perplexity", + "mode": "responses", + "supports_web_search": true, + "supports_preset": true + }, + "perplexity/openai/gpt-4o": { + "litellm_provider": "perplexity", + "mode": "responses", + "supports_web_search": true, + "supports_reasoning": false + }, + "perplexity/openai/gpt-4o-mini": { + "litellm_provider": "perplexity", + "mode": "responses", + "supports_web_search": true, + "supports_reasoning": false + }, + "perplexity/openai/gpt-5.2": { + "litellm_provider": "perplexity", + "mode": "responses", + "supports_web_search": true, + "supports_reasoning": true + }, + "perplexity/anthropic/claude-3-5-sonnet-20241022": { + "litellm_provider": "perplexity", + "mode": "responses", + "supports_web_search": true, + "supports_reasoning": false + }, + "perplexity/anthropic/claude-3-5-haiku-20241022": { + "litellm_provider": "perplexity", + "mode": "responses", + "supports_web_search": true, + "supports_reasoning": false + }, + "perplexity/google/gemini-2.0-flash-exp": { + "litellm_provider": "perplexity", + "mode": "responses", + "supports_web_search": true, + "supports_reasoning": false + }, + "perplexity/google/gemini-2.0-flash-thinking-exp": { + "litellm_provider": "perplexity", + "mode": "responses", + "supports_web_search": true, + "supports_reasoning": true + }, + "perplexity/xai/grok-2-1212": { + "litellm_provider": "perplexity", + "mode": "responses", + "supports_web_search": true, + "supports_reasoning": false + }, + "perplexity/xai/grok-2-vision-1212": { + "litellm_provider": "perplexity", + "mode": "responses", + "supports_web_search": true, + "supports_reasoning": false + }, "publicai/aisingapore/Qwen-SEA-LION-v4-32B-IT": { "input_cost_per_token": 0.0, "litellm_provider": "publicai", diff --git a/litellm/proxy/_experimental/mcp_server/discoverable_endpoints.py b/litellm/proxy/_experimental/mcp_server/discoverable_endpoints.py index 8b052dd0da1..bdf4cc312d9 100644 --- a/litellm/proxy/_experimental/mcp_server/discoverable_endpoints.py +++ b/litellm/proxy/_experimental/mcp_server/discoverable_endpoints.py @@ -16,6 +16,7 @@ from litellm.proxy.common_utils.encrypt_decrypt_utils import ( ) from litellm.proxy.common_utils.http_parsing_utils import _read_request_body from litellm.proxy.utils import get_server_root_path +from litellm.types.mcp import MCPAuth from litellm.types.mcp_server.mcp_server_manager import MCPServer router = APIRouter( @@ -125,6 +126,29 @@ def decode_state_hash(encrypted_state: str) -> dict: return state_data +def _resolve_oauth2_server_for_root_endpoints( + client_ip: Optional[str] = None, +) -> Optional[MCPServer]: + """ + Resolve the MCP server for root-level OAuth endpoints (no server name in path). + + When the MCP SDK hits root-level endpoints like /register, /authorize, /token + without a server name prefix, we try to find the right server automatically. + Returns the server if exactly one OAuth2 server is configured, else None. + """ + from litellm.proxy._experimental.mcp_server.mcp_server_manager import ( + global_mcp_server_manager, + ) + + registry = global_mcp_server_manager.get_filtered_registry(client_ip=client_ip) + oauth2_servers = [ + s for s in registry.values() if s.auth_type == MCPAuth.oauth2 + ] + if len(oauth2_servers) == 1: + return oauth2_servers[0] + return None + + async def authorize_with_server( request: Request, mcp_server: MCPServer, @@ -305,6 +329,8 @@ async def authorize( mcp_server = global_mcp_server_manager.get_mcp_server_by_name( lookup_name, client_ip=client_ip ) + if mcp_server is None and mcp_server_name is None: + mcp_server = _resolve_oauth2_server_for_root_endpoints() if mcp_server is None: raise HTTPException(status_code=404, detail="MCP server not found") return await authorize_with_server( @@ -350,6 +376,8 @@ async def token_endpoint( mcp_server = global_mcp_server_manager.get_mcp_server_by_name( lookup_name, client_ip=client_ip ) + if mcp_server is None and mcp_server_name is None: + mcp_server = _resolve_oauth2_server_for_root_endpoints() if mcp_server is None: raise HTTPException(status_code=404, detail="MCP server not found") return await exchange_token_with_server( @@ -430,6 +458,13 @@ def _build_oauth_protected_resource_response( ) request_base_url = get_request_base_url(request) + + # When no server name provided, try to resolve the single OAuth2 server + if mcp_server_name is None: + resolved = _resolve_oauth2_server_for_root_endpoints() + if resolved: + mcp_server_name = resolved.server_name or resolved.name + mcp_server: Optional[MCPServer] = None if mcp_server_name: client_ip = IPAddressUtils.get_mcp_client_ip(request) @@ -535,6 +570,12 @@ def _build_oauth_authorization_server_response( request_base_url = get_request_base_url(request) + # When no server name provided, try to resolve the single OAuth2 server + if mcp_server_name is None: + resolved = _resolve_oauth2_server_for_root_endpoints() + if resolved: + mcp_server_name = resolved.server_name or resolved.name + authorization_endpoint = ( f"{request_base_url}/{mcp_server_name}/authorize" if mcp_server_name @@ -640,6 +681,19 @@ async def register_client(request: Request, mcp_server_name: Optional[str] = Non "redirect_uris": [f"{request_base_url}/callback"], } if not mcp_server_name: + resolved = _resolve_oauth2_server_for_root_endpoints() + if resolved: + return await register_client_with_server( + request=request, + mcp_server=resolved, + client_name=data.get("client_name", ""), + grant_types=data.get("grant_types", []), + response_types=data.get("response_types", []), + token_endpoint_auth_method=data.get( + "token_endpoint_auth_method", "" + ), + fallback_client_id=resolved.server_name or resolved.name, + ) return dummy_return client_ip = IPAddressUtils.get_mcp_client_ip(request) diff --git a/litellm/proxy/_experimental/mcp_server/mcp_server_manager.py b/litellm/proxy/_experimental/mcp_server/mcp_server_manager.py index 532aea249bf..24eae430d85 100644 --- a/litellm/proxy/_experimental/mcp_server/mcp_server_manager.py +++ b/litellm/proxy/_experimental/mcp_server/mcp_server_manager.py @@ -341,7 +341,7 @@ class MCPServerManager: verbose_logger.info( f"Loading OpenAPI spec from {spec_path} for server {server_name}" ) - self._register_openapi_tools( + await self._register_openapi_tools( spec_path=spec_path, server=new_server, base_url=server_config.get("url", ""), @@ -353,7 +353,9 @@ class MCPServerManager: self.initialize_tool_name_to_mcp_server_name_mapping() - def _register_openapi_tools(self, spec_path: str, server: MCPServer, base_url: str): + async def _register_openapi_tools( + self, spec_path: str, server: MCPServer, base_url: str + ): """ Register tools from an OpenAPI specification for a given server. @@ -375,15 +377,15 @@ class MCPServerManager: get_base_url as get_openapi_base_url, ) from litellm.proxy._experimental.mcp_server.openapi_to_mcp_generator import ( - load_openapi_spec, + load_openapi_spec_async, ) from litellm.proxy._experimental.mcp_server.tool_registry import ( global_mcp_tool_registry, ) try: - # Load OpenAPI spec - spec = load_openapi_spec(spec_path) + # Load OpenAPI spec (async to avoid "called from within a running event loop") + spec = await load_openapi_spec_async(spec_path) # Use base_url from config if provided, otherwise extract from spec if not base_url: diff --git a/litellm/proxy/_experimental/mcp_server/openapi_to_mcp_generator.py b/litellm/proxy/_experimental/mcp_server/openapi_to_mcp_generator.py index b635f15ed09..deb0b4f9549 100644 --- a/litellm/proxy/_experimental/mcp_server/openapi_to_mcp_generator.py +++ b/litellm/proxy/_experimental/mcp_server/openapi_to_mcp_generator.py @@ -3,6 +3,8 @@ This module is used to generate MCP tools from OpenAPI specs. """ import json +import asyncio +import os from pathlib import PurePosixPath from typing import Any, Dict, Optional from urllib.parse import quote @@ -45,8 +47,36 @@ def _sanitize_path_parameter_value(param_value: Any, param_name: str) -> str: def load_openapi_spec(filepath: str) -> Dict[str, Any]: - """Load OpenAPI specification from JSON file.""" - with open(filepath, "r") as f: + """ + Sync wrapper. For URL specs, use the shared/custom MCP httpx client. + """ + try: + # If we're already inside an event loop, prefer the async function. + asyncio.get_running_loop() + raise RuntimeError( + "load_openapi_spec() was called from within a running event loop. " + "Use 'await load_openapi_spec_async(...)' instead." + ) + except RuntimeError as e: + # "no running event loop" is fine; other RuntimeErrors we re-raise + if "no running event loop" not in str(e).lower(): + raise + return asyncio.run(load_openapi_spec_async(filepath)) + +async def load_openapi_spec_async(filepath: str) -> Dict[str, Any]: + if filepath.startswith("http://") or filepath.startswith("https://"): + client = get_async_httpx_client(llm_provider=httpxSpecialProvider.MCP) + # NOTE: do not close shared client if get_async_httpx_client returns a shared singleton. + # If it returns a new client each time, consider wrapping it in an async context manager. + r = await client.get(filepath) + r.raise_for_status() + return r.json() + + # fallback: local file + # Local filesystem path + if not os.path.exists(filepath): + raise FileNotFoundError(f"OpenAPI spec not found at {filepath}") + with open(filepath, "r", encoding="utf-8") as f: return json.load(f) diff --git a/litellm/proxy/_experimental/mcp_server/rest_endpoints.py b/litellm/proxy/_experimental/mcp_server/rest_endpoints.py index 65d3a2caa16..43b388993d9 100644 --- a/litellm/proxy/_experimental/mcp_server/rest_endpoints.py +++ b/litellm/proxy/_experimental/mcp_server/rest_endpoints.py @@ -1,6 +1,6 @@ import importlib from datetime import datetime -from typing import Dict, List, Optional, Union +from typing import Any, Awaitable, Callable, Dict, List, Optional, Union from fastapi import APIRouter, Depends, HTTPException, Query, Request @@ -501,24 +501,50 @@ if MCP_AVAILABLE: NewMCPServerRequest, ) + def _extract_credentials( + request: NewMCPServerRequest, + ) -> tuple: + """ + Extract OAuth credentials from the nested ``request.credentials`` dict. + + Returns: + (client_id, client_secret, scopes) — any value may be ``None``. + """ + creds = request.credentials if isinstance(request.credentials, dict) else {} + client_id: Optional[str] = creds.get("client_id") + client_secret: Optional[str] = creds.get("client_secret") + scopes_raw = creds.get("scopes") + scopes: Optional[List[str]] = scopes_raw if isinstance(scopes_raw, list) else None + return client_id, client_secret, scopes + async def _execute_with_mcp_client( request: NewMCPServerRequest, - operation, + operation: Callable[..., Awaitable[Any]], mcp_auth_header: Optional[Union[str, Dict[str, str]]] = None, oauth2_headers: Optional[Dict[str, str]] = None, raw_headers: Optional[Dict[str, str]] = None, - ): + ) -> dict: """ - Common helper to create MCP client, execute operation, and ensure proper cleanup. + Create a temporary MCP client from *request*, run *operation*, and return the result. + + For M2M OAuth servers (those with ``client_id``, ``client_secret``, and + ``token_url``), the incoming ``oauth2_headers`` are dropped so that + ``resolve_mcp_auth`` can auto-fetch a token via ``client_credentials``. Args: - request: MCP server configuration - operation: Async function that takes a client and returns the operation result + request: MCP server configuration submitted by the UI. + operation: Async callable that receives the created client and returns a result dict. + mcp_auth_header: Pre-resolved credential header (API-key / bearer token). + oauth2_headers: Headers extracted from the incoming request (may contain the + litellm API key — must NOT be forwarded for M2M servers). + raw_headers: Raw request headers forwarded for stdio env construction. Returns: - Operation result or error response + The dict returned by *operation*, or an error dict on failure. """ try: + client_id, client_secret, scopes = _extract_credentials(request) + server_model = MCPServer( server_id=request.server_id or "", name=request.alias or request.server_name or "", @@ -530,14 +556,26 @@ if MCP_AVAILABLE: args=request.args, env=request.env, static_headers=request.static_headers, + client_id=client_id, + client_secret=client_secret, + token_url=request.token_url, + scopes=scopes, + authorization_url=request.authorization_url, + registration_url=request.registration_url, ) stdio_env = global_mcp_server_manager._build_stdio_env( server_model, raw_headers ) + # For M2M OAuth servers, drop the incoming Authorization header so that + # resolve_mcp_auth can auto-fetch a token via client_credentials. + effective_oauth2_headers = ( + None if server_model.has_client_credentials else oauth2_headers + ) + merged_headers = merge_mcp_headers( - extra_headers=oauth2_headers, + extra_headers=effective_oauth2_headers, static_headers=request.static_headers, ) @@ -550,11 +588,14 @@ if MCP_AVAILABLE: return await operation(client) - except Exception as e: - verbose_logger.error(f"Error in MCP operation: {e}", exc_info=True) + except (KeyboardInterrupt, SystemExit): + raise + except BaseException as e: + verbose_logger.error("Error in MCP operation: %s", e, exc_info=True) return { "status": "error", - "message": "An internal error has occurred while testing the MCP server.", + "error": True, + "message": "Failed to connect to MCP server. Check proxy logs for details.", } @router.post("/test/connection", dependencies=[Depends(user_api_key_auth)]) diff --git a/litellm/proxy/_experimental/mcp_server/server.py b/litellm/proxy/_experimental/mcp_server/server.py index 890c4ae8fb2..58cd8c99e7b 100644 --- a/litellm/proxy/_experimental/mcp_server/server.py +++ b/litellm/proxy/_experimental/mcp_server/server.py @@ -31,6 +31,9 @@ from litellm.litellm_core_utils.litellm_logging import Logging as LiteLLMLogging from litellm.proxy._experimental.mcp_server.auth.user_api_key_auth_mcp import ( MCPRequestHandler, ) +from litellm.proxy._experimental.mcp_server.discoverable_endpoints import ( + get_request_base_url, +) from litellm.proxy._experimental.mcp_server.utils import ( LITELLM_MCP_SERVER_DESCRIPTION, LITELLM_MCP_SERVER_NAME, @@ -1972,7 +1975,7 @@ if MCP_AVAILABLE: ) if server and server.auth_type == MCPAuth.oauth2 and not oauth2_headers: request = StarletteRequest(scope) - base_url = str(request.base_url).rstrip("/") + base_url = get_request_base_url(request) authorization_uri = ( f"Bearer authorization_uri=" diff --git a/litellm/proxy/_experimental/out/404.html b/litellm/proxy/_experimental/out/404/index.html similarity index 100% rename from litellm/proxy/_experimental/out/404.html rename to litellm/proxy/_experimental/out/404/index.html diff --git a/litellm/proxy/_experimental/out/_not-found.html b/litellm/proxy/_experimental/out/_not-found/index.html similarity index 100% rename from litellm/proxy/_experimental/out/_not-found.html rename to litellm/proxy/_experimental/out/_not-found/index.html diff --git a/litellm/proxy/_experimental/out/api-reference.html b/litellm/proxy/_experimental/out/api-reference/index.html similarity index 100% rename from litellm/proxy/_experimental/out/api-reference.html rename to litellm/proxy/_experimental/out/api-reference/index.html diff --git a/litellm/proxy/_experimental/out/experimental/api-playground.html b/litellm/proxy/_experimental/out/experimental/api-playground/index.html similarity index 100% rename from litellm/proxy/_experimental/out/experimental/api-playground.html rename to litellm/proxy/_experimental/out/experimental/api-playground/index.html diff --git a/litellm/proxy/_experimental/out/experimental/budgets.html b/litellm/proxy/_experimental/out/experimental/budgets/index.html similarity index 100% rename from litellm/proxy/_experimental/out/experimental/budgets.html rename to litellm/proxy/_experimental/out/experimental/budgets/index.html diff --git a/litellm/proxy/_experimental/out/experimental/caching.html b/litellm/proxy/_experimental/out/experimental/caching/index.html similarity index 100% rename from litellm/proxy/_experimental/out/experimental/caching.html rename to litellm/proxy/_experimental/out/experimental/caching/index.html diff --git a/litellm/proxy/_experimental/out/experimental/claude-code-plugins.html b/litellm/proxy/_experimental/out/experimental/claude-code-plugins/index.html similarity index 100% rename from litellm/proxy/_experimental/out/experimental/claude-code-plugins.html rename to litellm/proxy/_experimental/out/experimental/claude-code-plugins/index.html diff --git a/litellm/proxy/_experimental/out/experimental/old-usage.html b/litellm/proxy/_experimental/out/experimental/old-usage/index.html similarity index 100% rename from litellm/proxy/_experimental/out/experimental/old-usage.html rename to litellm/proxy/_experimental/out/experimental/old-usage/index.html diff --git a/litellm/proxy/_experimental/out/experimental/prompts.html b/litellm/proxy/_experimental/out/experimental/prompts/index.html similarity index 100% rename from litellm/proxy/_experimental/out/experimental/prompts.html rename to litellm/proxy/_experimental/out/experimental/prompts/index.html diff --git a/litellm/proxy/_experimental/out/experimental/tag-management.html b/litellm/proxy/_experimental/out/experimental/tag-management/index.html similarity index 100% rename from litellm/proxy/_experimental/out/experimental/tag-management.html rename to litellm/proxy/_experimental/out/experimental/tag-management/index.html diff --git a/litellm/proxy/_experimental/out/guardrails.html b/litellm/proxy/_experimental/out/guardrails/index.html similarity index 100% rename from litellm/proxy/_experimental/out/guardrails.html rename to litellm/proxy/_experimental/out/guardrails/index.html diff --git a/litellm/proxy/_experimental/out/login.html b/litellm/proxy/_experimental/out/login/index.html similarity index 100% rename from litellm/proxy/_experimental/out/login.html rename to litellm/proxy/_experimental/out/login/index.html diff --git a/litellm/proxy/_experimental/out/logs.html b/litellm/proxy/_experimental/out/logs/index.html similarity index 100% rename from litellm/proxy/_experimental/out/logs.html rename to litellm/proxy/_experimental/out/logs/index.html diff --git a/litellm/proxy/_experimental/out/mcp/oauth/callback.html b/litellm/proxy/_experimental/out/mcp/oauth/callback/index.html similarity index 100% rename from litellm/proxy/_experimental/out/mcp/oauth/callback.html rename to litellm/proxy/_experimental/out/mcp/oauth/callback/index.html diff --git a/litellm/proxy/_experimental/out/model-hub.html b/litellm/proxy/_experimental/out/model-hub/index.html similarity index 100% rename from litellm/proxy/_experimental/out/model-hub.html rename to litellm/proxy/_experimental/out/model-hub/index.html diff --git a/litellm/proxy/_experimental/out/model_hub.html b/litellm/proxy/_experimental/out/model_hub/index.html similarity index 100% rename from litellm/proxy/_experimental/out/model_hub.html rename to litellm/proxy/_experimental/out/model_hub/index.html diff --git a/litellm/proxy/_experimental/out/model_hub_table.html b/litellm/proxy/_experimental/out/model_hub_table/index.html similarity index 100% rename from litellm/proxy/_experimental/out/model_hub_table.html rename to litellm/proxy/_experimental/out/model_hub_table/index.html diff --git a/litellm/proxy/_experimental/out/models-and-endpoints.html b/litellm/proxy/_experimental/out/models-and-endpoints/index.html similarity index 100% rename from litellm/proxy/_experimental/out/models-and-endpoints.html rename to litellm/proxy/_experimental/out/models-and-endpoints/index.html diff --git a/litellm/proxy/_experimental/out/onboarding.html b/litellm/proxy/_experimental/out/onboarding/index.html similarity index 100% rename from litellm/proxy/_experimental/out/onboarding.html rename to litellm/proxy/_experimental/out/onboarding/index.html diff --git a/litellm/proxy/_experimental/out/organizations.html b/litellm/proxy/_experimental/out/organizations/index.html similarity index 100% rename from litellm/proxy/_experimental/out/organizations.html rename to litellm/proxy/_experimental/out/organizations/index.html diff --git a/litellm/proxy/_experimental/out/playground.html b/litellm/proxy/_experimental/out/playground/index.html similarity index 100% rename from litellm/proxy/_experimental/out/playground.html rename to litellm/proxy/_experimental/out/playground/index.html diff --git a/litellm/proxy/_experimental/out/policies.html b/litellm/proxy/_experimental/out/policies/index.html similarity index 100% rename from litellm/proxy/_experimental/out/policies.html rename to litellm/proxy/_experimental/out/policies/index.html diff --git a/litellm/proxy/_experimental/out/settings/admin-settings.html b/litellm/proxy/_experimental/out/settings/admin-settings/index.html similarity index 100% rename from litellm/proxy/_experimental/out/settings/admin-settings.html rename to litellm/proxy/_experimental/out/settings/admin-settings/index.html diff --git a/litellm/proxy/_experimental/out/settings/logging-and-alerts.html b/litellm/proxy/_experimental/out/settings/logging-and-alerts/index.html similarity index 100% rename from litellm/proxy/_experimental/out/settings/logging-and-alerts.html rename to litellm/proxy/_experimental/out/settings/logging-and-alerts/index.html diff --git a/litellm/proxy/_experimental/out/settings/router-settings.html b/litellm/proxy/_experimental/out/settings/router-settings/index.html similarity index 100% rename from litellm/proxy/_experimental/out/settings/router-settings.html rename to litellm/proxy/_experimental/out/settings/router-settings/index.html diff --git a/litellm/proxy/_experimental/out/settings/ui-theme.html b/litellm/proxy/_experimental/out/settings/ui-theme/index.html similarity index 100% rename from litellm/proxy/_experimental/out/settings/ui-theme.html rename to litellm/proxy/_experimental/out/settings/ui-theme/index.html diff --git a/litellm/proxy/_experimental/out/teams.html b/litellm/proxy/_experimental/out/teams/index.html similarity index 100% rename from litellm/proxy/_experimental/out/teams.html rename to litellm/proxy/_experimental/out/teams/index.html diff --git a/litellm/proxy/_experimental/out/test-key.html b/litellm/proxy/_experimental/out/test-key/index.html similarity index 100% rename from litellm/proxy/_experimental/out/test-key.html rename to litellm/proxy/_experimental/out/test-key/index.html diff --git a/litellm/proxy/_experimental/out/tools/mcp-servers.html b/litellm/proxy/_experimental/out/tools/mcp-servers/index.html similarity index 100% rename from litellm/proxy/_experimental/out/tools/mcp-servers.html rename to litellm/proxy/_experimental/out/tools/mcp-servers/index.html diff --git a/litellm/proxy/_experimental/out/tools/vector-stores.html b/litellm/proxy/_experimental/out/tools/vector-stores/index.html similarity index 100% rename from litellm/proxy/_experimental/out/tools/vector-stores.html rename to litellm/proxy/_experimental/out/tools/vector-stores/index.html diff --git a/litellm/proxy/_experimental/out/usage.html b/litellm/proxy/_experimental/out/usage/index.html similarity index 100% rename from litellm/proxy/_experimental/out/usage.html rename to litellm/proxy/_experimental/out/usage/index.html diff --git a/litellm/proxy/_experimental/out/users.html b/litellm/proxy/_experimental/out/users/index.html similarity index 100% rename from litellm/proxy/_experimental/out/users.html rename to litellm/proxy/_experimental/out/users/index.html diff --git a/litellm/proxy/_experimental/out/virtual-keys.html b/litellm/proxy/_experimental/out/virtual-keys/index.html similarity index 100% rename from litellm/proxy/_experimental/out/virtual-keys.html rename to litellm/proxy/_experimental/out/virtual-keys/index.html diff --git a/litellm/proxy/_types.py b/litellm/proxy/_types.py index ecefe772247..06bedfc6c09 100644 --- a/litellm/proxy/_types.py +++ b/litellm/proxy/_types.py @@ -420,6 +420,8 @@ class LiteLLMRoutes(enum.Enum): "/mcp/tools", "/mcp/tools/list", "/mcp/tools/call", + # Read-only MCP discovery endpoint (virtual keys may be allowed here) + "/v1/mcp/server", ] agent_routes = [ @@ -845,9 +847,9 @@ class GenerateRequestBase(LiteLLMPydanticObjectBase): allowed_cache_controls: Optional[list] = [] config: Optional[dict] = {} permissions: Optional[dict] = {} - model_max_budget: Optional[ - dict - ] = {} # {"gpt-4": 5.0, "gpt-3.5-turbo": 5.0}, defaults to {} + model_max_budget: Optional[dict] = ( + {} + ) # {"gpt-4": 5.0, "gpt-3.5-turbo": 5.0}, defaults to {} model_config = ConfigDict(protected_namespaces=()) model_rpm_limit: Optional[dict] = None @@ -1396,12 +1398,12 @@ class NewCustomerRequest(BudgetNewRequest): blocked: bool = False # allow/disallow requests for this end-user budget_id: Optional[str] = None # give either a budget_id or max_budget spend: Optional[float] = None - allowed_model_region: Optional[ - AllowedModelRegion - ] = None # require all user requests to use models in this specific region - default_model: Optional[ - str - ] = None # if no equivalent model in allowed region - default all requests to this model + allowed_model_region: Optional[AllowedModelRegion] = ( + None # require all user requests to use models in this specific region + ) + default_model: Optional[str] = ( + None # if no equivalent model in allowed region - default all requests to this model + ) @model_validator(mode="before") @classmethod @@ -1423,12 +1425,12 @@ class UpdateCustomerRequest(LiteLLMPydanticObjectBase): blocked: bool = False # allow/disallow requests for this end-user max_budget: Optional[float] = None budget_id: Optional[str] = None # give either a budget_id or max_budget - allowed_model_region: Optional[ - AllowedModelRegion - ] = None # require all user requests to use models in this specific region - default_model: Optional[ - str - ] = None # if no equivalent model in allowed region - default all requests to this model + allowed_model_region: Optional[AllowedModelRegion] = ( + None # require all user requests to use models in this specific region + ) + default_model: Optional[str] = ( + None # if no equivalent model in allowed region - default all requests to this model + ) class DeleteCustomerRequest(LiteLLMPydanticObjectBase): @@ -1516,15 +1518,15 @@ class NewTeamRequest(TeamBase): ] = None # raise an error if 'guaranteed_throughput' is set and we're overallocating tpm model_tpm_limit: Optional[Dict[str, int]] = None - team_member_budget: Optional[ - float - ] = None # allow user to set a budget for all team members - team_member_rpm_limit: Optional[ - int - ] = None # allow user to set RPM limit for all team members - team_member_tpm_limit: Optional[ - int - ] = None # allow user to set TPM limit for all team members + team_member_budget: Optional[float] = ( + None # allow user to set a budget for all team members + ) + team_member_rpm_limit: Optional[int] = ( + None # allow user to set RPM limit for all team members + ) + team_member_tpm_limit: Optional[int] = ( + None # allow user to set TPM limit for all team members + ) team_member_key_duration: Optional[str] = None # e.g. "1d", "1w", "1m" allowed_vector_store_indexes: Optional[List[AllowedVectorStoreIndexItem]] = None @@ -1615,9 +1617,9 @@ class BlockKeyRequest(LiteLLMPydanticObjectBase): class AddTeamCallback(LiteLLMPydanticObjectBase): callback_name: str - callback_type: Optional[ - Literal["success", "failure", "success_and_failure"] - ] = "success_and_failure" + callback_type: Optional[Literal["success", "failure", "success_and_failure"]] = ( + "success_and_failure" + ) callback_vars: Dict[str, str] @model_validator(mode="before") @@ -1908,6 +1910,10 @@ class PassThroughGenericEndpoint(LiteLLMPydanticObjectBase): default=None, description="Guardrails configuration for this passthrough endpoint. Dict keys are guardrail names, values are optional settings for field targeting. When set, all org/team/key level guardrails will also execute. Defaults to None (no guardrails execute).", ) + is_from_config: bool = Field( + default=False, + description="True if this endpoint is defined in the config file, False if from DB. Config-defined endpoints cannot be edited via the UI.", + ) class PassThroughEndpointResponse(LiteLLMPydanticObjectBase): @@ -1945,9 +1951,9 @@ class ConfigList(LiteLLMPydanticObjectBase): stored_in_db: Optional[bool] field_default_value: Any premium_field: bool = False - nested_fields: Optional[ - List[FieldDetail] - ] = None # For nested dictionary or Pydantic fields + nested_fields: Optional[List[FieldDetail]] = ( + None # For nested dictionary or Pydantic fields + ) class UserHeaderMapping(LiteLLMPydanticObjectBase): @@ -2386,9 +2392,9 @@ class LiteLLM_OrganizationMembershipTable(LiteLLMPydanticObjectBase): budget_id: Optional[str] = None created_at: datetime updated_at: datetime - user: Optional[ - Any - ] = None # You might want to replace 'Any' with a more specific type if available + user: Optional[Any] = ( + None # You might want to replace 'Any' with a more specific type if available + ) litellm_budget_table: Optional[LiteLLM_BudgetTable] = None model_config = ConfigDict(protected_namespaces=()) @@ -3364,9 +3370,9 @@ class TeamModelDeleteRequest(BaseModel): # Organization Member Requests class OrganizationMemberAddRequest(OrgMemberAddRequest): organization_id: str - max_budget_in_organization: Optional[ - float - ] = None # Users max budget within the organization + max_budget_in_organization: Optional[float] = ( + None # Users max budget within the organization + ) class OrganizationMemberDeleteRequest(MemberDeleteRequest): @@ -3584,9 +3590,9 @@ class ProviderBudgetResponse(LiteLLMPydanticObjectBase): Maps provider names to their budget configs. """ - providers: Dict[ - str, ProviderBudgetResponseObject - ] = {} # Dictionary mapping provider names to their budget configurations + providers: Dict[str, ProviderBudgetResponseObject] = ( + {} + ) # Dictionary mapping provider names to their budget configurations class ProxyStateVariables(TypedDict): @@ -3729,9 +3735,9 @@ class LiteLLM_JWTAuth(LiteLLMPydanticObjectBase): enforce_rbac: bool = False roles_jwt_field: Optional[str] = None # v2 on role mappings role_mappings: Optional[List[RoleMapping]] = None - object_id_jwt_field: Optional[ - str - ] = None # can be either user / team, inferred from the role mapping + object_id_jwt_field: Optional[str] = ( + None # can be either user / team, inferred from the role mapping + ) scope_mappings: Optional[List[ScopeMapping]] = None enforce_scope_based_access: bool = False enforce_team_based_model_access: bool = False diff --git a/litellm/proxy/auth/user_api_key_auth.py b/litellm/proxy/auth/user_api_key_auth.py index 05eeab3f611..42f10ff8598 100644 --- a/litellm/proxy/auth/user_api_key_auth.py +++ b/litellm/proxy/auth/user_api_key_auth.py @@ -8,6 +8,7 @@ Returns a UserAPIKeyAuth object if the API key is valid """ import asyncio +import re import secrets from datetime import datetime, timezone from typing import List, Optional, Tuple, cast @@ -115,6 +116,18 @@ def _get_bearer_token_or_received_api_key(api_key: str) -> str: api_key = api_key.replace("Basic ", "") # handle langfuse input elif api_key.startswith("bearer "): api_key = api_key.replace("bearer ", "") + elif api_key.startswith("AWS4-HMAC-SHA256"): + # Handle AWS Signature V4 format from LangChain + # Format: AWS4-HMAC-SHA256 Credential=Bearer sk-12345/date/region/service/aws4_request, SignedHeaders=..., Signature=... + # Extract the Bearer token from the Credential field + match = re.search(r'Credential=Bearer\s+([^/\s,]+)', api_key) + if match: + api_key = match.group(1) + else: + # If no Bearer token found in Credential, try to extract just the credential value + match = re.search(r'Credential=([^/\s,]+)', api_key) + if match: + api_key = match.group(1) return api_key @@ -128,6 +141,20 @@ def _get_bearer_token( api_key = api_key.replace("Basic ", "") # handle langfuse input elif api_key.startswith("bearer "): api_key = api_key.replace("bearer ", "") + elif api_key.startswith("AWS4-HMAC-SHA256"): + # Handle AWS Signature V4 format from LangChain + # Format: AWS4-HMAC-SHA256 Credential=Bearer sk-12345/date/region/service/aws4_request, SignedHeaders=..., Signature=... + # Extract the Bearer token from the Credential field + match = re.search(r'Credential=Bearer\s+([^/\s,]+)', api_key) + if match: + api_key = match.group(1) + else: + # If no Bearer token found in Credential, try to extract just the credential value + match = re.search(r'Credential=([^/\s,]+)', api_key) + if match: + api_key = match.group(1) + else: + api_key = "" else: api_key = "" return api_key diff --git a/litellm/proxy/custom_sso.py b/litellm/proxy/custom_sso.py index 210e9eea3dc..b2b028dfbe3 100644 --- a/litellm/proxy/custom_sso.py +++ b/litellm/proxy/custom_sso.py @@ -24,9 +24,12 @@ async def custom_sso_handler(userIDPInfo: OpenID) -> SSOUserDefinedValues: print(f"userIDPInfo: {userIDPInfo}") # noqa if userIDPInfo.id is None: - raise ValueError( - f"No ID found for user. userIDPInfo.id is None {userIDPInfo}" - ) + raise ValueError(f"No ID found for user. userIDPInfo.id is None {userIDPInfo}") + + # Access extra fields from the IDP response (requires GENERIC_USER_EXTRA_ATTRIBUTES env var) + # Example: Set GENERIC_USER_EXTRA_ATTRIBUTES="group,NTID,domain" to capture these fields + # extra_fields = getattr(userIDPInfo, 'extra_fields', None) or {} + # user_groups = extra_fields.get("group", []) # check if user exists in litellm proxy DB _user_info = await user_info(user_id=userIDPInfo.id) diff --git a/litellm/proxy/db/db_transaction_queue/base_update_queue.py b/litellm/proxy/db/db_transaction_queue/base_update_queue.py index 202829b78b6..a5ec1c3eaf4 100644 --- a/litellm/proxy/db/db_transaction_queue/base_update_queue.py +++ b/litellm/proxy/db/db_transaction_queue/base_update_queue.py @@ -10,14 +10,18 @@ from litellm._service_logger import ServiceLogging service_logger_obj = ( ServiceLogging() ) # used for tracking metrics for In memory buffer, redis buffer, pod lock manager -from litellm.constants import MAX_IN_MEMORY_QUEUE_FLUSH_COUNT, MAX_SIZE_IN_MEMORY_QUEUE +from litellm.constants import ( + LITELLM_ASYNCIO_QUEUE_MAXSIZE, + MAX_IN_MEMORY_QUEUE_FLUSH_COUNT, + MAX_SIZE_IN_MEMORY_QUEUE, +) class BaseUpdateQueue: """Base class for in memory buffer for database transactions""" def __init__(self): - self.update_queue = asyncio.Queue() + self.update_queue = asyncio.Queue(maxsize=LITELLM_ASYNCIO_QUEUE_MAXSIZE) self.MAX_SIZE_IN_MEMORY_QUEUE = MAX_SIZE_IN_MEMORY_QUEUE async def add_update(self, update): diff --git a/litellm/proxy/db/db_transaction_queue/daily_spend_update_queue.py b/litellm/proxy/db/db_transaction_queue/daily_spend_update_queue.py index c3074e641b2..5ba8fb13596 100644 --- a/litellm/proxy/db/db_transaction_queue/daily_spend_update_queue.py +++ b/litellm/proxy/db/db_transaction_queue/daily_spend_update_queue.py @@ -3,6 +3,7 @@ from copy import deepcopy from typing import Dict, List, Optional from litellm._logging import verbose_proxy_logger +from litellm.constants import LITELLM_ASYNCIO_QUEUE_MAXSIZE from litellm.proxy._types import BaseDailySpendTransaction from litellm.proxy.db.db_transaction_queue.base_update_queue import ( BaseUpdateQueue, @@ -54,7 +55,7 @@ class DailySpendUpdateQueue(BaseUpdateQueue): def __init__(self): super().__init__() self.update_queue: asyncio.Queue[Dict[str, BaseDailySpendTransaction]] = ( - asyncio.Queue() + asyncio.Queue(maxsize=LITELLM_ASYNCIO_QUEUE_MAXSIZE) ) async def add_update(self, update: Dict[str, BaseDailySpendTransaction]): diff --git a/litellm/proxy/db/db_transaction_queue/spend_update_queue.py b/litellm/proxy/db/db_transaction_queue/spend_update_queue.py index 9b0449bb9ab..c96564252d0 100644 --- a/litellm/proxy/db/db_transaction_queue/spend_update_queue.py +++ b/litellm/proxy/db/db_transaction_queue/spend_update_queue.py @@ -2,6 +2,7 @@ import asyncio from typing import Dict, List, Optional from litellm._logging import verbose_proxy_logger +from litellm.constants import LITELLM_ASYNCIO_QUEUE_MAXSIZE from litellm.proxy._types import ( DBSpendUpdateTransactions, Litellm_EntityType, @@ -21,7 +22,9 @@ class SpendUpdateQueue(BaseUpdateQueue): def __init__(self): super().__init__() - self.update_queue: asyncio.Queue[SpendUpdateQueueItem] = asyncio.Queue() + self.update_queue: asyncio.Queue[SpendUpdateQueueItem] = asyncio.Queue( + maxsize=LITELLM_ASYNCIO_QUEUE_MAXSIZE + ) async def flush_and_get_aggregated_db_spend_update_transactions( self, diff --git a/litellm/proxy/guardrails/guardrail_endpoints.py b/litellm/proxy/guardrails/guardrail_endpoints.py index a825ce22b25..07702252a10 100644 --- a/litellm/proxy/guardrails/guardrail_endpoints.py +++ b/litellm/proxy/guardrails/guardrail_endpoints.py @@ -15,6 +15,7 @@ from litellm.proxy._types import UserAPIKeyAuth from litellm.proxy.auth.user_api_key_auth import user_api_key_auth from litellm.proxy.guardrails.guardrail_registry import GuardrailRegistry from litellm.types.guardrails import ( + BaseLitellmParams, PII_ENTITY_CATEGORIES_MAP, ApplyGuardrailRequest, ApplyGuardrailResponse, @@ -150,6 +151,7 @@ async def list_guardrails_v2(): } ``` """ + from litellm.litellm_core_utils.litellm_logging import _get_masked_values from litellm.proxy.guardrails.guardrail_registry import IN_MEMORY_GUARDRAIL_HANDLER from litellm.proxy.proxy_server import prisma_client @@ -164,11 +166,29 @@ async def list_guardrails_v2(): guardrail_configs: List[GuardrailInfoResponse] = [] seen_guardrail_ids = set() for guardrail in guardrails: + litellm_params: Optional[Union[LitellmParams, dict]] = guardrail.get( + "litellm_params" + ) + litellm_params_dict = ( + litellm_params.model_dump(exclude_none=True) + if isinstance(litellm_params, LitellmParams) + else litellm_params + ) or {} + masked_litellm_params_dict = _get_masked_values( + litellm_params_dict, + unmasked_length=4, + number_of_asterisks=4, + ) + masked_litellm_params = ( + BaseLitellmParams(**masked_litellm_params_dict) + if masked_litellm_params_dict + else None + ) guardrail_configs.append( GuardrailInfoResponse( guardrail_id=guardrail.get("guardrail_id"), guardrail_name=guardrail.get("guardrail_name"), - litellm_params=guardrail.get("litellm_params"), + litellm_params=masked_litellm_params, guardrail_info=guardrail.get("guardrail_info"), created_at=guardrail.get("created_at"), updated_at=guardrail.get("updated_at"), @@ -182,11 +202,27 @@ async def list_guardrails_v2(): for guardrail in in_memory_guardrails: # only add guardrails that are not in DB guardrail list already if guardrail.get("guardrail_id") not in seen_guardrail_ids: + in_memory_litellm_params_raw = guardrail.get("litellm_params") + in_memory_litellm_params_dict = ( + in_memory_litellm_params_raw.model_dump(exclude_none=True) + if isinstance(in_memory_litellm_params_raw, LitellmParams) + else in_memory_litellm_params_raw + ) or {} + masked_in_memory_litellm_params = _get_masked_values( + in_memory_litellm_params_dict, + unmasked_length=4, + number_of_asterisks=4, + ) + masked_in_memory_litellm_params_typed = ( + BaseLitellmParams(**masked_in_memory_litellm_params) + if masked_in_memory_litellm_params + else None + ) guardrail_configs.append( GuardrailInfoResponse( guardrail_id=guardrail.get("guardrail_id"), guardrail_name=guardrail.get("guardrail_name"), - litellm_params=dict(guardrail.get("litellm_params") or {}), + litellm_params=masked_in_memory_litellm_params_typed, guardrail_info=dict(guardrail.get("guardrail_info") or {}), guardrail_definition_location="config", ) @@ -666,11 +702,16 @@ async def get_guardrail_info(guardrail_id: str): unmasked_length=4, number_of_asterisks=4, ) + masked_litellm_params = ( + BaseLitellmParams(**masked_litellm_params_dict) + if masked_litellm_params_dict + else None + ) return GuardrailInfoResponse( guardrail_id=result.get("guardrail_id"), guardrail_name=result.get("guardrail_name"), - litellm_params=masked_litellm_params_dict, + litellm_params=masked_litellm_params, guardrail_info=dict(result.get("guardrail_info") or {}), created_at=result.get("created_at"), updated_at=result.get("updated_at"), diff --git a/litellm/proxy/guardrails/guardrail_hooks/bedrock_guardrails.py b/litellm/proxy/guardrails/guardrail_hooks/bedrock_guardrails.py index 528857f5dd1..6800dff55ac 100644 --- a/litellm/proxy/guardrails/guardrail_hooks/bedrock_guardrails.py +++ b/litellm/proxy/guardrails/guardrail_hooks/bedrock_guardrails.py @@ -461,12 +461,37 @@ class BedrockGuardrail(CustomGuardrail, BaseAWSLLM): data=prepared_request.body, # type: ignore headers=prepared_request.headers, # type: ignore ) + except HTTPException: + # Propagate HTTPException (e.g. from non-200 path) as-is + raise except Exception as e: + # If this is an HTTP error with a response body (e.g. httpx.HTTPStatusError), + # extract the AWS error message and propagate it + response = getattr(e, "response", None) + if isinstance(response, httpx.Response): + try: + status_code, detail_message = ( + self._parse_bedrock_guardrail_error_response(response) + ) + self.add_standard_logging_guardrail_information_to_request_data( + guardrail_provider=self.guardrail_provider, + guardrail_json_response={"error": detail_message}, + request_data=request_data or {}, + guardrail_status="guardrail_failed_to_respond", + start_time=start_time.timestamp(), + end_time=datetime.now().timestamp(), + duration=(datetime.now() - start_time).total_seconds(), + event_type=event_type, + ) + raise HTTPException( + status_code=status_code, detail=detail_message + ) from e + except HTTPException: + raise # Endpoint down, timeout, or other HTTP/network errors verbose_proxy_logger.error( "Bedrock AI: failed to make guardrail request: %s", str(e) ) - # Add guardrail information with failure status self.add_standard_logging_guardrail_information_to_request_data( guardrail_provider=self.guardrail_provider, guardrail_json_response={"error": str(e)}, @@ -477,7 +502,6 @@ class BedrockGuardrail(CustomGuardrail, BaseAWSLLM): duration=(datetime.now() - start_time).total_seconds(), event_type=event_type, ) - # Re-raise the exception to maintain existing behavior raise ######################################################### @@ -509,11 +533,15 @@ class BedrockGuardrail(CustomGuardrail, BaseAWSLLM): bedrock_guardrail_response ) else: + status_code, detail_message = self._parse_bedrock_guardrail_error_response( + httpx_response + ) verbose_proxy_logger.error( "Bedrock AI: error in response. Status code: %s, response: %s", httpx_response.status_code, httpx_response.text, ) + raise HTTPException(status_code=status_code, detail=detail_message) return bedrock_guardrail_response @@ -579,6 +607,34 @@ class BedrockGuardrail(CustomGuardrail, BaseAWSLLM): return "success" return "guardrail_failed_to_respond" + def _parse_bedrock_guardrail_error_response( + self, response: httpx.Response + ) -> Tuple[int, str]: + """ + Parse AWS Bedrock guardrail error response body to extract status code and message. + + AWS may return shapes like {"message": "..."} or {"error": {"message": "..."}}. + Returns (status_code, message) for use in HTTPException. + """ + status_code = response.status_code + message = "Bedrock guardrail request failed" + try: + body = response.json() + except Exception: + text = getattr(response, "text", None) or "" + if isinstance(text, str) and text.strip(): + return (status_code, text.strip()) + return (status_code, message) + if isinstance(body, dict): + if isinstance(body.get("message"), str): + return (status_code, body["message"]) + err = body.get("error") + if isinstance(err, dict) and isinstance(err.get("message"), str): + return (status_code, err["message"]) + if isinstance(err, str): + return (status_code, err) + return (status_code, message) + def _get_http_exception_for_blocked_guardrail( self, response: BedrockGuardrailResponse ) -> Union[HTTPException, GuardrailInterventionNormalStringError]: @@ -739,9 +795,9 @@ class BedrockGuardrail(CustomGuardrail, BaseAWSLLM): ######################################################### ########## 1. Make the Bedrock API request ########## ######################################################### - bedrock_guardrail_response: Optional[ - Union[BedrockGuardrailResponse, str] - ] = None + bedrock_guardrail_response: Optional[Union[BedrockGuardrailResponse, str]] = ( + None + ) try: bedrock_guardrail_response = await self.make_bedrock_api_request( source="INPUT", messages=filtered_messages, request_data=data @@ -811,9 +867,9 @@ class BedrockGuardrail(CustomGuardrail, BaseAWSLLM): ######################################################### ########## 1. Make the Bedrock API request ########## ######################################################### - bedrock_guardrail_response: Optional[ - Union[BedrockGuardrailResponse, str] - ] = None + bedrock_guardrail_response: Optional[Union[BedrockGuardrailResponse, str]] = ( + None + ) try: bedrock_guardrail_response = await self.make_bedrock_api_request( source="INPUT", messages=filtered_messages, request_data=data diff --git a/litellm/proxy/guardrails/guardrail_hooks/custom_code/custom_code_guardrail.py b/litellm/proxy/guardrails/guardrail_hooks/custom_code/custom_code_guardrail.py index 68f9dfd7abc..66b80c10f18 100644 --- a/litellm/proxy/guardrails/guardrail_hooks/custom_code/custom_code_guardrail.py +++ b/litellm/proxy/guardrails/guardrail_hooks/custom_code/custom_code_guardrail.py @@ -35,7 +35,10 @@ from typing import TYPE_CHECKING, Any, Dict, Literal, Optional, Type, cast from fastapi import HTTPException from litellm._logging import verbose_proxy_logger -from litellm.integrations.custom_guardrail import CustomGuardrail +from litellm.integrations.custom_guardrail import ( + CustomGuardrail, + log_guardrail_information, +) from litellm.types.guardrails import GuardrailEventHooks from litellm.types.proxy.guardrails.guardrail_hooks.base import GuardrailConfigModel from litellm.types.utils import GenericGuardrailAPIInputs @@ -179,6 +182,7 @@ class CustomCodeGuardrail(CustomGuardrail): self._compile_error = f"Failed to compile custom code: {e}" raise CustomCodeCompilationError(self._compile_error) from e + @log_guardrail_information async def apply_guardrail( self, inputs: GenericGuardrailAPIInputs, diff --git a/litellm/proxy/guardrails/guardrail_hooks/enkryptai/enkryptai.py b/litellm/proxy/guardrails/guardrail_hooks/enkryptai/enkryptai.py index 8e992297e5d..63541a1e2f9 100644 --- a/litellm/proxy/guardrails/guardrail_hooks/enkryptai/enkryptai.py +++ b/litellm/proxy/guardrails/guardrail_hooks/enkryptai/enkryptai.py @@ -23,7 +23,10 @@ import httpx import litellm from litellm._logging import verbose_proxy_logger from litellm.caching.caching import DualCache -from litellm.integrations.custom_guardrail import CustomGuardrail +from litellm.integrations.custom_guardrail import ( + CustomGuardrail, + log_guardrail_information, +) from litellm.llms.custom_httpx.http_handler import ( get_async_httpx_client, httpxSpecialProvider, @@ -483,6 +486,7 @@ class EnkryptAIGuardrails(CustomGuardrail): request_data=data, guardrail_name=self.guardrail_name ) + @log_guardrail_information async def apply_guardrail( self, inputs: "GenericGuardrailAPIInputs", diff --git a/litellm/proxy/guardrails/guardrail_hooks/generic_guardrail_api/generic_guardrail_api.py b/litellm/proxy/guardrails/guardrail_hooks/generic_guardrail_api/generic_guardrail_api.py index b37074e25e7..9018675d7a5 100644 --- a/litellm/proxy/guardrails/guardrail_hooks/generic_guardrail_api/generic_guardrail_api.py +++ b/litellm/proxy/guardrails/guardrail_hooks/generic_guardrail_api/generic_guardrail_api.py @@ -10,7 +10,10 @@ from typing import TYPE_CHECKING, Any, Dict, Literal, Optional from litellm._logging import verbose_proxy_logger from litellm.exceptions import GuardrailRaisedException -from litellm.integrations.custom_guardrail import CustomGuardrail +from litellm.integrations.custom_guardrail import ( + CustomGuardrail, + log_guardrail_information, +) from litellm.llms.custom_httpx.http_handler import ( get_async_httpx_client, httpxSpecialProvider, @@ -150,6 +153,7 @@ class GenericGuardrailAPI(CustomGuardrail): return result_metadata + @log_guardrail_information async def apply_guardrail( self, inputs: GenericGuardrailAPIInputs, diff --git a/litellm/proxy/guardrails/guardrail_hooks/grayswan/grayswan.py b/litellm/proxy/guardrails/guardrail_hooks/grayswan/grayswan.py index 90f689ed23c..8955bffc125 100644 --- a/litellm/proxy/guardrails/guardrail_hooks/grayswan/grayswan.py +++ b/litellm/proxy/guardrails/guardrail_hooks/grayswan/grayswan.py @@ -9,7 +9,8 @@ from fastapi import HTTPException from litellm._logging import verbose_proxy_logger from litellm.integrations.custom_guardrail import ( CustomGuardrail, - ModifyResponseException + ModifyResponseException, + log_guardrail_information, ) from litellm.litellm_core_utils.safe_json_dumps import safe_dumps from litellm.litellm_core_utils.safe_json_loads import safe_json_loads @@ -108,7 +109,9 @@ class GraySwanGuardrail(CustomGuardrail): self.categories = categories self.policy_id = policy_id self.fail_open = True if fail_open is None else bool(fail_open) - self.guardrail_timeout = 30.0 if guardrail_timeout is None else float(guardrail_timeout) + self.guardrail_timeout = ( + 30.0 if guardrail_timeout is None else float(guardrail_timeout) + ) # Streaming configuration self.streaming_end_of_stream_only = streaming_end_of_stream_only @@ -155,6 +158,7 @@ class GraySwanGuardrail(CustomGuardrail): # Unified Guardrail Interface (works with ALL endpoints automatically) # ------------------------------------------------------------------ + @log_guardrail_information async def apply_guardrail( self, inputs: GenericGuardrailAPIInputs, @@ -208,7 +212,9 @@ class GraySwanGuardrail(CustomGuardrail): messages = [{"role": role, "content": text} for text in texts] # Get dynamic params from request metadata - dynamic_body = self.get_guardrail_dynamic_request_body_params(request_data) or {} + dynamic_body = ( + self.get_guardrail_dynamic_request_body_params(request_data) or {} + ) if dynamic_body: verbose_proxy_logger.debug( "Gray Swan Guardrail: dynamic extra_body=%s", safe_dumps(dynamic_body) @@ -271,12 +277,12 @@ class GraySwanGuardrail(CustomGuardrail): async def run_grayswan_guardrail(self, payload: dict) -> Dict[str, Any]: """ Run the GraySwan guardrail on a payload. - + This is a legacy method for testing purposes. - + Args: payload: The payload to scan - + Returns: Dict containing the GraySwan API response """ @@ -293,11 +299,11 @@ class GraySwanGuardrail(CustomGuardrail): ) -> None: """ Legacy method for processing GraySwan API responses. - + This method is maintained for backward compatibility with existing tests. It handles the test scenarios where responses need to be processed with knowledge of the request context (pre/during/post call hooks). - + Args: response_json: Response from GraySwan API data: Optional request data (for passthrough exceptions) @@ -365,7 +371,10 @@ class GraySwanGuardrail(CustomGuardrail): ) # If hook_type is provided and in pre/during call, raise exception - if hook_type in [GuardrailEventHooks.pre_call, GuardrailEventHooks.during_call]: + if hook_type in [ + GuardrailEventHooks.pre_call, + GuardrailEventHooks.during_call, + ]: # Raise ModifyResponseException to short-circuit LLM call if data is None: data = {} @@ -540,7 +549,9 @@ class GraySwanGuardrail(CustomGuardrail): if isinstance(litellm_metadata, dict) and litellm_metadata: cleaned_litellm_metadata = dict(litellm_metadata) # cleaned_litellm_metadata.pop("user_api_key_auth", None) - sanitized = safe_json_loads(safe_dumps(cleaned_litellm_metadata), default={}) + sanitized = safe_json_loads( + safe_dumps(cleaned_litellm_metadata), default={} + ) if isinstance(sanitized, dict) and sanitized: payload["litellm_metadata"] = sanitized @@ -566,7 +577,9 @@ class GraySwanGuardrail(CustomGuardrail): detection_info = detection_info[0] # Extract fields from detection_info dict - detection_dict: dict = detection_info if isinstance(detection_info, dict) else {} + detection_dict: dict = ( + detection_info if isinstance(detection_info, dict) else {} + ) violation_score = detection_dict.get("violation_score", 0.0) violated_rules = detection_dict.get("violated_rules", []) mutation = detection_dict.get("mutation", False) @@ -582,7 +595,9 @@ class GraySwanGuardrail(CustomGuardrail): if violated_rules: formatted_rules = self._format_violated_rules(violated_rules) if formatted_rules: - message_parts.append(f"It was violating the rule(s): {formatted_rules}.") + message_parts.append( + f"It was violating the rule(s): {formatted_rules}." + ) if mutation: message_parts.append( @@ -590,9 +605,7 @@ class GraySwanGuardrail(CustomGuardrail): ) if ipi: - message_parts.append( - "Indirect Prompt Injection was DETECTED." - ) + message_parts.append("Indirect Prompt Injection was DETECTED.") return "\n".join(message_parts) diff --git a/litellm/proxy/guardrails/guardrail_hooks/hiddenlayer/hiddenlayer.py b/litellm/proxy/guardrails/guardrail_hooks/hiddenlayer/hiddenlayer.py index e2c20604880..b907fbbcbda 100644 --- a/litellm/proxy/guardrails/guardrail_hooks/hiddenlayer/hiddenlayer.py +++ b/litellm/proxy/guardrails/guardrail_hooks/hiddenlayer/hiddenlayer.py @@ -10,7 +10,10 @@ from httpx import HTTPStatusError from requests.auth import HTTPBasicAuth from litellm._logging import verbose_proxy_logger -from litellm.integrations.custom_guardrail import CustomGuardrail +from litellm.integrations.custom_guardrail import ( + CustomGuardrail, + log_guardrail_information, +) from litellm.litellm_core_utils.litellm_logging import Logging as LiteLLMLoggingObj from litellm.llms.custom_httpx.http_handler import ( get_async_httpx_client, @@ -110,6 +113,7 @@ class HiddenlayerGuardrail(CustomGuardrail): ) super().__init__(**kwargs) + @log_guardrail_information async def apply_guardrail( self, inputs: GenericGuardrailAPIInputs, diff --git a/litellm/proxy/guardrails/guardrail_hooks/litellm_content_filter/content_filter.py b/litellm/proxy/guardrails/guardrail_hooks/litellm_content_filter/content_filter.py index 083a407e9cf..263b6eee768 100644 --- a/litellm/proxy/guardrails/guardrail_hooks/litellm_content_filter/content_filter.py +++ b/litellm/proxy/guardrails/guardrail_hooks/litellm_content_filter/content_filter.py @@ -28,7 +28,10 @@ from fastapi import HTTPException from litellm import Router from litellm._logging import verbose_proxy_logger -from litellm.integrations.custom_guardrail import CustomGuardrail +from litellm.integrations.custom_guardrail import ( + CustomGuardrail, + log_guardrail_information, +) from litellm.proxy._types import UserAPIKeyAuth from litellm.types.utils import ModelResponseStream @@ -50,6 +53,7 @@ from litellm.types.proxy.guardrails.guardrail_hooks.litellm_content_filter impor ContentFilterDetection, PatternDetection, ) + from .patterns import PATTERN_EXTRA_CONFIG, get_compiled_pattern MAX_KEYWORD_VALUE_GAP_WORDS = 1 @@ -168,9 +172,9 @@ class ContentFilterGuardrail(CustomGuardrail): self.image_model = image_model # Store loaded categories self.loaded_categories: Dict[str, CategoryConfig] = {} - self.category_keywords: Dict[ - str, Tuple[str, str, ContentFilterAction] - ] = {} # keyword -> (category, severity, action) + self.category_keywords: Dict[str, Tuple[str, str, ContentFilterAction]] = ( + {} + ) # keyword -> (category, severity, action) # Load categories if provided if categories: @@ -994,6 +998,7 @@ class ContentFilterGuardrail(CustomGuardrail): masked_entity_count=masked_entity_count, ) + @log_guardrail_information async def apply_guardrail( self, inputs: "GenericGuardrailAPIInputs", diff --git a/litellm/proxy/guardrails/guardrail_hooks/onyx/onyx.py b/litellm/proxy/guardrails/guardrail_hooks/onyx/onyx.py index 3598dbe741e..1cfc805dbf9 100644 --- a/litellm/proxy/guardrails/guardrail_hooks/onyx/onyx.py +++ b/litellm/proxy/guardrails/guardrail_hooks/onyx/onyx.py @@ -12,7 +12,10 @@ import httpx from fastapi import HTTPException from litellm._logging import verbose_proxy_logger -from litellm.integrations.custom_guardrail import CustomGuardrail +from litellm.integrations.custom_guardrail import ( + CustomGuardrail, + log_guardrail_information, +) from litellm.litellm_core_utils.litellm_logging import Logging as LiteLLMLoggingObj from litellm.llms.custom_httpx.http_handler import ( get_async_httpx_client, @@ -26,7 +29,11 @@ if TYPE_CHECKING: class OnyxGuardrail(CustomGuardrail): def __init__( - self, api_base: Optional[str] = None, api_key: Optional[str] = None, timeout: Optional[float] = 10.0, **kwargs + self, + api_base: Optional[str] = None, + api_key: Optional[str] = None, + timeout: Optional[float] = 10.0, + **kwargs, ): timeout = timeout or int(os.getenv("ONYX_TIMEOUT", 10.0)) self.async_handler = get_async_httpx_client( @@ -79,6 +86,7 @@ class OnyxGuardrail(CustomGuardrail): ) return result + @log_guardrail_information async def apply_guardrail( self, inputs: GenericGuardrailAPIInputs, diff --git a/litellm/proxy/guardrails/guardrail_hooks/openai/moderations.py b/litellm/proxy/guardrails/guardrail_hooks/openai/moderations.py index 030b6036815..a196937ef6c 100644 --- a/litellm/proxy/guardrails/guardrail_hooks/openai/moderations.py +++ b/litellm/proxy/guardrails/guardrail_hooks/openai/moderations.py @@ -58,7 +58,9 @@ class OpenAIModerationGuardrail(OpenAIGuardrailBase, CustomGuardrail): guardrail_name: str, api_key: Optional[str] = None, api_base: Optional[str] = None, - model: Optional[Literal["omni-moderation-latest", "text-moderation-latest"]] = None, + model: Optional[ + Literal["omni-moderation-latest", "text-moderation-latest"] + ] = None, **kwargs, ): """Initialize OpenAI Moderation guardrail handler.""" @@ -75,7 +77,7 @@ class OpenAIModerationGuardrail(OpenAIGuardrailBase, CustomGuardrail): supported_event_hooks=supported_event_hooks, **kwargs, ) - + self.async_handler = get_async_httpx_client( llm_provider=httpxSpecialProvider.GuardrailCallback ) @@ -83,10 +85,14 @@ class OpenAIModerationGuardrail(OpenAIGuardrailBase, CustomGuardrail): # Store configuration self.api_key = api_key or self._get_api_key() self.api_base = api_base or "https://api.openai.com/v1" - self.model: Literal["omni-moderation-latest", "text-moderation-latest"] = model or "omni-moderation-latest" + self.model: Literal["omni-moderation-latest", "text-moderation-latest"] = ( + model or "omni-moderation-latest" + ) if not self.api_key: - raise ValueError("OpenAI Moderation: api_key is required. Set OPENAI_API_KEY environment variable or pass it in configuration.") + raise ValueError( + "OpenAI Moderation: api_key is required. Set OPENAI_API_KEY environment variable or pass it in configuration." + ) verbose_proxy_logger.debug( f"Initialized OpenAI Moderation Guardrail: {guardrail_name} with model: {self.model}" @@ -98,7 +104,7 @@ class OpenAIModerationGuardrail(OpenAIGuardrailBase, CustomGuardrail): import litellm from litellm.secret_managers.main import get_secret_str - + return ( os.environ.get("OPENAI_API_KEY") or litellm.api_key @@ -106,21 +112,14 @@ class OpenAIModerationGuardrail(OpenAIGuardrailBase, CustomGuardrail): or get_secret_str("OPENAI_API_KEY") ) - async def async_make_request( - self, input_text: str - ) -> "OpenAIModerationResponse": + async def async_make_request(self, input_text: str) -> "OpenAIModerationResponse": """ Make a request to the OpenAI Moderation API. """ - request_body = { - "model": self.model, - "input": input_text - } - - verbose_proxy_logger.debug( - "OpenAI Moderation guard request: %s", request_body - ) - + request_body = {"model": self.model, "input": input_text} + + verbose_proxy_logger.debug("OpenAI Moderation guard request: %s", request_body) + response = await self.async_handler.post( url=f"{self.api_base}/moderations", headers={ @@ -133,7 +132,7 @@ class OpenAIModerationGuardrail(OpenAIGuardrailBase, CustomGuardrail): verbose_proxy_logger.debug( "OpenAI Moderation guard response: %s", response.json() ) - + if response.status_code != 200: raise HTTPException( status_code=response.status_code, @@ -144,9 +143,12 @@ class OpenAIModerationGuardrail(OpenAIGuardrailBase, CustomGuardrail): ) from litellm.types.llms.openai import OpenAIModerationResponse + return OpenAIModerationResponse(**response.json()) - def _check_moderation_result(self, moderation_response: "OpenAIModerationResponse") -> None: + def _check_moderation_result( + self, moderation_response: "OpenAIModerationResponse" + ) -> None: """ Check if the moderation response indicates harmful content and raise exception if needed. """ @@ -168,10 +170,10 @@ class OpenAIModerationGuardrail(OpenAIGuardrailBase, CustomGuardrail): } verbose_proxy_logger.warning( - "OpenAI Moderation: Content flagged for violations: %s", - violation_details + "OpenAI Moderation: Content flagged for violations: %s", + violation_details, ) - + raise HTTPException( status_code=400, detail={ @@ -180,6 +182,7 @@ class OpenAIModerationGuardrail(OpenAIGuardrailBase, CustomGuardrail): }, ) + @log_guardrail_information async def apply_guardrail( self, inputs: GenericGuardrailAPIInputs, @@ -189,51 +192,50 @@ class OpenAIModerationGuardrail(OpenAIGuardrailBase, CustomGuardrail): ) -> GenericGuardrailAPIInputs: """ Apply OpenAI moderation guardrail using the unified guardrail interface. - + This method is called by the UnifiedLLMGuardrails system for all endpoint types (chat completions, embeddings, responses API, etc.). - + Args: inputs: GenericGuardrailAPIInputs containing texts and/or structured_messages request_data: The original request data input_type: Whether this is a "request" (pre-call) or "response" (post-call) logging_obj: Optional logging object - + Returns: The inputs unchanged (moderation doesn't modify content, only blocks) - + Raises: HTTPException: If content violates moderation policy """ # Extract text to moderate from inputs text_to_moderate: Optional[str] = None - + # Prefer structured_messages if available (has role context) if structured_messages := inputs.get("structured_messages"): text_to_moderate = self.get_user_prompt(structured_messages) - + # Fall back to texts if not text_to_moderate: if texts := inputs.get("texts"): # Join all texts for moderation text_to_moderate = "\n".join(texts) - + if not text_to_moderate: verbose_proxy_logger.debug( "OpenAI Moderation: No text content to moderate in inputs" ) return inputs - + # Make moderation request moderation_response = await self.async_make_request(input_text=text_to_moderate) - + # Check if content is flagged and raise exception if needed self._check_moderation_result(moderation_response) - + # Moderation doesn't modify content, just blocks - return inputs unchanged return inputs - @log_guardrail_information async def async_post_call_streaming_iterator_hook( self, @@ -252,9 +254,7 @@ class OpenAIModerationGuardrail(OpenAIGuardrailBase, CustomGuardrail): from litellm.main import stream_chunk_builder from litellm.types.utils import TextCompletionResponse - verbose_proxy_logger.debug( - "OpenAI Moderation: Running streaming response scan" - ) + verbose_proxy_logger.debug("OpenAI Moderation: Running streaming response scan") # Collect all chunks to process them together all_chunks: List["ModelResponseStream"] = [] @@ -269,7 +269,7 @@ class OpenAIModerationGuardrail(OpenAIGuardrailBase, CustomGuardrail): ) if isinstance(assembled_model_response, (type(None), TextCompletionResponse)): - # If we can't assemble a ModelResponse or it's a text completion, + # If we can't assemble a ModelResponse or it's a text completion, # just yield the original chunks without moderation verbose_proxy_logger.warning( "OpenAI Moderation: Could not assemble ModelResponse from chunks, skipping moderation" @@ -284,19 +284,17 @@ class OpenAIModerationGuardrail(OpenAIGuardrailBase, CustomGuardrail): verbose_proxy_logger.debug( f"OpenAI Moderation: Streaming response text: {response_text[:100]}..." # Log first 100 chars ) - + # Make moderation request - this will raise HTTPException if content is flagged moderation_response = await self.async_make_request( input_text=response_text, ) - + # Check if content is flagged and raise exception if needed self._check_moderation_result(moderation_response) # If we reach here, content passed moderation - yield the original chunks - mock_response = MockResponseIterator( - model_response=assembled_model_response - ) + mock_response = MockResponseIterator(model_response=assembled_model_response) # Return the reconstructed stream async for chunk in mock_response: @@ -306,34 +304,34 @@ class OpenAIModerationGuardrail(OpenAIGuardrailBase, CustomGuardrail): """ Extract text content from the model response for moderation. """ - if not hasattr(response, 'choices') or not response.choices: + if not hasattr(response, "choices") or not response.choices: return None response_texts = [] for choice in response.choices: try: # Try to get content from message (chat completion) - message = getattr(choice, 'message', None) + message = getattr(choice, "message", None) if message: - content = getattr(message, 'content', None) + content = getattr(message, "content", None) if content and isinstance(content, str): response_texts.append(content) continue - + # Try to get text (text completion) - text = getattr(choice, 'text', None) + text = getattr(choice, "text", None) if text and isinstance(text, str): response_texts.append(text) continue - + # Try to get content from delta (streaming) - delta = getattr(choice, 'delta', None) + delta = getattr(choice, "delta", None) if delta: - content = getattr(delta, 'content', None) + content = getattr(delta, "content", None) if content and isinstance(content, str): response_texts.append(content) continue - + except (AttributeError, TypeError): # Skip choices that don't have expected attributes continue diff --git a/litellm/proxy/guardrails/guardrail_hooks/presidio.py b/litellm/proxy/guardrails/guardrail_hooks/presidio.py index 71ad9819146..3984384aae4 100644 --- a/litellm/proxy/guardrails/guardrail_hooks/presidio.py +++ b/litellm/proxy/guardrails/guardrail_hooks/presidio.py @@ -9,10 +9,10 @@ import asyncio -import threading import json -from datetime import datetime +import threading from contextlib import asynccontextmanager +from datetime import datetime from typing import ( TYPE_CHECKING, Any, @@ -38,8 +38,11 @@ if TYPE_CHECKING: from litellm._uuid import uuid from litellm.caching.caching import DualCache -from litellm.exceptions import BlockedPiiEntityError -from litellm.integrations.custom_guardrail import CustomGuardrail +from litellm.exceptions import BlockedPiiEntityError, GuardrailRaisedException +from litellm.integrations.custom_guardrail import ( + CustomGuardrail, + log_guardrail_information, +) from litellm.proxy._types import UserAPIKeyAuth from litellm.types.guardrails import ( GuardrailEventHooks, @@ -229,6 +232,14 @@ class _OPTIONAL_PresidioPIIMasking(CustomGuardrail): """Cleanup: we try to close, but doing async cleanup in __del__ is risky.""" pass + def _has_block_action(self) -> bool: + """Return True if pii_entities_config has any BLOCK action (fail-closed on analyzer errors).""" + if not self.pii_entities_config: + return False + return any( + action == PiiAction.BLOCK for action in self.pii_entities_config.values() + ) + def _get_presidio_analyze_request_payload( self, text: str, @@ -313,13 +324,30 @@ class _OPTIONAL_PresidioPIIMasking(CustomGuardrail): # Handle error responses from Presidio (e.g., {'error': 'No text provided'}) # Presidio may return a dict instead of a list when errors occur + def _fail_on_invalid_response( + reason: str, + ) -> List[PresidioAnalyzeResponseItem]: + should_fail_closed = ( + bool(self.pii_entities_config) + or self.output_parse_pii + or self.apply_to_output + ) + if should_fail_closed: + raise GuardrailRaisedException( + guardrail_name=self.guardrail_name, + message=f"Presidio analyzer returned invalid response; cannot verify PII when PII protection is configured: {reason}", + should_wrap_with_default_message=False, + ) + verbose_proxy_logger.warning( + "Presidio analyzer %s, returning empty list", reason + ) + return [] + if isinstance(analyze_results, dict): if "error" in analyze_results: - verbose_proxy_logger.warning( - "Presidio analyzer returned error: %s, returning empty list", - analyze_results.get("error"), + return _fail_on_invalid_response( + f"error: {analyze_results.get('error')}" ) - return [] # If it's a dict but not an error, try to process it as a single item verbose_proxy_logger.debug( "Presidio returned dict (not list), attempting to process as single item" @@ -327,23 +355,33 @@ class _OPTIONAL_PresidioPIIMasking(CustomGuardrail): try: return [PresidioAnalyzeResponseItem(**analyze_results)] except Exception as e: - verbose_proxy_logger.warning( - "Failed to parse Presidio dict response: %s, returning empty list", - e, + return _fail_on_invalid_response( + f"failed to parse dict response: {e}" ) - return [] + + # Handle unexpected types (str, None, etc.) - e.g. from malformed/error + if not isinstance(analyze_results, list): + return _fail_on_invalid_response( + f"unexpected type {type(analyze_results).__name__} (expected list or dict), response: {str(analyze_results)[:200]}" + ) # Normal case: list of results final_results = [] for item in analyze_results: + if not isinstance(item, dict): + verbose_proxy_logger.warning( + "Skipping invalid Presidio result item (expected dict, got %s): %s", + type(item).__name__, + str(item)[:100], + ) + continue try: final_results.append(PresidioAnalyzeResponseItem(**item)) - except TypeError as te: - # Handle case where item is not a dict (shouldn't happen, but be defensive) + except Exception as e: verbose_proxy_logger.warning( - "Skipping invalid Presidio result item: %s (error: %s)", + "Failed to parse Presidio result item: %s (error: %s)", item, - te, + e, ) continue return final_results @@ -568,9 +606,9 @@ class _OPTIONAL_PresidioPIIMasking(CustomGuardrail): if messages is None: return data tasks = [] - task_mappings: List[ - Tuple[int, Optional[int]] - ] = [] # Track (message_index, content_index) for each task + task_mappings: List[Tuple[int, Optional[int]]] = ( + [] + ) # Track (message_index, content_index) for each task for msg_idx, m in enumerate(messages): content = m.get("content", None) @@ -671,9 +709,9 @@ class _OPTIONAL_PresidioPIIMasking(CustomGuardrail): ): # /chat/completions requests messages: Optional[List] = kwargs.get("messages", None) tasks = [] - task_mappings: List[ - Tuple[int, Optional[int]] - ] = [] # Track (message_index, content_index) for each task + task_mappings: List[Tuple[int, Optional[int]]] = ( + [] + ) # Track (message_index, content_index) for each task if messages is None: return kwargs, result @@ -792,11 +830,11 @@ class _OPTIONAL_PresidioPIIMasking(CustomGuardrail): # Type narrowing: StreamingChoices doesn't have .message attribute if not hasattr(choice, "message"): continue - content = getattr(choice.message, "content", None) + content = getattr(choice.message, "content", None) # type: ignore if content is None: continue if isinstance(content, str): - choice.message.content = await self.check_pii( + choice.message.content = await self.check_pii( # type: ignore text=content, output_parse_pii=False, presidio_config=presidio_config, @@ -989,6 +1027,7 @@ class _OPTIONAL_PresidioPIIMasking(CustomGuardrail): except Exception: pass + @log_guardrail_information async def apply_guardrail( self, inputs: "GenericGuardrailAPIInputs", diff --git a/litellm/proxy/guardrails/guardrail_hooks/prompt_security/prompt_security.py b/litellm/proxy/guardrails/guardrail_hooks/prompt_security/prompt_security.py index 5ebc7b96eb8..b3e761869b0 100644 --- a/litellm/proxy/guardrails/guardrail_hooks/prompt_security/prompt_security.py +++ b/litellm/proxy/guardrails/guardrail_hooks/prompt_security/prompt_security.py @@ -6,7 +6,10 @@ from typing import TYPE_CHECKING, Any, List, Literal, Optional, Type from fastapi import HTTPException from litellm._logging import verbose_proxy_logger -from litellm.integrations.custom_guardrail import CustomGuardrail +from litellm.integrations.custom_guardrail import ( + CustomGuardrail, + log_guardrail_information, +) from litellm.llms.custom_httpx.http_handler import ( get_async_httpx_client, httpxSpecialProvider, @@ -67,6 +70,7 @@ class PromptSecurityGuardrail(CustomGuardrail): super().__init__(**kwargs) + @log_guardrail_information async def apply_guardrail( self, inputs: GenericGuardrailAPIInputs, diff --git a/litellm/proxy/guardrails/guardrail_hooks/qualifire/qualifire.py b/litellm/proxy/guardrails/guardrail_hooks/qualifire/qualifire.py index 87da11efad0..6486da7f714 100644 --- a/litellm/proxy/guardrails/guardrail_hooks/qualifire/qualifire.py +++ b/litellm/proxy/guardrails/guardrail_hooks/qualifire/qualifire.py @@ -12,10 +12,11 @@ from typing import Any, Dict, List, Literal, Optional, Type from fastapi import HTTPException from litellm._logging import verbose_proxy_logger -from litellm.integrations.custom_guardrail import CustomGuardrail -from litellm.litellm_core_utils.litellm_logging import ( - Logging as LiteLLMLoggingObj, +from litellm.integrations.custom_guardrail import ( + CustomGuardrail, + log_guardrail_information, ) +from litellm.litellm_core_utils.litellm_logging import Logging as LiteLLMLoggingObj from litellm.llms.custom_httpx.http_handler import ( get_async_httpx_client, httpxSpecialProvider, @@ -343,9 +344,7 @@ class QualifireGuardrail(CustomGuardrail): ) url = f"{self.qualifire_api_base}/api/evaluation/evaluate" - verbose_proxy_logger.debug( - f"Qualifire Guardrail: Making request to {url}" - ) + verbose_proxy_logger.debug(f"Qualifire Guardrail: Making request to {url}") # Make the API request response = await self.async_handler.post( @@ -393,6 +392,7 @@ class QualifireGuardrail(CustomGuardrail): verbose_proxy_logger.exception(f"Qualifire Guardrail error: {e}") raise + @log_guardrail_information async def apply_guardrail( self, inputs: GenericGuardrailAPIInputs, diff --git a/litellm/proxy/guardrails/guardrail_hooks/zscaler_ai_guard/zscaler_ai_guard.py b/litellm/proxy/guardrails/guardrail_hooks/zscaler_ai_guard/zscaler_ai_guard.py index c60752d7952..ff00cd73ca5 100644 --- a/litellm/proxy/guardrails/guardrail_hooks/zscaler_ai_guard/zscaler_ai_guard.py +++ b/litellm/proxy/guardrails/guardrail_hooks/zscaler_ai_guard/zscaler_ai_guard.py @@ -9,7 +9,10 @@ from typing import TYPE_CHECKING, Literal, Optional from fastapi import HTTPException from litellm._logging import verbose_proxy_logger -from litellm.integrations.custom_guardrail import CustomGuardrail +from litellm.integrations.custom_guardrail import ( + CustomGuardrail, + log_guardrail_information, +) from litellm.llms.custom_httpx.http_handler import ( get_async_httpx_client, httpxSpecialProvider, @@ -70,6 +73,7 @@ class ZscalerAIGuard(CustomGuardrail): return str(value).strip() return "N/A" + @log_guardrail_information async def apply_guardrail( self, inputs: "GenericGuardrailAPIInputs", @@ -92,7 +96,7 @@ class ZscalerAIGuard(CustomGuardrail): Raises: Exception: If content is blocked by Zscaler AI Guard """ - + texts = inputs.get("texts", []) try: verbose_proxy_logger.debug(f"ZscalerAIGuard: Checking {len(texts)} text(s)") @@ -102,8 +106,8 @@ class ZscalerAIGuard(CustomGuardrail): team_metadata = metadata.get("team_metadata", {}) or {} # Precedence for policy_id: - # 1. metadata.zguard_policy_id # request level - # 2. user_api_key_metadata.zguard_policy_id # Key level + # 1. metadata.zguard_policy_id # request level + # 2. user_api_key_metadata.zguard_policy_id # Key level # 3. team_metadata.zguard_policy_id # Team level # 4. self.policy_id (from environment) # Global policy_id = ( @@ -154,9 +158,7 @@ class ZscalerAIGuard(CustomGuardrail): zscaler_ai_guard_result and zscaler_ai_guard_result.get("action") == "BLOCK" ): - blocking_info = zscaler_ai_guard_result.get( - "zscaler_ai_guard_response" - ) + blocking_info = zscaler_ai_guard_result.get("zscaler_ai_guard_response") error_message = f"Content blocked by Zscaler AI Guard: {self.extract_blocking_info(blocking_info)}" raise Exception(error_message) except Exception as e: diff --git a/litellm/proxy/hooks/batch_rate_limiter.py b/litellm/proxy/hooks/batch_rate_limiter.py index ecad8bc1b11..78a371ad66e 100644 --- a/litellm/proxy/hooks/batch_rate_limiter.py +++ b/litellm/proxy/hooks/batch_rate_limiter.py @@ -17,7 +17,7 @@ Quick summary: - async_log_success_event() fires on GET /v1/batches/{id} (batch completion) """ -from typing import TYPE_CHECKING, Any, Dict, List, Literal, Union +from typing import TYPE_CHECKING, Any, Dict, List, Literal, Optional, Union from fastapi import HTTPException from pydantic import BaseModel @@ -241,6 +241,7 @@ class _PROXY_BatchRateLimiter(CustomLogger): self, file_id: str, custom_llm_provider: Literal["openai", "azure", "vertex_ai"] = "openai", + user_api_key_dict: Optional[UserAPIKeyAuth] = None, ) -> BatchFileUsage: """ Count number of requests and tokens in a batch input file. @@ -248,6 +249,7 @@ class _PROXY_BatchRateLimiter(CustomLogger): Args: file_id: The file ID to read custom_llm_provider: The custom LLM provider to use for token encoding + user_api_key_dict: User authentication information for file access (required for managed files) Returns: BatchFileUsage with total_tokens and request_count @@ -257,6 +259,7 @@ class _PROXY_BatchRateLimiter(CustomLogger): file_content = await litellm.afile_content( file_id=file_id, custom_llm_provider=custom_llm_provider, + user_api_key_dict=user_api_key_dict, ) file_content_as_dict = _get_file_content_as_dictionary( @@ -336,6 +339,7 @@ class _PROXY_BatchRateLimiter(CustomLogger): batch_usage = await self.count_input_file_usage( file_id=input_file_id, custom_llm_provider=custom_llm_provider, + user_api_key_dict=user_api_key_dict, ) verbose_proxy_logger.debug( diff --git a/litellm/proxy/management_endpoints/mcp_management_endpoints.py b/litellm/proxy/management_endpoints/mcp_management_endpoints.py index 90c2f7fdf62..597521ae773 100644 --- a/litellm/proxy/management_endpoints/mcp_management_endpoints.py +++ b/litellm/proxy/management_endpoints/mcp_management_endpoints.py @@ -262,6 +262,67 @@ if MCP_AVAILABLE: ) -> List[LiteLLM_MCPServerTable]: return [_redact_mcp_credentials(server) for server in mcp_servers] + def _is_restricted_virtual_key_request(user_api_key_dict: UserAPIKeyAuth) -> bool: + """Best-effort detection for route-restricted virtual keys. + + We treat a requestor as a "restricted" virtual key if `allowed_routes` + is a non-empty list. This matches the auth gate that blocks routes with + the error: "Virtual key is not allowed to call this route...". + """ + + allowed_routes = getattr(user_api_key_dict, "allowed_routes", None) + return isinstance(allowed_routes, list) and len(allowed_routes) > 0 + + def _sanitize_mcp_server_for_virtual_key( + mcp_server: LiteLLM_MCPServerTable, + ) -> LiteLLM_MCPServerTable: + """Return a minimally sufficient MCP server view for virtual keys. + + Security model: + - Virtual keys should be able to *discover* accessible servers. + - They should NOT receive sensitive configuration details like upstream + URLs, env vars, headers, commands/args, access-group names, or + credentials. + """ + + sanitized = _redact_mcp_credentials(mcp_server) + + # Remove potentially sensitive config + identity fields. + sanitized.url = None + sanitized.static_headers = None + sanitized.env = {} + sanitized.command = None + sanitized.args = [] + sanitized.extra_headers = [] + sanitized.allowed_tools = [] + sanitized.mcp_access_groups = [] + sanitized.teams = [] + + sanitized.authorization_url = None + sanitized.token_url = None + sanitized.registration_url = None + + sanitized.health_check_error = None + sanitized.last_health_check = None + + sanitized.created_by = None + sanitized.updated_by = None + sanitized.created_at = None + sanitized.updated_at = None + + # `mcp_info` is arbitrary metadata; keep only an explicit safe subset. + is_public = False + if isinstance(sanitized.mcp_info, dict): + is_public = bool(sanitized.mcp_info.get("is_public")) + sanitized.mcp_info = {"is_public": True} if is_public else None + + return sanitized + + def _sanitize_mcp_server_list_for_virtual_key( + mcp_servers: Iterable[LiteLLM_MCPServerTable], + ) -> List[LiteLLM_MCPServerTable]: + return [_sanitize_mcp_server_for_virtual_key(server) for server in mcp_servers] + def _inherit_credentials_from_existing_server( payload: NewMCPServerRequest, ) -> NewMCPServerRequest: @@ -504,8 +565,11 @@ if MCP_AVAILABLE: """ user_mcp_management_mode = _get_user_mcp_management_mode() + is_restricted_virtual_key = _is_restricted_virtual_key_request( + user_api_key_dict + ) - if user_mcp_management_mode == "view_all": + if user_mcp_management_mode == "view_all" and not is_restricted_virtual_key: servers = await global_mcp_server_manager.get_all_mcp_servers_unfiltered() redacted_mcp_servers = _redact_mcp_credentials_list(servers) else: @@ -531,6 +595,11 @@ if MCP_AVAILABLE: if server.mcp_info is None: server.mcp_info = {} server.mcp_info["is_public"] = True + + # Virtual keys only get a sanitized discovery view. + if is_restricted_virtual_key: + return _sanitize_mcp_server_list_for_virtual_key(redacted_mcp_servers) + return redacted_mcp_servers @router.get( @@ -625,6 +694,34 @@ if MCP_AVAILABLE: detail={"error": f"MCP Server with id {server_id} not found"}, ) + # Implement authz restriction from requested user + is_admin_view = _user_has_admin_view(user_api_key_dict) + is_restricted_virtual_key = _is_restricted_virtual_key_request( + user_api_key_dict + ) + + if not is_admin_view: + # Perform authz check BEFORE any health check (avoid side-effects for + # unauthorized callers). + mcp_server_records = await get_all_mcp_servers_for_user( + prisma_client, user_api_key_dict + ) + exists = does_mcp_server_exist(mcp_server_records, server_id) + + if not exists: + raise HTTPException( + status_code=status.HTTP_403_FORBIDDEN, + detail={ + "error": ( + f"User does not have permission to view mcp server with id {server_id}. " + "You can only view mcp servers that you have access to." + ) + }, + ) + + # At this point caller is authorized to view the server. + await global_mcp_server_manager.add_server(mcp_server) + # Perform health check on the server using server manager try: health_result = await global_mcp_server_manager.health_check_server( @@ -644,26 +741,10 @@ if MCP_AVAILABLE: mcp_server.last_health_check = datetime.now() mcp_server.health_check_error = str(e) - # Implement authz restriction from requested user - if _user_has_admin_view(user_api_key_dict): - return _redact_mcp_credentials(mcp_server) - - # Perform authz check to filter the mcp servers user has access to - mcp_server_records = await get_all_mcp_servers_for_user( - prisma_client, user_api_key_dict - ) - exists = does_mcp_server_exist(mcp_server_records, server_id) - - if exists: - await global_mcp_server_manager.add_server(mcp_server) - return _redact_mcp_credentials(mcp_server) - else: - raise HTTPException( - status_code=status.HTTP_403_FORBIDDEN, - detail={ - "error": f"User does not have permission to view mcp server with id {server_id}. You can only view mcp servers that you have access to." - }, - ) + redacted = _redact_mcp_credentials(mcp_server) + if is_restricted_virtual_key: + return _sanitize_mcp_server_for_virtual_key(redacted) + return redacted @router.post( "/server", diff --git a/litellm/proxy/management_endpoints/types.py b/litellm/proxy/management_endpoints/types.py index a35fc4a5f3f..295c2ad50b3 100644 --- a/litellm/proxy/management_endpoints/types.py +++ b/litellm/proxy/management_endpoints/types.py @@ -4,7 +4,7 @@ Types for the management endpoints Might include fastapi/proxy requirements.txt related imports """ -from typing import List, Optional, cast +from typing import Any, Dict, List, Optional, cast from fastapi_sso.sso.base import OpenID @@ -56,3 +56,4 @@ def get_litellm_user_role(role_str) -> Optional[LitellmUserRoles]: class CustomOpenID(OpenID): team_ids: List[str] user_role: Optional[LitellmUserRoles] = None + extra_fields: Optional[Dict[str, Any]] = None diff --git a/litellm/proxy/management_endpoints/ui_sso.py b/litellm/proxy/management_endpoints/ui_sso.py index 278f3bdaafd..7274b389a92 100644 --- a/litellm/proxy/management_endpoints/ui_sso.py +++ b/litellm/proxy/management_endpoints/ui_sso.py @@ -190,9 +190,15 @@ def process_sso_jwt_access_token( if access_token_str and result: import jwt - access_token_payload = jwt.decode( - access_token_str, options={"verify_signature": False} - ) + try: + access_token_payload = jwt.decode( + access_token_str, options={"verify_signature": False} + ) + except jwt.exceptions.DecodeError: + verbose_proxy_logger.debug( + "Access token is not a valid JWT (possibly an opaque token), skipping JWT-based extraction" + ) + return # Extract team IDs from access token if sso_jwt_handler is available if sso_jwt_handler: @@ -401,6 +407,8 @@ def generic_response_convertor( generic_user_role_attribute_name = os.getenv("GENERIC_USER_ROLE_ATTRIBUTE", "role") + generic_user_extra_attributes = os.getenv("GENERIC_USER_EXTRA_ATTRIBUTES", None) + verbose_proxy_logger.debug( f" generic_user_id_attribute_name: {generic_user_id_attribute_name}\n generic_user_email_attribute_name: {generic_user_email_attribute_name}" ) @@ -473,6 +481,14 @@ def generic_response_convertor( f"Found valid LitellmUserRoles '{role.value}' from SSO attribute '{generic_user_role_attribute_name}'" ) + # Build extra_fields dict from GENERIC_USER_EXTRA_ATTRIBUTES if specified + extra_fields: Optional[Dict[str, Any]] = None + if generic_user_extra_attributes: + extra_fields = {} + for attr_name in generic_user_extra_attributes.split(","): + attr_name = attr_name.strip() + extra_fields[attr_name] = get_nested_value(response, attr_name) + return CustomOpenID( id=get_nested_value(response, generic_user_id_attribute_name), display_name=get_nested_value( @@ -484,6 +500,7 @@ def generic_response_convertor( provider=get_nested_value(response, generic_provider_attribute_name), team_ids=all_teams, user_role=user_role, + extra_fields=extra_fields, ) diff --git a/litellm/proxy/pass_through_endpoints/llm_passthrough_endpoints.py b/litellm/proxy/pass_through_endpoints/llm_passthrough_endpoints.py index 3dab6ea14f8..81144ad9f31 100644 --- a/litellm/proxy/pass_through_endpoints/llm_passthrough_endpoints.py +++ b/litellm/proxy/pass_through_endpoints/llm_passthrough_endpoints.py @@ -642,10 +642,10 @@ def _extract_model_from_bedrock_endpoint(endpoint: str) -> str: by finding the action in the endpoint and extracting everything between "model" and the action. Args: - endpoint: The endpoint path (e.g., "/model/aws/anthropic/model-name/invoke") + endpoint: The endpoint path (e.g., "/model/aws/anthropic/model-name/invoke" or "v2/model/model-name/invoke") Returns: - The extracted model name (e.g., "aws/anthropic/model-name") + The extracted model name (e.g., "aws/anthropic/model-name" or "model-name") Raises: ValueError: If model cannot be extracted from endpoint @@ -657,7 +657,34 @@ def _extract_model_from_bedrock_endpoint(endpoint: str) -> str: # Format: model/application-inference-profile/{profile-id}/{action} return "/".join(endpoint_parts[1:3]) - # Format: model/{modelId}/{action} + # Format: model/{modelId}/{action} or v2/model/{modelId}/{action} + # Find the index of "model" in the endpoint parts + model_index = None + for idx, part in enumerate(endpoint_parts): + if part == "model": + model_index = idx + break + + # If "model" keyword not found, try to extract model from the endpoint + # by finding the action and taking everything before it + if model_index is None: + # Find the index of the action in the endpoint parts + action_index = None + for idx, part in enumerate(endpoint_parts): + if part in BEDROCK_ENDPOINT_ACTIONS: + action_index = idx + break + + if action_index is not None and action_index > 1: + # Join all parts before the action (excluding empty strings) + model_parts = [p for p in endpoint_parts[1:action_index] if p] + if model_parts: + return "/".join(model_parts) + + raise ValueError( + f"'model' keyword not found and unable to extract model from endpoint. Expected format: /model/{{modelId}}/{{action}}. Got: {endpoint}" + ) + # Find the index of the action in the endpoint parts action_index = None for idx, part in enumerate(endpoint_parts): @@ -665,13 +692,22 @@ def _extract_model_from_bedrock_endpoint(endpoint: str) -> str: action_index = idx break - if action_index is not None and action_index > 1: - # Join all parts between "model" and the action - return "/".join(endpoint_parts[1:action_index]) + if action_index is not None and action_index > model_index + 1: + # Join all parts between "model" and the action (excluding "model" itself) + return "/".join(endpoint_parts[model_index + 1:action_index]) # Fallback to taking everything after "model" if no action found - return "/".join(endpoint_parts[1:]) + model_parts = [p for p in endpoint_parts[model_index + 1:] if p] + if model_parts: + return "/".join(model_parts) + raise ValueError( + f"No model ID found after 'model' keyword. Expected format: /model/{{modelId}}/{{action}}. Got: {endpoint}" + ) + + except ValueError: + # Re-raise ValueError as-is + raise except Exception as e: raise ValueError( f"Model missing from endpoint. Expected format: /model/{{modelId}}/{{action}}. Got: {endpoint}" diff --git a/litellm/proxy/pass_through_endpoints/pass_through_endpoints.py b/litellm/proxy/pass_through_endpoints/pass_through_endpoints.py index 51a7c37717c..a7b60c8b185 100644 --- a/litellm/proxy/pass_through_endpoints/pass_through_endpoints.py +++ b/litellm/proxy/pass_through_endpoints/pass_through_endpoints.py @@ -2201,6 +2201,41 @@ async def initialize_pass_through_endpoints( InitPassThroughEndpointHelpers.remove_endpoint_routes(endpoint_key) +def _get_pass_through_endpoints_from_config() -> List[PassThroughGenericEndpoint]: + """ + Get pass-through endpoints defined in the config file. + These are read-only and cannot be edited via the UI. + Malformed endpoints are logged and skipped; they do not crash the function. + """ + from pydantic import ValidationError + + from litellm.proxy.proxy_server import config_passthrough_endpoints + + if config_passthrough_endpoints is None or len(config_passthrough_endpoints) == 0: + return [] + + returned_endpoints: List[PassThroughGenericEndpoint] = [] + for endpoint in config_passthrough_endpoints: + try: + if isinstance(endpoint, dict): + endpoint_dict = dict(endpoint) + endpoint_dict["is_from_config"] = True + returned_endpoints.append(PassThroughGenericEndpoint(**endpoint_dict)) + elif isinstance(endpoint, PassThroughGenericEndpoint): + # Create a copy with is_from_config=True + endpoint_dict = endpoint.model_dump() + endpoint_dict["is_from_config"] = True + returned_endpoints.append(PassThroughGenericEndpoint(**endpoint_dict)) + except ValidationError as e: + verbose_proxy_logger.warning( + "Skipping malformed pass-through endpoint from config: %s", + e, + exc_info=False, + ) + + return returned_endpoints + + async def _get_pass_through_endpoints_from_db( endpoint_id: Optional[str] = None, user_api_key_dict: Optional[UserAPIKeyAuth] = None, @@ -2223,17 +2258,27 @@ async def _get_pass_through_endpoints_from_db( returned_endpoints: List[PassThroughGenericEndpoint] = [] if endpoint_id is None: - # Return all endpoints + # Return all endpoints from DB, mark as not from config for endpoint in pass_through_endpoint_data: if isinstance(endpoint, dict): - returned_endpoints.append(PassThroughGenericEndpoint(**endpoint)) + endpoint_dict = dict(endpoint) + endpoint_dict["is_from_config"] = False + returned_endpoints.append(PassThroughGenericEndpoint(**endpoint_dict)) elif isinstance(endpoint, PassThroughGenericEndpoint): - returned_endpoints.append(endpoint) + endpoint_dict = endpoint.model_dump() + endpoint_dict["is_from_config"] = False + returned_endpoints.append(PassThroughGenericEndpoint(**endpoint_dict)) else: # Find specific endpoint by ID found_endpoint = _find_endpoint_by_id(pass_through_endpoint_data, endpoint_id) if found_endpoint is not None: - returned_endpoints.append(found_endpoint) + endpoint_dict = ( + found_endpoint.model_dump() + if isinstance(found_endpoint, PassThroughGenericEndpoint) + else dict(found_endpoint) + ) + endpoint_dict["is_from_config"] = False + returned_endpoints.append(PassThroughGenericEndpoint(**endpoint_dict)) return returned_endpoints @@ -2312,10 +2357,25 @@ async def get_pass_through_endpoints( detail={"error": CommonProxyErrors.db_not_connected_error.value}, ) - pass_through_endpoints = await _get_pass_through_endpoints_from_db( + # Get endpoints from DB (editable via UI) + db_endpoints = await _get_pass_through_endpoints_from_db( endpoint_id=endpoint_id, user_api_key_dict=user_api_key_dict ) + # Get endpoints from config file (read-only, not editable via UI) + config_endpoints = _get_pass_through_endpoints_from_config() + + # Merge: config endpoints not in DB + all DB endpoints (DB overrides config for same path) + db_paths = {ep.path for ep in db_endpoints} + config_only_endpoints = [ + ep for ep in config_endpoints if ep.path not in db_paths + ] + if endpoint_id is not None: + # When filtering by endpoint_id, only return if found in DB (config endpoints use generated IDs) + pass_through_endpoints = db_endpoints + else: + pass_through_endpoints = config_only_endpoints + db_endpoints + if team_id is not None: pass_through_endpoints = await _filter_endpoints_by_team_allowed_routes( team_id=team_id, @@ -2392,7 +2452,8 @@ async def update_pass_through_endpoints( ) # Get the update data as dict, excluding None values for partial updates - update_data = data.model_dump(exclude_none=True) + # Exclude is_from_config as it's a response-only field (computed at read time) + update_data = data.model_dump(exclude_none=True, exclude={"is_from_config"}) # Start with existing endpoint data endpoint_dict = found_endpoint.model_dump() @@ -2404,6 +2465,9 @@ async def update_pass_through_endpoints( if "id" not in update_data and found_endpoint.id is not None: endpoint_dict["id"] = found_endpoint.id + # Remove is_from_config before saving - it's a response-only field (computed at read time) + endpoint_dict.pop("is_from_config", None) + # Create updated endpoint object updated_endpoint = PassThroughGenericEndpoint(**endpoint_dict) @@ -2490,7 +2554,8 @@ async def create_pass_through_endpoints( ) ## Auto-generate ID if not provided - data_dict = data.model_dump() + # Exclude is_from_config as it's a response-only field (computed at read time) + data_dict = data.model_dump(exclude={"is_from_config"}) if data_dict.get("id") is None: data_dict["id"] = str(uuid.uuid4()) diff --git a/litellm/proxy/policy_engine/policy_resolve_endpoints.py b/litellm/proxy/policy_engine/policy_resolve_endpoints.py index eb4d3fc5845..947c659abe4 100644 --- a/litellm/proxy/policy_engine/policy_resolve_endpoints.py +++ b/litellm/proxy/policy_engine/policy_resolve_endpoints.py @@ -7,7 +7,7 @@ Policy resolve and attachment impact estimation endpoints. import json -from fastapi import APIRouter, Depends, HTTPException, Query +from litellm.proxy.auth.route_checks import RouteChecks from litellm._logging import verbose_proxy_logger from litellm.constants import MAX_POLICY_ESTIMATE_IMPACT_ROWS diff --git a/litellm/proxy/proxy_config.yaml b/litellm/proxy/proxy_config.yaml index d87ae8b14ca..a094eb84bf3 100644 --- a/litellm/proxy/proxy_config.yaml +++ b/litellm/proxy/proxy_config.yaml @@ -46,6 +46,7 @@ mcp_servers: transport: "http" url: "https://mcp.deepwiki.com/mcp" + # General Settings general_settings: master_key: sk-1234 diff --git a/litellm/proxy/spend_tracking/spend_management_endpoints.py b/litellm/proxy/spend_tracking/spend_management_endpoints.py index 6e49e4244e7..c05eda85c81 100644 --- a/litellm/proxy/spend_tracking/spend_management_endpoints.py +++ b/litellm/proxy/spend_tracking/spend_management_endpoints.py @@ -1672,6 +1672,9 @@ async def ui_view_spend_logs( # noqa: PLR0915 model: Optional[str] = fastapi.Query( default=None, description="Filter logs by model" ), + model_id: Optional[str] = fastapi.Query( + default=None, description="Filter logs by model ID (litellm model deployment id)" + ), key_alias: Optional[str] = fastapi.Query( default=None, description="Filter logs by key alias" ), @@ -1763,6 +1766,9 @@ async def ui_view_spend_logs( # noqa: PLR0915 if model is not None: where_conditions["model"] = model + if model_id is not None: + where_conditions["model_id"] = model_id + # Build metadata filters metadata_filters = [] if key_alias is not None: diff --git a/litellm/proxy/spend_tracking/spend_tracking_utils.py b/litellm/proxy/spend_tracking/spend_tracking_utils.py index cb8b9ec0395..2f392c48e9d 100644 --- a/litellm/proxy/spend_tracking/spend_tracking_utils.py +++ b/litellm/proxy/spend_tracking/spend_tracking_utils.py @@ -641,7 +641,9 @@ def _sanitize_request_body_for_spend_logs_payload( return {k: _sanitize_value(v) for k, v in request_body.items()} -def _convert_to_json_serializable_dict(obj: Any) -> Any: +def _convert_to_json_serializable_dict( + obj: Any, visited: Optional[set] = None, max_depth: int = 20 +) -> Any: """ Convert object to JSON-serializable dict, handling Pydantic models safely. @@ -650,23 +652,55 @@ def _convert_to_json_serializable_dict(obj: Any) -> Any: Args: obj: Object to convert (dict, list, Pydantic model, or primitive) + visited: Set of object IDs to track circular references + max_depth: Maximum recursion depth to prevent infinite recursion Returns: JSON-serializable version of the object """ - if isinstance(obj, BaseModel): - # Use Pydantic's model_dump() instead of pickle - return obj.model_dump() - elif isinstance(obj, dict): - return {k: _convert_to_json_serializable_dict(v) for k, v in obj.items()} - elif isinstance(obj, list): - return [_convert_to_json_serializable_dict(item) for item in obj] - elif hasattr(obj, "__dict__"): - # Handle objects with __dict__ attribute - return _convert_to_json_serializable_dict(obj.__dict__) - else: - # Primitives (str, int, float, bool, None) pass through - return obj + if max_depth <= 0: + # Return a placeholder if max depth is exceeded + return "" + + if visited is None: + visited = set() + + # Get the object's memory address to track visited objects + obj_id = id(obj) + if obj_id in visited: + # Circular reference detected, return placeholder + return "" + + # Only track mutable objects (dict, list, objects with __dict__) + if isinstance(obj, (dict, list)) or hasattr(obj, "__dict__"): + visited.add(obj_id) + + try: + if isinstance(obj, BaseModel): + # Use Pydantic's model_dump() instead of pickle + result = obj.model_dump() + # Recursively process the dumped dict + return _convert_to_json_serializable_dict(result, visited, max_depth - 1) + elif isinstance(obj, dict): + return { + k: _convert_to_json_serializable_dict(v, visited, max_depth - 1) + for k, v in obj.items() + } + elif isinstance(obj, list): + return [ + _convert_to_json_serializable_dict(item, visited, max_depth - 1) + for item in obj + ] + elif hasattr(obj, "__dict__"): + # Handle objects with __dict__ attribute + return _convert_to_json_serializable_dict(obj.__dict__, visited, max_depth - 1) + else: + # Primitives (str, int, float, bool, None) pass through + return obj + finally: + # Remove from visited set when done processing this object + if obj_id in visited: + visited.remove(obj_id) def _get_proxy_server_request_for_spend_logs_payload( diff --git a/litellm/responses/litellm_completion_transformation/transformation.py b/litellm/responses/litellm_completion_transformation/transformation.py index df298f7c448..2bff4e23c78 100644 --- a/litellm/responses/litellm_completion_transformation/transformation.py +++ b/litellm/responses/litellm_completion_transformation/transformation.py @@ -602,18 +602,53 @@ class LiteLLMCompletionResponsesConfig: } return None + @staticmethod + def _get_mapping_or_attr_value(obj: Any, key: str, default: Any = None) -> Any: + """ + Safely read a field from dict-like or attribute-based objects. + """ + if obj is None: + return default + + if isinstance(obj, dict): + return obj.get(key, default) + + getter = getattr(obj, "get", None) + if callable(getter): + try: + return getter(key, default) + except (TypeError, AttributeError): + pass + + return getattr(obj, key, default) + @staticmethod def _create_tool_call_chunk( tool_use_definition: Dict[str, Any], tool_call_id: str, index: int ) -> ChatCompletionToolCallChunk: """Create a ChatCompletionToolCallChunk from tool_use_definition.""" - function_raw = tool_use_definition.get("function") - function: Dict[str, Any] = function_raw if isinstance(function_raw, dict) else {} - tool_use_id_raw = tool_use_definition.get("id") + function_raw = LiteLLMCompletionResponsesConfig._get_mapping_or_attr_value( + tool_use_definition, "function" + ) + function_name_raw = LiteLLMCompletionResponsesConfig._get_mapping_or_attr_value( + function_raw, "name" + ) + function_arguments_raw = LiteLLMCompletionResponsesConfig._get_mapping_or_attr_value( + function_raw, "arguments" + ) + function: Dict[str, Any] = { + "name": function_name_raw or "", + "arguments": function_arguments_raw or "{}", + } + tool_use_id_raw = LiteLLMCompletionResponsesConfig._get_mapping_or_attr_value( + tool_use_definition, "id" + ) tool_use_id: str = ( str(tool_use_id_raw) if tool_use_id_raw is not None else str(tool_call_id) ) - tool_use_type_raw = tool_use_definition.get("type") + tool_use_type_raw = LiteLLMCompletionResponsesConfig._get_mapping_or_attr_value( + tool_use_definition, "type" + ) tool_use_type: str = ( str(tool_use_type_raw) if tool_use_type_raw is not None else "function" ) @@ -627,6 +662,63 @@ class LiteLLMCompletionResponsesConfig: index=index, ) + @staticmethod + def _normalize_tool_use_definition( + tool_use_definition: Any, tool_call_id: str + ) -> Optional[Dict[str, Any]]: + """ + Normalize cached tool_call definitions to a dict-like shape consumed by _create_tool_call_chunk. + """ + if not tool_use_definition: + return None + + if isinstance(tool_use_definition, dict): + normalized_definition: Dict[str, Any] = dict(tool_use_definition) + else: + tool_use_id_raw = LiteLLMCompletionResponsesConfig._get_mapping_or_attr_value( + tool_use_definition, "id" + ) + tool_use_type_raw = LiteLLMCompletionResponsesConfig._get_mapping_or_attr_value( + tool_use_definition, "type" + ) + function_raw = LiteLLMCompletionResponsesConfig._get_mapping_or_attr_value( + tool_use_definition, "function" + ) + + # Object does not expose the expected tool_call fields. + if ( + tool_use_id_raw is None + and tool_use_type_raw is None + and function_raw is None + ): + return None + + normalized_definition = { + "id": tool_use_id_raw, + "type": tool_use_type_raw, + "function": function_raw, + } + + function_raw = normalized_definition.get("function") + if function_raw is not None and not isinstance(function_raw, dict): + function_name_raw = LiteLLMCompletionResponsesConfig._get_mapping_or_attr_value( + function_raw, "name" + ) + function_arguments_raw = LiteLLMCompletionResponsesConfig._get_mapping_or_attr_value( + function_raw, "arguments" + ) + if function_name_raw is not None or function_arguments_raw is not None: + normalized_definition["function"] = { + "name": function_name_raw, + "arguments": function_arguments_raw, + } + + normalized_definition["id"] = normalized_definition.get("id") or tool_call_id + normalized_definition["type"] = ( + normalized_definition.get("type") or "function" + ) + return normalized_definition + @staticmethod def _add_tool_call_to_assistant( assistant_message: Any, tool_call_chunk: ChatCompletionToolCallChunk @@ -740,13 +832,19 @@ class LiteLLMCompletionResponsesConfig: tool_call_id, tools ) ) - - if _tool_use_definition: - if not isinstance(_tool_use_definition, dict): - _tool_use_definition = {} + + normalized_tool_use_definition = ( + LiteLLMCompletionResponsesConfig._normalize_tool_use_definition( + _tool_use_definition, tool_call_id + ) + ) + + if normalized_tool_use_definition: tool_call_chunk = ( LiteLLMCompletionResponsesConfig._create_tool_call_chunk( - _tool_use_definition, tool_call_id, len(tool_calls) + normalized_tool_use_definition, + tool_call_id, + len(tool_calls), ) ) LiteLLMCompletionResponsesConfig._add_tool_call_to_assistant( @@ -1794,6 +1892,12 @@ class LiteLLMCompletionResponsesConfig: ): output_details_dict["text_tokens"] = completion_details.text_tokens + if ( + hasattr(completion_details, "image_tokens") + and completion_details.image_tokens is not None + ): + output_details_dict["image_tokens"] = completion_details.image_tokens + if output_details_dict: response_usage.output_tokens_details = OutputTokensDetails( **output_details_dict diff --git a/litellm/router.py b/litellm/router.py index 0f49cf90510..d9de7e7fc5a 100644 --- a/litellm/router.py +++ b/litellm/router.py @@ -1924,6 +1924,17 @@ class Router: "deployment_model_name": deployment_model_name, } ) + + ## DEPLOYMENT-LEVEL TAGS + deployment_tags = deployment.get("litellm_params", {}).get("tags") + if deployment_tags: + existing_tags = kwargs[metadata_variable_name].get("tags") or [] + merged_tags = list(existing_tags) + for tag in deployment_tags: + if tag not in merged_tags: + merged_tags.append(tag) + kwargs[metadata_variable_name]["tags"] = merged_tags + kwargs["model_info"] = model_info kwargs["timeout"] = self._get_timeout( @@ -5925,9 +5936,20 @@ class Router: deployment.litellm_params.custom_llm_provider + "/" + _model_name ) + # For the shared backend key, strip custom pricing fields so that + # one deployment's pricing overrides don't pollute another + # deployment sharing the same backend model name. + # Each deployment's full pricing is already stored under its + # unique model_id above. + _custom_pricing_fields = CustomPricingLiteLLMParams.model_fields.keys() + _shared_model_info = { + k: v + for k, v in _model_info.items() + if k not in _custom_pricing_fields + } litellm.register_model( model_cost={ - _model_name: _model_info, + _model_name: _shared_model_info, } ) diff --git a/litellm/types/integrations/cloudzero.py b/litellm/types/integrations/cloudzero.py index aeda76aa5f9..e79500e08db 100644 --- a/litellm/types/integrations/cloudzero.py +++ b/litellm/types/integrations/cloudzero.py @@ -1,4 +1,4 @@ -from typing import Any, Dict, Optional +from typing import Any, Dict class CBFRecord(Dict[str, Any]): @@ -9,19 +9,23 @@ class CBFRecord(Dict[str, Any]): (e.g., 'time/usage_start', 'cost/cost'), we use a Dict base class rather than TypedDict to accommodate the special characters in field names. - Expected CBF fields: + Expected CBF fields (per LIT-1907): - time/usage_start: ISO-formatted UTC datetime (Optional[str]) - cost/cost: Billed cost (float) - - resource/id: CloudZero Resource Name (CZRN) (str) + - resource/id: Model name (str) - usage/amount: Numeric value of tokens consumed (int) - usage/units: Description of units, e.g., 'tokens' (str) - - resource/service: Maps to CZRN service-type, e.g., 'litellm' (str) - - resource/account: Maps to CZRN owner-account-id (entity_id) (str) + - resource/service: Model group (str) + - resource/account: api_key_alias|api_key_prefix (str) - resource/region: Maps to CZRN region, e.g., 'cross-region' (str) - - resource/usage_family: Maps to CZRN resource-type, e.g., 'llm-usage' (str) + - resource/usage_family: Provider (str) + - action/operation: Team ID (str) - lineitem/type: Standard usage line item, e.g., 'Usage' (str) - resource/tag:provider: CZRN provider component (str) - resource/tag:model: CZRN cloud-local-id component (model) (str) + - resource/tag:organization_alias: Organization alias if available (Optional[str]) + - resource/tag:project_alias: Project alias if available (Optional[str]) + - resource/tag:user_alias: User alias if available (Optional[str]) - resource/tag:{key}: Various resource tags for dimensions and metrics (Optional[str]) """ pass diff --git a/litellm/utils.py b/litellm/utils.py index 6fdd2d88bca..ed0d6ee930d 100644 --- a/litellm/utils.py +++ b/litellm/utils.py @@ -8243,6 +8243,8 @@ class ProviderConfigManager: return litellm.VolcEngineResponsesAPIConfig() elif litellm.LlmProviders.MANUS == provider: return litellm.ManusResponsesAPIConfig() + elif litellm.LlmProviders.PERPLEXITY == provider: + return litellm.PerplexityResponsesConfig() return None @staticmethod diff --git a/model_prices_and_context_window.json b/model_prices_and_context_window.json index 815d29c7964..35538ab1003 100644 --- a/model_prices_and_context_window.json +++ b/model_prices_and_context_window.json @@ -5848,6 +5848,19 @@ "output_cost_per_token": 7e-07, "supports_tool_choice": true }, + "azure_ai/kimi-k2.5": { + "input_cost_per_token": 6e-07, + "litellm_provider": "azure_ai", + "max_input_tokens": 262144, + "max_output_tokens": 262144, + "max_tokens": 262144, + "mode": "chat", + "output_cost_per_token": 3e-06, + "source": "https://techcommunity.microsoft.com/blog/azure-ai-foundry-blog/kimi-k2-5-now-in-microsoft-foundry/4492321", + "supports_function_calling": true, + "supports_tool_choice": true, + "supports_vision": true + }, "azure_ai/ministral-3b": { "input_cost_per_token": 4e-08, "litellm_provider": "azure_ai", @@ -6091,6 +6104,39 @@ "output_cost_per_token": 2.4e-05, "supports_tool_choice": true }, + "bedrock/ap-northeast-1/moonshotai.kimi-k2-thinking": { + "input_cost_per_token": 7.3e-07, + "litellm_provider": "bedrock", + "max_input_tokens": 262144, + "max_output_tokens": 262144, + "max_tokens": 262144, + "mode": "chat", + "output_cost_per_token": 3.03e-06, + "supports_function_calling": true, + "supports_reasoning": true + }, + "bedrock/moonshotai.kimi-k2-thinking": { + "input_cost_per_token": 7.3e-07, + "litellm_provider": "bedrock", + "max_input_tokens": 262144, + "max_output_tokens": 262144, + "max_tokens": 262144, + "mode": "chat", + "output_cost_per_token": 3.03e-06, + "supports_function_calling": true, + "supports_reasoning": true + }, + "bedrock/moonshotai.kimi-k2.5": { + "input_cost_per_token": 7.3e-07, + "litellm_provider": "bedrock", + "max_input_tokens": 262144, + "max_output_tokens": 262144, + "max_tokens": 262144, + "mode": "chat", + "output_cost_per_token": 3.03e-06, + "supports_function_calling": true, + "supports_reasoning": true + }, "bedrock/ap-south-1/meta.llama3-70b-instruct-v1:0": { "input_cost_per_token": 3.18e-06, "litellm_provider": "bedrock", @@ -6109,6 +6155,17 @@ "mode": "chat", "output_cost_per_token": 7.2e-07 }, + "bedrock/ap-south-1/moonshotai.kimi-k2-thinking": { + "input_cost_per_token": 7.1e-07, + "litellm_provider": "bedrock", + "max_input_tokens": 262144, + "max_output_tokens": 262144, + "max_tokens": 262144, + "mode": "chat", + "output_cost_per_token": 2.94e-06, + "supports_function_calling": true, + "supports_reasoning": true + }, "bedrock/ca-central-1/meta.llama3-70b-instruct-v1:0": { "input_cost_per_token": 3.05e-06, "litellm_provider": "bedrock", @@ -6314,6 +6371,17 @@ "mode": "chat", "output_cost_per_token": 1.01e-06 }, + "bedrock/sa-east-1/moonshotai.kimi-k2-thinking": { + "input_cost_per_token": 7.3e-07, + "litellm_provider": "bedrock", + "max_input_tokens": 262144, + "max_output_tokens": 262144, + "max_tokens": 262144, + "mode": "chat", + "output_cost_per_token": 3.03e-06, + "supports_function_calling": true, + "supports_reasoning": true + }, "bedrock/us-east-1/1-month-commitment/anthropic.claude-instant-v1": { "input_cost_per_second": 0.011, "litellm_provider": "bedrock", @@ -6450,6 +6518,28 @@ "output_cost_per_token": 7e-07, "supports_tool_choice": true }, + "bedrock/us-east-1/moonshotai.kimi-k2-thinking": { + "input_cost_per_token": 6e-07, + "litellm_provider": "bedrock", + "max_input_tokens": 262144, + "max_output_tokens": 262144, + "max_tokens": 262144, + "mode": "chat", + "output_cost_per_token": 2.5e-06, + "supports_function_calling": true, + "supports_reasoning": true + }, + "bedrock/us-east-2/moonshotai.kimi-k2-thinking": { + "input_cost_per_token": 6e-07, + "litellm_provider": "bedrock", + "max_input_tokens": 262144, + "max_output_tokens": 262144, + "max_tokens": 262144, + "mode": "chat", + "output_cost_per_token": 2.5e-06, + "supports_function_calling": true, + "supports_reasoning": true + }, "bedrock/us-gov-east-1/amazon.nova-pro-v1:0": { "input_cost_per_token": 9.6e-07, "litellm_provider": "bedrock", @@ -6856,6 +6946,17 @@ "output_cost_per_token": 7e-07, "supports_tool_choice": true }, + "bedrock/us-west-2/moonshotai.kimi-k2-thinking": { + "input_cost_per_token": 6e-07, + "litellm_provider": "bedrock", + "max_input_tokens": 262144, + "max_output_tokens": 262144, + "max_tokens": 262144, + "mode": "chat", + "output_cost_per_token": 2.5e-06, + "supports_function_calling": true, + "supports_reasoning": true + }, "bedrock/us.anthropic.claude-3-5-haiku-20241022-v1:0": { "cache_creation_input_token_cost": 1e-06, "cache_read_input_token_cost": 8e-08, @@ -25614,6 +25715,66 @@ "supports_function_calling": true, "supports_tool_choice": true }, + "perplexity/preset/pro-search": { + "litellm_provider": "perplexity", + "mode": "responses", + "supports_web_search": true, + "supports_preset": true + }, + "perplexity/openai/gpt-4o": { + "litellm_provider": "perplexity", + "mode": "responses", + "supports_web_search": true, + "supports_reasoning": false + }, + "perplexity/openai/gpt-4o-mini": { + "litellm_provider": "perplexity", + "mode": "responses", + "supports_web_search": true, + "supports_reasoning": false + }, + "perplexity/openai/gpt-5.2": { + "litellm_provider": "perplexity", + "mode": "responses", + "supports_web_search": true, + "supports_reasoning": true + }, + "perplexity/anthropic/claude-3-5-sonnet-20241022": { + "litellm_provider": "perplexity", + "mode": "responses", + "supports_web_search": true, + "supports_reasoning": false + }, + "perplexity/anthropic/claude-3-5-haiku-20241022": { + "litellm_provider": "perplexity", + "mode": "responses", + "supports_web_search": true, + "supports_reasoning": false + }, + "perplexity/google/gemini-2.0-flash-exp": { + "litellm_provider": "perplexity", + "mode": "responses", + "supports_web_search": true, + "supports_reasoning": false + }, + "perplexity/google/gemini-2.0-flash-thinking-exp": { + "litellm_provider": "perplexity", + "mode": "responses", + "supports_web_search": true, + "supports_reasoning": true + }, + "perplexity/xai/grok-2-1212": { + "litellm_provider": "perplexity", + "mode": "responses", + "supports_web_search": true, + "supports_reasoning": false + }, + "perplexity/xai/grok-2-vision-1212": { + "litellm_provider": "perplexity", + "mode": "responses", + "supports_web_search": true, + "supports_reasoning": false + }, "publicai/aisingapore/Qwen-SEA-LION-v4-32B-IT": { "input_cost_per_token": 0.0, "litellm_provider": "publicai", diff --git a/pyproject.toml b/pyproject.toml index f26e49093de..acb8bc2ada3 100644 --- a/pyproject.toml +++ b/pyproject.toml @@ -1,6 +1,6 @@ [tool.poetry] name = "litellm" -version = "1.81.9" +version = "1.81.10" description = "Library to easily interface with LLM API providers" authors = ["BerriAI"] license = "MIT" @@ -61,7 +61,7 @@ boto3 = { version = "1.40.76", optional = true } redisvl = {version = "^0.4.1", optional = true, markers = "python_version >= '3.9' and python_version < '3.14'"} mcp = {version = ">=1.25.0,<2.0.0", optional = true, python = ">=3.10"} a2a-sdk = {version = "^0.3.22", optional = true, python = ">=3.10"} -litellm-proxy-extras = {version = "0.4.33", optional = true} +litellm-proxy-extras = {version = "0.4.34", optional = true} rich = {version = "13.7.1", optional = true} litellm-enterprise = {version = "0.1.31", optional = true} diskcache = {version = "^5.6.1", optional = true} @@ -175,7 +175,7 @@ requires = ["poetry-core", "wheel"] build-backend = "poetry.core.masonry.api" [tool.commitizen] -version = "1.81.9" +version = "1.81.10" version_files = [ "pyproject.toml:^version" ] diff --git a/requirements.txt b/requirements.txt index f680de120c5..bd313b105e5 100644 --- a/requirements.txt +++ b/requirements.txt @@ -32,7 +32,7 @@ async_generator==1.10.0 # for async ollama calls langfuse==2.59.7 # for langfuse self-hosted logging prometheus_client==0.20.0 # for /metrics endpoint on proxy ddtrace==2.19.0 # for advanced DD tracing / profiling -orjson==3.11.2 # fast /embedding responses +orjson==3.11.7 # fast /embedding responses polars==1.31.0 # for data processing apscheduler==3.10.4 # for resetting budget in background fastapi-sso==0.19.0 # admin UI, SSO @@ -55,7 +55,7 @@ sentry_sdk==2.21.0 # for sentry error handling detect-secrets==1.5.0 # Enterprise - secret detection / masking in LLM requests cryptography==44.0.1 tzdata==2025.1 # IANA time zone database -litellm-proxy-extras==0.4.33 # for proxy extras - e.g. prisma migrations +litellm-proxy-extras==0.4.34 # for proxy extras - e.g. prisma migrations llm-sandbox==0.3.31 # for skill execution in sandbox ### LITELLM PACKAGE DEPENDENCIES python-dotenv==1.0.1 # for env diff --git a/tests/batches_tests/test_batch_rate_limits.py b/tests/batches_tests/test_batch_rate_limits.py index 776aba438c2..13241e94d51 100644 --- a/tests/batches_tests/test_batch_rate_limits.py +++ b/tests/batches_tests/test_batch_rate_limits.py @@ -389,3 +389,233 @@ async def test_batch_rate_limit_multiple_requests(): print(f" Error: {exc_info.value.detail}") finally: os.unlink(file_path_2) + + +@pytest.mark.asyncio() +@pytest.mark.skipif( + os.environ.get("OPENAI_API_KEY") is None, + reason="OPENAI_API_KEY not set - skipping integration test" +) +async def test_batch_rate_limiter_with_managed_files(): + """ + Test for GEN-2166: Verify batch rate limiter can read user files when managed files are enabled. + + This test ensures that: + 1. The batch rate limiter passes user_api_key_dict to afile_content() + 2. The managed files hook can verify file ownership correctly + 3. Rate limiting is enforced (not silently bypassed) + 4. No 403 Permission Denied errors occur for files owned by the user + """ + import tempfile + from unittest.mock import AsyncMock, MagicMock, patch + + CUSTOM_LLM_PROVIDER = "openai" + + # Setup: Create internal usage cache and rate limiter + dual_cache = DualCache() + internal_usage_cache = InternalUsageCache(dual_cache=dual_cache) + rate_limiter = _PROXY_MaxParallelRequestsHandler_v3( + internal_usage_cache=internal_usage_cache + ) + + # Setup: Get batch rate limiter + batch_limiter = rate_limiter._get_batch_rate_limiter() + assert batch_limiter is not None, "Batch rate limiter should be available" + + # Setup: Create user API key with TPM = 500, RPM = 10 + test_user_id = "test-user-abc123" + user_api_key_dict = UserAPIKeyAuth( + api_key="test-key-managed-files", + user_id=test_user_id, + tpm_limit=500, + rpm_limit=10, + ) + + print(f"\n=== Testing Batch Rate Limiter with Managed Files ===") + print(f"User ID: {test_user_id}") + + # Create a batch file with ~200 tokens + import json as json_lib + message = "This is a test message for batch rate limiting with managed files. " * 5 + requests = [] + for i in range(1, 4): + request_obj = { + "custom_id": f"request-{i}", + "method": "POST", + "url": "/v1/chat/completions", + "body": { + "model": "gpt-3.5-turbo", + "messages": [{"role": "user", "content": message}] + } + } + requests.append(json_lib.dumps(request_obj)) + + batch_content = "\n".join(requests) + + with tempfile.NamedTemporaryFile(mode='w', suffix='.jsonl', delete=False) as f: + f.write(batch_content) + file_path = f.name + + try: + # Step 1: Upload file to OpenAI (simulating user upload) + print("\n1. Uploading batch input file...") + file_obj = await litellm.acreate_file( + file=open(file_path, "rb"), + purpose="batch", + custom_llm_provider=CUSTOM_LLM_PROVIDER, + ) + print(f" ✓ File uploaded: {file_obj.id}") + await asyncio.sleep(1) # Give API time to process + + # Step 2: Mock managed files hook to simulate file ownership check + # In a real scenario, the managed files hook would check if the user owns the file + # For this test, we'll verify that user_api_key_dict is passed correctly + print("\n2. Testing rate limiter file access with user context...") + + # Track if user_api_key_dict was passed to afile_content + original_afile_content = litellm.afile_content + user_context_passed = {"value": False} + + async def mock_afile_content(*args, **kwargs): + # Check if user_api_key_dict was passed + if "user_api_key_dict" in kwargs and kwargs["user_api_key_dict"] is not None: + user_context_passed["value"] = True + print(f" ✓ user_api_key_dict passed to afile_content") + print(f" User ID: {kwargs['user_api_key_dict'].user_id}") + else: + print(f" ✗ user_api_key_dict NOT passed to afile_content (BUG!)") + + # Call original function + return await original_afile_content(*args, **kwargs) + + # Patch afile_content to track the call + with patch('litellm.afile_content', side_effect=mock_afile_content): + data = { + "model": "gpt-3.5-turbo", + "input_file_id": file_obj.id, + "custom_llm_provider": CUSTOM_LLM_PROVIDER, + } + + # Step 3: Submit batch and verify rate limiting works + print("\n3. Submitting batch with rate limiting...") + result = await batch_limiter.async_pre_call_hook( + user_api_key_dict=user_api_key_dict, + cache=dual_cache, + data=data, + call_type="acreate_batch", + ) + + tokens_used = result.get('_batch_token_count', 0) + requests_count = result.get('_batch_request_count', 0) + print(f" ✓ Batch submitted successfully") + print(f" Tokens counted: {tokens_used}") + print(f" Requests counted: {requests_count}") + print(f" Rate limit usage: {tokens_used}/500 TPM, {requests_count}/10 RPM") + + # Step 4: Verify user context was passed + print("\n4. Verifying fix for GEN-2166...") + assert user_context_passed["value"], ( + "FAILED: user_api_key_dict was not passed to afile_content(). " + "This means the bug GEN-2166 is not fixed!" + ) + print(" ✓ Fix verified: user_api_key_dict is correctly passed") + + # Step 5: Verify rate limiting is actually enforced (not bypassed) + print("\n5. Verifying rate limiting is enforced...") + assert tokens_used > 0, "Token count should be greater than 0" + assert requests_count > 0, "Request count should be greater than 0" + print(" ✓ Rate limiting is active (not silently bypassed)") + + print("\n=== Test Passed: GEN-2166 Fix Verified ===") + print("✓ Batch rate limiter can access user files") + print("✓ User context is correctly passed") + print("✓ Rate limiting is enforced") + print("✓ No silent failures") + + except HTTPException as e: + if e.status_code == 403: + pytest.fail( + f"FAILED: Got 403 Permission Denied error. " + f"This indicates the bug GEN-2166 is not fixed. " + f"Error: {e.detail}" + ) + else: + raise + except Exception as e: + pytest.fail(f"Unexpected error: {str(e)}") + finally: + os.unlink(file_path) + + +@pytest.mark.asyncio() +async def test_batch_rate_limiter_without_user_context(): + """ + Test that verifies the bug scenario from GEN-2166. + + When user_api_key_dict is NOT passed to count_input_file_usage(), + the function should still work for non-managed files, but would fail + for managed files (which is the bug we fixed). + + This test documents the expected behavior with and without user context. + """ + import tempfile + + CUSTOM_LLM_PROVIDER = "openai" + + # Setup + BATCH_LIMITER = _PROXY_BatchRateLimiter( + internal_usage_cache=None, + parallel_request_limiter=None, + ) + + # Create a simple batch file + batch_content = """{"custom_id": "request-1", "method": "POST", "url": "/v1/chat/completions", "body": {"model": "gpt-3.5-turbo", "messages": [{"role": "user", "content": "Hello"}]}}""" + + with tempfile.NamedTemporaryFile(mode='w', suffix='.jsonl', delete=False) as f: + f.write(batch_content) + file_path = f.name + + try: + # Upload file + file_obj = await litellm.acreate_file( + file=open(file_path, "rb"), + purpose="batch", + custom_llm_provider=CUSTOM_LLM_PROVIDER, + ) + await asyncio.sleep(1) + + # Test 1: Without user context (old behavior - would fail with managed files) + print("\n=== Test 1: count_input_file_usage WITHOUT user context ===") + try: + usage_without_context = await BATCH_LIMITER.count_input_file_usage( + file_id=file_obj.id, + custom_llm_provider=CUSTOM_LLM_PROVIDER, + user_api_key_dict=None, # Explicitly passing None + ) + print(f"✓ Works for non-managed files (tokens: {usage_without_context.total_tokens})") + print(" Note: Would fail with 403 for managed files (GEN-2166 bug)") + except Exception as e: + print(f"✗ Failed: {str(e)}") + + # Test 2: With user context (new behavior - works with managed files) + print("\n=== Test 2: count_input_file_usage WITH user context ===") + user_api_key_dict = UserAPIKeyAuth( + api_key="test-key", + user_id="test-user-123", + ) + + usage_with_context = await BATCH_LIMITER.count_input_file_usage( + file_id=file_obj.id, + custom_llm_provider=CUSTOM_LLM_PROVIDER, + user_api_key_dict=user_api_key_dict, # Passing user context + ) + print(f"✓ Works with user context (tokens: {usage_with_context.total_tokens})") + print(" Note: This fixes GEN-2166 for managed files") + + # Verify both return the same results + assert usage_with_context.total_tokens == usage_without_context.total_tokens + assert usage_with_context.request_count == usage_without_context.request_count + print("\n✓ Both methods return identical results for non-managed files") + + finally: + os.unlink(file_path) diff --git a/tests/code_coverage_tests/check_guardrail_apply_decorator.py b/tests/code_coverage_tests/check_guardrail_apply_decorator.py new file mode 100644 index 00000000000..18a86277aa9 --- /dev/null +++ b/tests/code_coverage_tests/check_guardrail_apply_decorator.py @@ -0,0 +1,126 @@ +""" +Test that all guardrail hooks with async def apply_guardrail use @log_guardrail_information decorator. + +This ensures consistent logging and observability across all guardrail implementations. +""" + +import ast +from pathlib import Path +from typing import List, Tuple + + +def find_apply_guardrail_methods(file_path: Path) -> List[Tuple[str, int, bool]]: + """ + Find all apply_guardrail methods and check if they have the decorator. + + Returns: + List of tuples: (class_name, line_number, has_decorator) + """ + with open(file_path, "r") as f: + content = f.read() + + try: + tree = ast.parse(content) + except SyntaxError: + return [] + + results = [] + + for node in ast.walk(tree): + if isinstance(node, ast.ClassDef): + class_name = node.name + + # Check if this class has apply_guardrail method + for item in node.body: + if ( + isinstance(item, ast.AsyncFunctionDef) + and item.name == "apply_guardrail" + ): + # Check if it has the log_guardrail_information decorator + has_decorator = False + for decorator in item.decorator_list: + if ( + isinstance(decorator, ast.Name) + and decorator.id == "log_guardrail_information" + ): + has_decorator = True + break + + results.append((class_name, item.lineno, has_decorator)) + + return results + + +def test_guardrail_apply_decorator(): + """Test that all guardrail hooks with apply_guardrail have the decorator.""" + # Path to the guardrail hooks directory + guardrail_hooks_dir = ( + Path(__file__).parent.parent.parent + / "litellm" + / "proxy" + / "guardrails" + / "guardrail_hooks" + ) + + # Find all Python files in the guardrail hooks directory + python_files = list(guardrail_hooks_dir.rglob("*.py")) + + # Track violations + violations = [] + + for python_file in python_files: + # Skip __init__.py files and test files + if python_file.name == "__init__.py" or python_file.name.startswith("test_"): + continue + + # Skip base files and primitives + if python_file.name in ["base.py", "primitives.py", "patterns.py"]: + continue + + # Skip bedrock_guardrails.py - it implements logging differently via + # add_standard_logging_guardrail_information_to_request_data calls + # in make_bedrock_api_request method instead of using the decorator + if python_file.name == "bedrock_guardrails.py": + continue + + results = find_apply_guardrail_methods(python_file) + + for class_name, line_num, has_decorator in results: + if not has_decorator: + relative_path = python_file.relative_to( + Path(__file__).parent.parent.parent + ) + violations.append((relative_path, class_name, line_num)) + + # Assert no violations found + if violations: + print( + f"\nFound {len(violations)} guardrail hook(s) without @log_guardrail_information decorator:" + ) + print( + "\nAll guardrail hooks must use @log_guardrail_information decorator on their apply_guardrail method." + ) + print( + "This ensures consistent logging and observability across all guardrails.\n" + ) + + for file_path, class_name, line_num in violations: + print(f" - {file_path}:{line_num} ({class_name}.apply_guardrail)") + + print("\nTo fix, add the decorator:") + print( + " from litellm.integrations.custom_guardrail import log_guardrail_information" + ) + print(" ") + print(" @log_guardrail_information") + print(" async def apply_guardrail(self, ...):") + print(" ...") + + raise AssertionError( + f"Found {len(violations)} guardrail hook(s) without @log_guardrail_information decorator" + ) + + +if __name__ == "__main__": + test_guardrail_apply_decorator() + print("✓ All guardrail hooks have @log_guardrail_information decorator") diff --git a/tests/code_coverage_tests/check_licenses.py b/tests/code_coverage_tests/check_licenses.py index dc7cc7c54ac..f49f6807a02 100644 --- a/tests/code_coverage_tests/check_licenses.py +++ b/tests/code_coverage_tests/check_licenses.py @@ -1,7 +1,8 @@ #!/usr/bin/env python3 import sys -import pkg_resources + import requests +from packaging.requirements import Requirement from pathlib import Path import json from typing import Dict, List, Optional, Set, Tuple @@ -213,7 +214,7 @@ class LicenseChecker: try: with open(requirements_file) as f: requirements = [ - pkg_resources.Requirement.parse(line) + Requirement(line.strip()) for line in f if line.strip() and not line.startswith("#") ] @@ -225,8 +226,10 @@ class LicenseChecker: for req in requirements: try: - version = next(iter(req.specs))[1] if req.specs else None - except Exception: + version = ( + next(iter(req.specifier)).version if req.specifier else None + ) + except StopIteration: version = None if not self.check_package(req.name, version): diff --git a/tests/code_coverage_tests/liccheck.ini b/tests/code_coverage_tests/liccheck.ini index dc46f83366a..e6e9d761ad5 100644 --- a/tests/code_coverage_tests/liccheck.ini +++ b/tests/code_coverage_tests/liccheck.ini @@ -111,6 +111,7 @@ ddtrace: >=2.19.0 # Unknown license orjson: >=3.10.12 # Unknown license apscheduler: >=3.10.4 # Unknown license fastapi-sso: >=0.16.0 # Unknown license +filelock: >=3.20.0 # Unlicense (public domain) - https://unlicense.org / https://github.com/tox-dev/filelock pyjwt: >=2.9.0 # Unknown license python-multipart: >=0.0.18 # Unknown license pillow: >=11.0.0 # Unknown license @@ -123,6 +124,7 @@ opentelemetry-exporter-otlp: >=1.25.0 # Unknown license sentry_sdk: >=2.21.0 # Unknown license cryptography: >=43.0.1 # Unknown license tzdata: >=2025.1 # Unknown license +urllib3: >=2.0.0 # MIT license - https://github.com/urllib3/urllib3 python-dotenv: >=1.0.0 # Unknown license tiktoken: >=0.8.0 # Unknown license click: >=8.1.7 # Unknown license diff --git a/tests/code_coverage_tests/recursive_detector.py b/tests/code_coverage_tests/recursive_detector.py index 71e7798b09e..d6bf1941a08 100644 --- a/tests/code_coverage_tests/recursive_detector.py +++ b/tests/code_coverage_tests/recursive_detector.py @@ -42,6 +42,7 @@ IGNORE_FUNCTIONS = [ "_validate_inheritance_chain", # max depth set (default 100) to prevent infinite recursion in policy inheritance validation. "_basic_json_schema_validate", # max depth set. "extract_text_from_a2a_message", # max depth set (default 10) to prevent infinite recursion in A2A message parsing. + "_convert_to_json_serializable_dict", # max depth set (default 20) and circular reference protection to prevent infinite recursion. ] diff --git a/tests/enterprise/litellm_enterprise/proxy/guardrails/test_bedrock_apply_guardrail.py b/tests/enterprise/litellm_enterprise/proxy/guardrails/test_bedrock_apply_guardrail.py index 4bea2255b56..dff444168c2 100644 --- a/tests/enterprise/litellm_enterprise/proxy/guardrails/test_bedrock_apply_guardrail.py +++ b/tests/enterprise/litellm_enterprise/proxy/guardrails/test_bedrock_apply_guardrail.py @@ -78,7 +78,7 @@ async def test_bedrock_apply_guardrail_blocked(): }, ) - # HTTPException must propagate as-is (not wrapped in a generic Exception) + # Test the apply_guardrail method propagates HTTPException (AWS error) to the client with pytest.raises(HTTPException) as exc_info: await guardrail.apply_guardrail( inputs={"texts": ["This is blocked content"]}, @@ -276,7 +276,7 @@ async def test_bedrock_apply_guardrail_filters_request_messages_when_flag_enable }, ) - with pytest.raises(HTTPException) as exc_info: + with pytest.raises(HTTPException, match="policy") as exc_info: await guardrail.apply_guardrail( inputs={"texts": ["blocked"]}, request_data=request_data, @@ -286,9 +286,9 @@ async def test_bedrock_apply_guardrail_filters_request_messages_when_flag_enable assert mock_api.called _, kwargs = mock_api.call_args assert kwargs["messages"] == [request_messages[-1]] - # HTTPException must propagate as-is (not wrapped) + # HTTPException from guardrail is propagated so the client gets the AWS message assert exc_info.value.status_code == 400 - assert "Violated guardrail policy" in str(exc_info.value.detail) + assert "policy" in str(exc_info.value.detail) def test_bedrock_guardrail_filters_latest_user_message_when_enabled(): diff --git a/tests/enterprise/litellm_enterprise/proxy/hooks/test_managed_files.py b/tests/enterprise/litellm_enterprise/proxy/hooks/test_managed_files.py index 946c5ad1729..58efa854e7c 100644 --- a/tests/enterprise/litellm_enterprise/proxy/hooks/test_managed_files.py +++ b/tests/enterprise/litellm_enterprise/proxy/hooks/test_managed_files.py @@ -192,7 +192,7 @@ async def test_async_post_call_success_hook_for_unified_finetuning_job(): "model_id": "gpt-3.5-turbo-0613", } proxy_managed_files = _PROXY_LiteLLMManagedFiles( - DualCache(), prisma_client=MagicMock() + DualCache(), prisma_client=AsyncMock() ) data = { "user_api_key_dict": {"parent_otel_span": MagicMock()}, @@ -373,6 +373,91 @@ async def test_output_file_id_for_batch_retrieve(): assert not cast(LiteLLMBatch, response).output_file_id.startswith("file-") +@pytest.mark.asyncio +async def test_error_file_id_for_failed_batch(): + """ + Test that the error_file_id is properly managed when a batch fails + """ + from typing import cast + + from openai.types.batch import BatchRequestCounts + + from litellm.proxy._types import UserAPIKeyAuth + from litellm.types.llms.openai import OpenAIFileObject + from litellm.types.utils import LiteLLMBatch + + batch = LiteLLMBatch( + id="bGl0ZWxsbV9wcm94eTttb2RlbF9pZDoxMjM0NTY3OTtsbG1fYmF0Y2hfaWQ6YmF0Y2hfYWJjMTIz", + completion_window="24h", + created_at=1714508499, + endpoint="/v1/chat/completions", + input_file_id="file-abc123", + object="batch", + status="failed", + cancelled_at=None, + cancelling_at=None, + completed_at=None, + error_file_id="error-abc123", + errors=None, + expired_at=None, + expires_at=1714536634, + failed_at=None, + finalizing_at=None, + in_progress_at=None, + metadata=None, + output_file_id=None, + request_counts=BatchRequestCounts(completed=0, failed=0, total=0), + usage=None, + ) + + batch._hidden_params = { + "litellm_call_id": "test-call-id", + "api_base": "https://api.openai.com", + "model_id": "test-model-id", + "model_name": "gpt-4o", + "response_cost": 0.0, + "additional_headers": {}, + "litellm_model_name": "gpt-4o", + "unified_batch_id": "litellm_proxy;model_id:test-model-id;llm_batch_id:batch_abc123", + } + + proxy_managed_files = _PROXY_LiteLLMManagedFiles( + DualCache(), prisma_client=AsyncMock() + ) + + # Create a proper OpenAIFileObject for the error file + error_file_object = OpenAIFileObject( + id="error-abc123", + object="file", + bytes=1234, + created_at=1714508500, + filename="error.jsonl", + purpose="batch_output", + status="processed", + ) + + # Mock the afile_retrieve to simulate retrieving error file metadata + with patch("litellm.afile_retrieve", new_callable=AsyncMock) as mock_retrieve: + mock_retrieve.return_value = error_file_object + + user_api_key_dict = UserAPIKeyAuth( + user_id="test-user-123", + parent_otel_span=MagicMock() + ) + + response = await proxy_managed_files.async_post_call_success_hook( + data={}, + user_api_key_dict=user_api_key_dict, + response=batch, + ) + + # Verify that error_file_id was transformed to a managed file ID + assert cast(LiteLLMBatch, response).error_file_id is not None + assert not cast(LiteLLMBatch, response).error_file_id.startswith("error-") + # Verify it's a base64 encoded managed file ID + assert _is_base64_encoded_unified_file_id(cast(LiteLLMBatch, response).error_file_id) + + @pytest.mark.asyncio async def test_async_post_call_success_hook_twice_assert_no_unique_violation(): import asyncio @@ -849,7 +934,7 @@ async def test_check_file_ids_access_with_unified_file_ids(): Test that check_file_ids_access validates user access to managed file IDs. """ from litellm.proxy._types import UserAPIKeyAuth - + # Create a unified file ID unified_file_id = "bGl0ZWxsbV9wcm94eTphcHBsaWNhdGlvbi9wZGY7dW5pZmllZF9pZCw2YzBiNTg5MC04OTE0LTQ4ZTAtYjhmNC0wYWU1ZWQzYzE0YTU7dGFyZ2V0X21vZGVsX25hbWVzLGdwdC00bztsbG1fb3V0cHV0X2ZpbGVfaWQsZmlsZS1FQ0JQVzdNTDlnN1hIZHdHZ1VQWmFNO2xsbV9vdXRwdXRfZmlsZV9tb2RlbF9pZCxlMjY0NTNmOWU3NmU3OTkzNjgwZDAwNjhkOThjMWY0Y2MyMDViYmFkMDk2N2EzM2M2NjQ4OTM1NjhjYTc0M2My" regular_file_id = "file-abc123" diff --git a/tests/image_gen_tests/test_image_generation.py b/tests/image_gen_tests/test_image_generation.py index 0567f60ecfc..3b4abeeb82f 100644 --- a/tests/image_gen_tests/test_image_generation.py +++ b/tests/image_gen_tests/test_image_generation.py @@ -5,7 +5,7 @@ import logging import os import sys import traceback -from unittest.mock import AsyncMock, patch +from unittest.mock import AsyncMock, MagicMock, patch sys.path.insert( @@ -25,7 +25,7 @@ import pytest import litellm import json import tempfile -from base_image_generation_test import BaseImageGenTest +from base_image_generation_test import BaseImageGenTest, TestCustomLogger import logging from litellm._logging import verbose_logger @@ -182,6 +182,86 @@ class TestAimlImageGeneration(BaseImageGenTest): def get_base_image_generation_call_args(self) -> dict: return {"model": "aiml/flux-pro/v1.1"} + @pytest.mark.asyncio(scope="module") + @pytest.mark.flaky(retries=0) + async def test_basic_image_generation(self): + """Test basic image generation""" + from unittest.mock import AsyncMock, patch + + mock_aiml_response = { + "created": 1703658209, + "data": [{"url": "https://example.com/generated_image.png"}], + } + mock_response = MagicMock() + mock_response.status_code = 200 + mock_response.json.return_value = mock_aiml_response + mock_response.text = json.dumps(mock_aiml_response) + mock_response.headers = {} + + with patch( + "litellm.llms.custom_httpx.http_handler.AsyncHTTPHandler.post", + new_callable=AsyncMock, + ) as mock_async_post, patch( + "litellm.llms.custom_httpx.http_handler.HTTPHandler.post", + ) as mock_sync_post: + mock_async_post.return_value = mock_response + mock_sync_post.return_value = mock_response + + try: + litellm._turn_on_debug() + custom_logger = TestCustomLogger() + litellm.logging_callback_manager._reset_all_callbacks() + litellm.callbacks = [custom_logger] + base_image_generation_call_args = self.get_base_image_generation_call_args() + litellm.set_verbose = True + # Pass dummy api_key so validate_environment passes; HTTP is mocked + response = await litellm.aimage_generation( + **base_image_generation_call_args, + prompt="A image of a otter", + api_key="test-key-mocked-no-credits-needed", + ) + print("FAL AI RESPONSE: ", response) + + await asyncio.sleep(1) + + # assert response._hidden_params["response_cost"] is not None + # assert response._hidden_params["response_cost"] > 0 + # print("response_cost", response._hidden_params["response_cost"]) + + logged_standard_logging_payload = custom_logger.standard_logging_payload + print("logged_standard_logging_payload", logged_standard_logging_payload) + assert logged_standard_logging_payload is not None + assert logged_standard_logging_payload["response_cost"] is not None + assert logged_standard_logging_payload["response_cost"] > 0 + import openai + from openai.types.images_response import ImagesResponse + + # print openai version + print("openai version=", openai.__version__) + + response_dict = dict(response) + if "usage" in response_dict: + response_dict["usage"] = dict(response_dict["usage"]) + print("response usage=", response_dict.get("usage")) + + assert response.data is not None # type guard for iteration (base fails here if None) + for d in response.data: + assert isinstance(d, Image) + print("data in response.data", d) + assert d.b64_json is not None or d.url is not None + except litellm.RateLimitError as e: + pass + except litellm.ContentPolicyViolationError: + pass # Azure randomly raises these errors - skip when they occur + except litellm.InternalServerError: + pass + except Exception as e: + if "Your task failed as a result of our safety system." in str(e): + pass + else: + pytest.fail(f"An exception occurred - {str(e)}") + + class TestGoogleImageGen(BaseImageGenTest): def get_base_image_generation_call_args(self) -> dict: return {"model": "gemini/imagen-4.0-generate-001"} diff --git a/tests/llm_translation/test_model_cost_map_resilience.py b/tests/llm_translation/test_model_cost_map_resilience.py new file mode 100644 index 00000000000..61e375eabeb --- /dev/null +++ b/tests/llm_translation/test_model_cost_map_resilience.py @@ -0,0 +1,291 @@ +""" +Tests for model cost map resilience. + +Simulates: +- A bad (invalid JSON) model cost map upstream +- A bad (empty/missing) backup model cost map +- Verifies litellm.completion() still works even with a broken cost map +- Verifies litellm.get_model_info() raises the expected error for unmapped models +- Verifies the integrity validation helper catches corrupted maps +""" + +import json +import os +import sys +from unittest.mock import MagicMock, patch + +import pytest + +sys.path.insert( + 0, os.path.abspath(os.path.join(os.path.dirname(__file__), "../..")) +) + +import litellm +from litellm.litellm_core_utils.get_model_cost_map import ( + GetModelCostMap, + get_model_cost_map, +) + + +class TestCheckIsValidDict: + """Unit tests for _check_is_valid_dict.""" + + def test_should_reject_non_dict(self): + """Non-dict should fail.""" + assert GetModelCostMap._check_is_valid_dict("not a dict") is False + + def test_should_reject_empty_dict(self): + """Empty dict should fail.""" + assert GetModelCostMap._check_is_valid_dict({}) is False + + def test_should_reject_list(self): + """List should fail.""" + assert GetModelCostMap._check_is_valid_dict([1, 2, 3]) is False + + def test_should_reject_none(self): + """None should fail.""" + assert GetModelCostMap._check_is_valid_dict(None) is False + + def test_should_accept_non_empty_dict(self): + """Non-empty dict should pass.""" + assert GetModelCostMap._check_is_valid_dict({"model": {}}) is True + + +class TestCheckModelCountNotReduced: + """Unit tests for _check_model_count_not_reduced.""" + + def test_should_reject_too_few_models(self): + """Fetched map with fewer models than min_model_count should fail.""" + small_map = {f"model-{i}": {} for i in range(5)} + assert ( + GetModelCostMap._check_model_count_not_reduced( + fetched_map=small_map, backup_model_count=0, min_model_count=10 + ) + is False + ) + + def test_should_reject_significant_shrinkage(self): + """Fetched map that shrunk >50% vs backup should fail.""" + fetched = {f"model-{i}": {} for i in range(40)} # 40% of 100 + assert ( + GetModelCostMap._check_model_count_not_reduced( + fetched_map=fetched, backup_model_count=100, min_model_count=10 + ) + is False + ) + + def test_should_accept_when_above_threshold(self): + """Fetched map at 60% of backup (above 50% threshold) should pass.""" + fetched = {f"model-{i}": {} for i in range(60)} + assert ( + GetModelCostMap._check_model_count_not_reduced( + fetched_map=fetched, backup_model_count=100, min_model_count=10 + ) + is True + ) + + def test_should_accept_growth(self): + """Fetched map larger than backup should pass.""" + fetched = {f"model-{i}": {} for i in range(120)} + assert ( + GetModelCostMap._check_model_count_not_reduced( + fetched_map=fetched, backup_model_count=100, min_model_count=10 + ) + is True + ) + + def test_should_accept_with_empty_backup(self): + """When backup is empty, only min_model_count matters.""" + fetched = {f"model-{i}": {} for i in range(15)} + assert ( + GetModelCostMap._check_model_count_not_reduced( + fetched_map=fetched, backup_model_count=0, min_model_count=10 + ) + is True + ) + + +class TestValidateModelCostMap: + """Unit tests for validate_model_cost_map (combines both checks).""" + + def test_should_reject_non_dict(self): + """Non-dict should fail at check 1.""" + assert GetModelCostMap.validate_model_cost_map(fetched_map="not a dict", backup_model_count=0) is False + + def test_should_reject_empty_map(self): + """Empty dict should fail at check 1.""" + assert GetModelCostMap.validate_model_cost_map(fetched_map={}, backup_model_count=0) is False + + def test_should_reject_significant_shrinkage(self): + """Should fail at check 2 (shrinkage).""" + fetched = {f"model-{i}": {} for i in range(40)} + assert ( + GetModelCostMap.validate_model_cost_map( + fetched_map=fetched, backup_model_count=100, min_model_count=10 + ) + is False + ) + + def test_should_accept_valid_map(self): + """Should pass both checks.""" + fetched = {f"model-{i}": {} for i in range(120)} + assert ( + GetModelCostMap.validate_model_cost_map( + fetched_map=fetched, backup_model_count=100, min_model_count=10 + ) + is True + ) + + def test_should_accept_equal_size_map(self): + """Equal size should pass both checks.""" + fetched = {f"model-{i}": {} for i in range(100)} + assert ( + GetModelCostMap.validate_model_cost_map( + fetched_map=fetched, backup_model_count=100, min_model_count=10 + ) + is True + ) + + +class TestGetModelCostMapFallback: + """Tests for get_model_cost_map fallback behavior with bad upstream.""" + + def test_should_fallback_to_backup_on_invalid_json(self): + """When upstream returns invalid JSON, should fall back to local backup.""" + mock_response = MagicMock() + mock_response.raise_for_status = MagicMock() + mock_response.json.side_effect = json.JSONDecodeError("bad json", "", 0) + + with patch("httpx.get", return_value=mock_response): + result = get_model_cost_map("https://fake-url.com/model_prices.json") + + # Should have fallen back to backup — backup always has models + assert isinstance(result, dict) + assert len(result) > 0 + + def test_should_fallback_to_backup_on_network_error(self): + """When upstream is unreachable, should fall back to local backup.""" + with patch("httpx.get", side_effect=Exception("Connection refused")): + result = get_model_cost_map("https://fake-url.com/model_prices.json") + + assert isinstance(result, dict) + assert len(result) > 0 + + def test_should_fallback_when_fetched_map_is_empty(self): + """When upstream returns valid JSON but empty dict, should fall back.""" + mock_response = MagicMock() + mock_response.raise_for_status = MagicMock() + mock_response.json.return_value = {} # empty map + + with patch("httpx.get", return_value=mock_response): + result = get_model_cost_map("https://fake-url.com/model_prices.json") + + # Should have fallen back to backup since empty map fails validation + assert isinstance(result, dict) + assert len(result) > 0 + + def test_should_fallback_when_fetched_map_shrinks_dramatically(self): + """When upstream returns far fewer models than backup, should fall back.""" + tiny_map = {f"model-{i}": {"litellm_provider": "test"} for i in range(11)} + mock_response = MagicMock() + mock_response.raise_for_status = MagicMock() + mock_response.json.return_value = tiny_map + + with patch("httpx.get", return_value=mock_response): + result = get_model_cost_map("https://fake-url.com/model_prices.json") + + # Backup has thousands of models; 11 is a massive shrinkage → fallback + assert len(result) > 11 + + def test_should_use_local_map_when_env_var_set(self): + """LITELLM_LOCAL_MODEL_COST_MAP=True should skip remote fetch entirely.""" + with patch.dict(os.environ, {"LITELLM_LOCAL_MODEL_COST_MAP": "True"}): + with patch("httpx.get") as mock_get: + result = get_model_cost_map( + "https://fake-url.com/model_prices.json" + ) + mock_get.assert_not_called() + + assert isinstance(result, dict) + assert len(result) > 0 + + +class TestBackupModelCostMapExists: + """Validates the local backup file is always present and valid.""" + + def test_should_have_backup_file(self): + """The backup model cost map must exist and be loadable.""" + backup = GetModelCostMap.load_local_model_cost_map() + assert isinstance(backup, dict) + assert len(backup) > 0, "Backup model cost map is empty" + + def test_should_have_minimum_models_in_backup(self): + """The backup must contain a reasonable number of models.""" + backup = GetModelCostMap.load_local_model_cost_map() + assert len(backup) > 100, ( + f"Backup has only {len(backup)} models, expected > 100" + ) + + +class TestBadHostedModelCostMap: + """ + Simulates the hosted model cost map being bad (invalid JSON / corrupted). + + When the hosted map is bad, get_model_cost_map() falls back to the local + backup. These tests verify that after fallback: + - get_model_info() still works for models in the backup + - litellm.completion() still works + """ + + def test_should_model_info_pass_after_bad_hosted_map(self): + """ + If the hosted map is bad, get_model_cost_map falls back to the local + backup. get_model_info should still work for models in the backup. + """ + mock_response = MagicMock() + mock_response.raise_for_status = MagicMock() + mock_response.json.side_effect = json.JSONDecodeError("bad json", "", 0) + + with patch("httpx.get", return_value=mock_response): + fallback_map = get_model_cost_map("https://fake-url.com/bad.json") + + original = litellm.model_cost + litellm.model_cost = fallback_map + try: + # gpt-4o is in every backup — should work fine + info = litellm.get_model_info("gpt-4o") + assert info is not None + assert info["input_cost_per_token"] > 0 + finally: + litellm.model_cost = original + + def test_should_completion_pass_after_bad_hosted_map(self): + """ + If the hosted map is bad, litellm.completion() should still work. + + Uses litellm's built-in mock_response param so the real completion + path is exercised (routing, cost calculator, logging) without + needing API credentials. + """ + # Simulate bad hosted map → fallback to backup + mock_http = MagicMock() + mock_http.raise_for_status = MagicMock() + mock_http.json.side_effect = json.JSONDecodeError("bad json", "", 0) + + with patch("httpx.get", return_value=mock_http): + fallback_map = get_model_cost_map("https://fake-url.com/bad.json") + + original = litellm.model_cost + litellm.model_cost = fallback_map + try: + # mock_response goes through the real completion path — + # routing, cost calculator, logging — but skips the HTTP call + response = litellm.completion( + model="gpt-4o-mini", + messages=[{"role": "user", "content": "say hi"}], + mock_response="hello from mock", + ) + assert response is not None + assert response.choices[0].message.content == "hello from mock" + finally: + litellm.model_cost = original diff --git a/tests/local_testing/test_amazing_vertex_completion.py b/tests/local_testing/test_amazing_vertex_completion.py index e1c24ea46fe..0078483c734 100644 --- a/tests/local_testing/test_amazing_vertex_completion.py +++ b/tests/local_testing/test_amazing_vertex_completion.py @@ -3714,6 +3714,60 @@ def test_vertex_schema_test(): print(response) +def test_gemini_nullable_object_tool_schema_httpx(): + """ + Ensure nullable object tool params preserve nested properties in Vertex schema conversion. + """ + load_vertex_ai_credentials() + litellm._turn_on_debug() + + + tools = [{ + "type": "function", + "strict": True, + "function": { + "name": "create_support_ticket", + "description": "Create a paid user support ticket", + "parameters": { + "type": "object", + "additionalProperties": False, + "required": ["ticket_id", "customer_context"], + "properties": { + "ticket_id": { + "type": "string", + "description": "Unique identifier for the support ticket" + }, + "customer_context": { + "type": ["object", "null"], + "description": "Context about the paid customer, if available", + "additionalProperties": False, + "required": ["user_id", "plan"], + "properties": { + "user_id": { + "type": "string", + "description": "Internal user identifier" + }, + "plan": { + "type": "string", + "description": "Subscription plan name (e.g. pro, enterprise)" + } + } + } + } + } + } + }] + + response = litellm.completion( + model="vertex_ai/gemini-2.5-flash", + messages=[{"role": "user", "content": "call the tool"}], + tools=tools, + tool_choice="required", + ) + + print(response) + + def test_vertex_ai_response_id(): """Test that litellm preserves the response ID from Vertex AI's API for non-streaming responses""" from litellm.llms.custom_httpx.http_handler import HTTPHandler diff --git a/tests/local_testing/test_get_model_info.py b/tests/local_testing/test_get_model_info.py index b84fc22af01..d46a087eb73 100644 --- a/tests/local_testing/test_get_model_info.py +++ b/tests/local_testing/test_get_model_info.py @@ -311,7 +311,15 @@ def test_get_model_info_bedrock_models(): for commitment in potential_commitments: k = k.replace(f"{commitment}/", "") base_model = BedrockModelInfo.get_base_model(k) - base_model_info = litellm.model_cost[base_model] + # get_base_model() returns model id without "bedrock/" prefix; cost map keys use "bedrock/" + base_model_key = ( + base_model + if base_model in litellm.model_cost + else f"bedrock/{base_model}" + ) + if base_model_key not in litellm.model_cost: + continue + base_model_info = litellm.model_cost[base_model_key] for base_model_key, base_model_value in base_model_info.items(): if "invoke/" in k: continue diff --git a/tests/mcp_tests/test_oauth2_mcp_config.yaml b/tests/mcp_tests/test_oauth2_mcp_config.yaml new file mode 100644 index 00000000000..c2704c5fe71 --- /dev/null +++ b/tests/mcp_tests/test_oauth2_mcp_config.yaml @@ -0,0 +1,14 @@ +model_list: + - model_name: fake-model + litellm_params: + model: openai/fake + api_key: fake-key + +mcp_servers: + test_oauth2_server: + url: "http://localhost:8765/mcp" + transport: "http" + auth_type: "oauth2" + client_id: "test-client" + client_secret: "test-secret" + token_url: "http://localhost:8765/oauth/token" diff --git a/tests/mcp_tests/test_openapi_spec_path_url.py b/tests/mcp_tests/test_openapi_spec_path_url.py new file mode 100644 index 00000000000..03e9db94967 --- /dev/null +++ b/tests/mcp_tests/test_openapi_spec_path_url.py @@ -0,0 +1,92 @@ +from __future__ import annotations + +from typing import Any, Dict + +import httpx +import pytest + +from litellm.proxy._experimental.mcp_server import openapi_to_mcp_generator as gen + + +class _FakeAsyncHTTPHandler: + """ + Minimal stand-in for the object returned by get_async_httpx_client(). + openapi_to_mcp_generator.load_openapi_spec_async() calls: + + client = get_async_httpx_client(...) + r = await client.get(url, timeout=30.0) + + So we must implement async get(). + """ + + def __init__(self, response: httpx.Response, expected_url: str): + self._response = response + self._expected_url = expected_url + self.calls = 0 + + async def get(self, request_url: str, timeout: float = 30.0): + self.calls += 1 + assert request_url == self._expected_url + assert timeout == 30.0 + return self._response + + +def test_load_openapi_spec_supports_http_url(monkeypatch: pytest.MonkeyPatch) -> None: + url = "http://example.local/openapi.json" + expected: Dict[str, Any] = { + "openapi": "3.0.0", + "info": {"title": "Test API", "version": "1.0.0"}, + "paths": {}, + } + + # httpx.Response must include a Request for raise_for_status() to work. + req = httpx.Request("GET", url) + resp = httpx.Response(status_code=200, json=expected, request=req) + + calls = {"get_async_httpx_client": 0} + handler_holder: Dict[str, Any] = {} + + def fake_get_async_httpx_client(*args, **kwargs): + calls["get_async_httpx_client"] += 1 + h = _FakeAsyncHTTPHandler(resp, expected_url=url) + handler_holder["handler"] = h + return h + + # Ensure shared/custom client path is used + monkeypatch.setattr(gen, "get_async_httpx_client", fake_get_async_httpx_client) + + # Fail loudly if someone reintroduces direct httpx.get() + def boom(*args, **kwargs): + raise AssertionError("Direct httpx.get() must not be used for URL spec loading") + + monkeypatch.setattr(httpx, "get", boom) + + spec = gen.load_openapi_spec(url) + + assert spec == expected + assert calls["get_async_httpx_client"] == 1 + assert handler_holder["handler"].calls == 1 + + +def test_load_openapi_spec_supports_local_file_path(tmp_path, monkeypatch: pytest.MonkeyPatch) -> None: + expected: Dict[str, Any] = { + "openapi": "3.0.0", + "info": {"title": "Local API", "version": "1.0.0"}, + "paths": {}, + } + + p = tmp_path / "openapi.json" + p.write_text( + '{"openapi":"3.0.0","info":{"title":"Local API","version":"1.0.0"},"paths":{}}', + encoding="utf-8", + ) + + # For local files, shared client must NOT be used. + def boom_client(*args, **kwargs): + raise AssertionError("get_async_httpx_client() must not be called for local file paths") + + monkeypatch.setattr(gen, "get_async_httpx_client", boom_client) + + spec = gen.load_openapi_spec(str(p)) + assert spec == expected + diff --git a/tests/test_litellm/integrations/test_opentelemetry.py b/tests/test_litellm/integrations/test_opentelemetry.py index 95fa6ed8f60..3da9d9857c9 100644 --- a/tests/test_litellm/integrations/test_opentelemetry.py +++ b/tests/test_litellm/integrations/test_opentelemetry.py @@ -274,6 +274,26 @@ class TestOpenTelemetry(unittest.TestCase): self.assertEqual(config.deployment_environment, "production") self.assertEqual(config.model_id, "custom-service") + @patch.dict(os.environ, {}, clear=True) + def test_open_telemetry_config_auto_infer_otlp_http_when_endpoint_set(self): + """When endpoint is set but exporter is default 'console', auto-infer 'otlp_http'. + + This fixes an issue where UI-configured OTEL settings would default to console + output instead of sending traces to the configured endpoint. + See: https://github.com/BerriAI/litellm/issues/XXXX + """ + # When endpoint is specified without explicit exporter, should auto-infer otlp_http + config = OpenTelemetryConfig(endpoint="https://otel-collector.example.com:443") + self.assertEqual(config.exporter, "otlp_http") + + # When exporter is explicitly set to something other than console, should not override + config_grpc = OpenTelemetryConfig(exporter="grpc", endpoint="https://otel-collector.example.com:443") + self.assertEqual(config_grpc.exporter, "grpc") + + # When no endpoint is set, should keep console as default + config_no_endpoint = OpenTelemetryConfig() + self.assertEqual(config_no_endpoint.exporter, "console") + def wait_for_spans(self, exporter: InMemorySpanExporter, prefix: str): """Poll until we see at least one span with an attribute key starting with `prefix`.""" deadline = time.time() + self.POLL_TIMEOUT diff --git a/tests/test_litellm/litellm_core_utils/prompt_templates/test_litellm_core_utils_prompt_templates_common_utils.py b/tests/test_litellm/litellm_core_utils/prompt_templates/test_litellm_core_utils_prompt_templates_common_utils.py index 980693aa73a..f566f91841d 100644 --- a/tests/test_litellm/litellm_core_utils/prompt_templates/test_litellm_core_utils_prompt_templates_common_utils.py +++ b/tests/test_litellm/litellm_core_utils/prompt_templates/test_litellm_core_utils_prompt_templates_common_utils.py @@ -12,6 +12,7 @@ sys.path.insert( from litellm.litellm_core_utils.prompt_templates.common_utils import ( get_format_from_file_id, handle_any_messages_to_chat_completion_str_messages_conversion, + split_concatenated_json_objects, update_messages_with_model_file_ids, ) @@ -143,3 +144,58 @@ def test_convert_prefix_message_to_non_prefix_messages(): }, {"role": "assistant", "content": "value"}, ] + + +# ── split_concatenated_json_objects tests ── + + +def test_split_concatenated_json_single_object(): + """A single valid JSON object is returned as a one-element list.""" + result = split_concatenated_json_objects('{"location": "Boston"}') + assert result == [{"location": "Boston"}] + + +def test_split_concatenated_json_multiple_objects(): + """ + Multiple JSON objects concatenated without separators are split correctly. + This is the exact pattern from issue #20543 where Bedrock Claude Sonnet 4.5 + returns concatenated JSON in a single tool call arguments string. + """ + raw = ( + '{"command": ["curl", "-i", "http://localhost:9009"]}' + '{"command": ["curl", "-i", "http://localhost:9009/robots.txt"]}' + '{"command": ["curl", "-i", "http://localhost:9009/sitemap.xml"]}' + ) + result = split_concatenated_json_objects(raw) + assert len(result) == 3 + assert result[0] == {"command": ["curl", "-i", "http://localhost:9009"]} + assert result[1] == {"command": ["curl", "-i", "http://localhost:9009/robots.txt"]} + assert result[2] == {"command": ["curl", "-i", "http://localhost:9009/sitemap.xml"]} + + +def test_split_concatenated_json_with_whitespace(): + """Objects separated by whitespace are handled correctly.""" + raw = '{"a": 1} {"b": 2}\n{"c": 3}' + result = split_concatenated_json_objects(raw) + assert len(result) == 3 + assert result[0] == {"a": 1} + assert result[1] == {"b": 2} + assert result[2] == {"c": 3} + + +def test_split_concatenated_json_empty_string(): + """Empty or whitespace-only strings return an empty list.""" + assert split_concatenated_json_objects("") == [] + assert split_concatenated_json_objects(" ") == [] + + +def test_split_concatenated_json_non_dict_value(): + """Non-dict JSON values (e.g. arrays, strings) are replaced with {}.""" + result = split_concatenated_json_objects('[1, 2, 3]') + assert result == [{}] + + +def test_split_concatenated_json_invalid_raises(): + """Completely invalid JSON raises JSONDecodeError.""" + with pytest.raises(json.JSONDecodeError): + split_concatenated_json_objects("not json at all") diff --git a/tests/test_litellm/litellm_core_utils/prompt_templates/test_litellm_core_utils_prompt_templates_factory.py b/tests/test_litellm/litellm_core_utils/prompt_templates/test_litellm_core_utils_prompt_templates_factory.py index e87233a52a3..707b5bdc777 100644 --- a/tests/test_litellm/litellm_core_utils/prompt_templates/test_litellm_core_utils_prompt_templates_factory.py +++ b/tests/test_litellm/litellm_core_utils/prompt_templates/test_litellm_core_utils_prompt_templates_factory.py @@ -8,6 +8,7 @@ from litellm.litellm_core_utils.prompt_templates.factory import ( BAD_MESSAGE_ERROR_STR, BedrockConverseMessagesProcessor, BedrockImageProcessor, + _convert_to_bedrock_tool_call_invoke, ollama_pt, ) @@ -1590,3 +1591,153 @@ def test_bedrock_tools_unpack_defs_no_oom_with_nested_refs(): # Verify $defs have been removed (Bedrock doesn't support them) tool_schema = result[0]["toolSpec"].get("inputSchema", {}).get("json", {}) assert "$defs" not in tool_schema, "$defs should be removed after expansion" + + +# ── _convert_to_bedrock_tool_call_invoke tests ── + + +def test_bedrock_tool_call_invoke_normal_single_tool(): + """Normal single tool call with valid JSON arguments.""" + tool_calls = [ + { + "id": "call_abc123", + "type": "function", + "function": { + "name": "get_weather", + "arguments": '{"location": "Boston, MA"}', + }, + } + ] + result = _convert_to_bedrock_tool_call_invoke(tool_calls) + assert len(result) == 1 + assert result[0]["toolUse"]["toolUseId"] == "call_abc123" + assert result[0]["toolUse"]["name"] == "get_weather" + assert result[0]["toolUse"]["input"] == {"location": "Boston, MA"} + + +def test_bedrock_tool_call_invoke_empty_arguments(): + """Tool call with empty arguments produces an empty dict input.""" + tool_calls = [ + { + "id": "call_empty", + "type": "function", + "function": {"name": "do_something", "arguments": ""}, + } + ] + result = _convert_to_bedrock_tool_call_invoke(tool_calls) + assert len(result) == 1 + assert result[0]["toolUse"]["input"] == {} + + +def test_bedrock_tool_call_invoke_concatenated_json(): + """ + Tool call whose arguments contain multiple concatenated JSON objects + (the bug from issue #20543) is split into separate Bedrock toolUse blocks. + + Bedrock Claude Sonnet 4.5 sometimes returns multiple tool call arguments + concatenated in a single string like: + '{"command":["curl",...]}{"command":["curl",...]}{"command":["curl",...]}' + """ + tool_calls = [ + { + "id": "tooluse_L7I3TewYAUhoheJZQEuwVN", + "type": "function", + "function": { + "name": "shell", + "arguments": ( + '{"command": ["curl", "-i", "http://localhost:9009", "-m", "10"]}' + '{"command": ["curl", "-i", "http://localhost:9009/robots.txt", "-m", "5"]}' + '{"command": ["curl", "-i", "http://localhost:9009/sitemap.xml", "-m", "5"]}' + ), + }, + } + ] + result = _convert_to_bedrock_tool_call_invoke(tool_calls) + + # Should produce 3 separate toolUse blocks + assert len(result) == 3 + + # First block keeps original tool id + assert result[0]["toolUse"]["toolUseId"] == "tooluse_L7I3TewYAUhoheJZQEuwVN" + assert result[0]["toolUse"]["name"] == "shell" + assert result[0]["toolUse"]["input"] == { + "command": ["curl", "-i", "http://localhost:9009", "-m", "10"] + } + + # Subsequent blocks get suffixed ids + assert result[1]["toolUse"]["toolUseId"] == "tooluse_L7I3TewYAUhoheJZQEuwVN_1" + assert result[1]["toolUse"]["name"] == "shell" + assert result[1]["toolUse"]["input"] == { + "command": ["curl", "-i", "http://localhost:9009/robots.txt", "-m", "5"] + } + + assert result[2]["toolUse"]["toolUseId"] == "tooluse_L7I3TewYAUhoheJZQEuwVN_2" + assert result[2]["toolUse"]["name"] == "shell" + assert result[2]["toolUse"]["input"] == { + "command": ["curl", "-i", "http://localhost:9009/sitemap.xml", "-m", "5"] + } + + +def test_bedrock_tool_call_invoke_concatenated_json_with_cache_control(): + """ + When a tool call has cache_control AND concatenated JSON arguments, + the cachePoint block is appended after the last split block. + """ + tool_calls = [ + { + "id": "call_cached", + "type": "function", + "cache_control": {"type": "default"}, + "function": { + "name": "shell", + "arguments": '{"a": 1}{"b": 2}', + }, + } + ] + result = _convert_to_bedrock_tool_call_invoke(tool_calls) + + # 2 toolUse blocks + 1 cachePoint block + assert len(result) == 3 + assert "toolUse" in result[0] + assert "toolUse" in result[1] + assert "cachePoint" in result[2] + + +def test_bedrock_tool_call_invoke_non_dict_arguments(): + """Arguments that parse to a non-dict (e.g. '""') produce empty dict input.""" + tool_calls = [ + { + "id": "call_non_dict", + "type": "function", + "function": {"name": "tool", "arguments": '""'}, + } + ] + result = _convert_to_bedrock_tool_call_invoke(tool_calls) + assert len(result) == 1 + assert result[0]["toolUse"]["input"] == {} + + +def test_bedrock_tool_call_invoke_multiple_normal_tools(): + """Multiple separate tool calls (normal parallel calling) work correctly.""" + tool_calls = [ + { + "id": "call_1", + "type": "function", + "function": { + "name": "get_weather", + "arguments": '{"city": "NYC"}', + }, + }, + { + "id": "call_2", + "type": "function", + "function": { + "name": "get_weather", + "arguments": '{"city": "LA"}', + }, + }, + ] + result = _convert_to_bedrock_tool_call_invoke(tool_calls) + assert len(result) == 2 + assert result[0]["toolUse"]["toolUseId"] == "call_1" + assert result[1]["toolUse"]["toolUseId"] == "call_2" diff --git a/tests/test_litellm/llms/anthropic/experimental_pass_through/messages/test_anthropic_experimental_pass_through_messages_handler.py b/tests/test_litellm/llms/anthropic/experimental_pass_through/messages/test_anthropic_experimental_pass_through_messages_handler.py index 66d62aae1ec..80fd3ab698a 100644 --- a/tests/test_litellm/llms/anthropic/experimental_pass_through/messages/test_anthropic_experimental_pass_through_messages_handler.py +++ b/tests/test_litellm/llms/anthropic/experimental_pass_through/messages/test_anthropic_experimental_pass_through_messages_handler.py @@ -185,7 +185,14 @@ def test_openai_model_with_thinking_converts_to_reasoning_effort(): # Verify reasoning_effort is set (converted from thinking) assert "reasoning_effort" in call_kwargs, "reasoning_effort should be passed to completion" - assert call_kwargs["reasoning_effort"] == "minimal", f"reasoning_effort should be 'minimal' for budget_tokens=1024, got {call_kwargs.get('reasoning_effort')}" + assert call_kwargs["reasoning_effort"] == { + "effort": "minimal", + "summary": "detailed", + }, f"reasoning_effort should request a reasoning summary for OpenAI responses API, got {call_kwargs.get('reasoning_effort')}" + + # Verify OpenAI thinking requests are routed to the Responses API + assert call_kwargs.get("model") == "responses/gpt-5.2" + # Verify thinking is NOT passed (non-Claude model) assert "thinking" not in call_kwargs, "thinking should NOT be passed for non-Claude models" diff --git a/tests/test_litellm/llms/custom_httpx/test_http_handler.py b/tests/test_litellm/llms/custom_httpx/test_http_handler.py index 65f08ef5021..c249bd9970c 100644 --- a/tests/test_litellm/llms/custom_httpx/test_http_handler.py +++ b/tests/test_litellm/llms/custom_httpx/test_http_handler.py @@ -140,6 +140,83 @@ async def test_ssl_verification_with_aiohttp_transport(): litellm.disable_aiohttp_transport = original_disable +@pytest.mark.asyncio +async def test_ssl_verification_with_shared_session(): + """ + Test that ssl_verify=False is respected even with shared sessions. + + This was a bug where shared sessions bypassed SSL configuration because + _create_aiohttp_transport returned immediately without passing ssl_verify + to the LiteLLMAiohttpTransport constructor. + + The fix stores ssl_verify in the transport and passes it per-request. + """ + import aiohttp + + # Ensure aiohttp transport is enabled for this test + original_disable = litellm.disable_aiohttp_transport + litellm.disable_aiohttp_transport = False + + try: + # Create a shared session (simulating what happens in production) + shared_session = aiohttp.ClientSession() + + try: + # Create transport with shared session and ssl_verify=False + transport = AsyncHTTPHandler._create_aiohttp_transport( + ssl_verify=False, + shared_session=shared_session, + ) + + # Verify the transport uses the shared session + assert transport.client is shared_session + + # Verify the SSL setting is stored in the transport for per-request use + assert transport._ssl_verify is False + finally: + await shared_session.close() + finally: + # Restore original setting + litellm.disable_aiohttp_transport = original_disable + + +@pytest.mark.asyncio +async def test_ssl_context_with_shared_session(): + """ + Test that ssl_context is respected even with shared sessions. + """ + import aiohttp + + # Ensure aiohttp transport is enabled for this test + original_disable = litellm.disable_aiohttp_transport + litellm.disable_aiohttp_transport = False + + try: + # Create a custom SSL context + custom_ssl_context = ssl.create_default_context() + + # Create a shared session + shared_session = aiohttp.ClientSession() + + try: + # Create transport with shared session and custom ssl_context + transport = AsyncHTTPHandler._create_aiohttp_transport( + ssl_context=custom_ssl_context, + shared_session=shared_session, + ) + + # Verify the transport uses the shared session + assert transport.client is shared_session + + # Verify the SSL context is stored in the transport for per-request use + assert transport._ssl_verify is custom_ssl_context + finally: + await shared_session.close() + finally: + # Restore original setting + litellm.disable_aiohttp_transport = original_disable + + @pytest.mark.asyncio async def test_aiohttp_transport_trust_env_setting(monkeypatch): """Test that trust_env setting is properly configured in aiohttp transport""" diff --git a/tests/test_litellm/llms/databricks/chat/test_databricks_chat_transformation.py b/tests/test_litellm/llms/databricks/chat/test_databricks_chat_transformation.py index f437b8405f7..f9b5b5fe29c 100644 --- a/tests/test_litellm/llms/databricks/chat/test_databricks_chat_transformation.py +++ b/tests/test_litellm/llms/databricks/chat/test_databricks_chat_transformation.py @@ -13,6 +13,7 @@ from unittest.mock import MagicMock, patch from litellm.llms.databricks.chat.transformation import ( DatabricksChatResponseIterator, DatabricksConfig, + _sanitize_empty_content, ) @@ -215,3 +216,45 @@ def test_chunk_parser_with_citation(): "end_char_index": 50, } } + + +def test_sanitize_empty_content_pops_none(): + message = {"role": "user", "content": None} + _sanitize_empty_content(message) + assert "content" not in message + + +def test_sanitize_empty_content_pops_empty_string(): + message = {"role": "user", "content": ""} + _sanitize_empty_content(message) + assert "content" not in message + + +def test_sanitize_empty_content_pops_single_empty_text_block(): + message = {"role": "user", "content": [{"type": "text", "text": ""}]} + _sanitize_empty_content(message) + assert "content" not in message + + +def test_sanitize_empty_content_filters_empty_blocks_keeps_non_empty(): + message = { + "role": "user", + "content": [ + {"type": "text", "text": ""}, + {"type": "text", "text": "Hello"}, + {"type": "text", "text": " "}, + ], + } + _sanitize_empty_content(message) + assert message["content"] == [{"type": "text", "text": "Hello"}] + + +def test_transform_messages_sanitizes_empty_content(): + config = DatabricksConfig() + messages = [ + {"role": "user", "content": [{"type": "text", "text": ""}]}, + {"role": "user", "content": "Hi"}, + ] + result = config._transform_messages(messages=messages, model="databricks-claude", is_async=False) + assert "content" not in result[0] + assert result[1]["content"] == "Hi" diff --git a/tests/test_litellm/llms/ollama/test_ollama_chat_transformation.py b/tests/test_litellm/llms/ollama/test_ollama_chat_transformation.py index af6481a6cb0..02495106a84 100644 --- a/tests/test_litellm/llms/ollama/test_ollama_chat_transformation.py +++ b/tests/test_litellm/llms/ollama/test_ollama_chat_transformation.py @@ -10,7 +10,8 @@ sys.path.insert( 0, os.path.abspath(os.path.join(os.path.dirname(__file__), "../../../../..")) ) -from litellm.llms.ollama.chat.transformation import OllamaChatConfig +from litellm.llms.ollama.chat.transformation import OllamaChatConfig, OllamaChatCompletionResponseIterator + from litellm.types.llms.openai import AllMessageValues from litellm.utils import get_optional_params @@ -473,3 +474,130 @@ class TestOllamaToolCalling: # finish_reason should be "stop" (default behavior) assert result.choices[0].finish_reason == "stop" assert result.choices[0].message.tool_calls is None + + +class TestOllamaReasoningContentStreaming: + """Test that reasoning_content is properly extracted from all thinking chunks.""" + + def test_multiple_thinking_chunks_all_returned_as_reasoning_content(self): + """ + Test that more than 2 consecutive thinking chunks are all returned as reasoning_content. + + Previously, the code had a bug where finished_reasoning_content was set to True + after just 2 chunks with 'thinking', causing subsequent thinking content to be lost. + """ + iterator = OllamaChatCompletionResponseIterator( + streaming_response=iter([]), # Not used in chunk_parser + sync_stream=True, + ) + + # Simulate 5 consecutive chunks with 'thinking' content + thinking_chunks = [ + { + "model": "deepseek-r1", + "message": {"role": "assistant", "thinking": f"Thinking chunk {i}"}, + "done": False, + } + for i in range(1, 6) + ] + + # Process all thinking chunks + reasoning_contents = [] + for chunk in thinking_chunks: + result = iterator.chunk_parser(chunk) + rc = result.choices[0].delta.reasoning_content + reasoning_contents.append(rc) + + # ALL chunks should have reasoning_content, not just the first 2 + assert len(reasoning_contents) == 5 + assert reasoning_contents[0] == "Thinking chunk 1" + assert reasoning_contents[1] == "Thinking chunk 2" + assert reasoning_contents[2] == "Thinking chunk 3" # This was previously None + assert reasoning_contents[3] == "Thinking chunk 4" # This was previously None + assert reasoning_contents[4] == "Thinking chunk 5" # This was previously None + + # Verify none of them are None + for i, rc in enumerate(reasoning_contents): + assert rc is not None, f"Chunk {i+1} reasoning_content should not be None" + + def test_thinking_to_content_transition(self): + """ + Test that transition from thinking to regular content works correctly. + """ + iterator = OllamaChatCompletionResponseIterator( + streaming_response=iter([]), + sync_stream=True, + ) + + # First: thinking chunks + thinking_chunk = { + "model": "deepseek-r1", + "message": {"role": "assistant", "thinking": "Let me think about this..."}, + "done": False, + } + result1 = iterator.chunk_parser(thinking_chunk) + assert result1.choices[0].delta.reasoning_content == "Let me think about this..." + assert result1.choices[0].delta.content is None + + # Then: regular content chunk + content_chunk = { + "model": "deepseek-r1", + "message": {"role": "assistant", "content": "Here is my answer."}, + "done": False, + } + result2 = iterator.chunk_parser(content_chunk) + assert result2.choices[0].delta.content == "Here is my answer." + # reasoning_content is not set when there's no thinking in the chunk + assert getattr(result2.choices[0].delta, 'reasoning_content', None) is None + + def test_think_tags_in_content(self): + """ + Test that tags embedded in content are properly parsed. + """ + iterator = OllamaChatCompletionResponseIterator( + streaming_response=iter([]), + sync_stream=True, + ) + + # Content with tag + chunk1 = { + "model": "deepseek-r1", + "message": {"role": "assistant", "content": "I need to analyze this"}, + "done": False, + } + result1 = iterator.chunk_parser(chunk1) + assert result1.choices[0].delta.reasoning_content == "I need to analyze this" + assert result1.choices[0].delta.content is None + + # Content with tag (end of thinking) + chunk2 = { + "model": "deepseek-r1", + "message": {"role": "assistant", "content": "The answer is 42."}, + "done": False, + } + result2 = iterator.chunk_parser(chunk2) + assert result2.choices[0].delta.content == "The answer is 42." + # reasoning_content is not set when it's regular content + assert getattr(result2.choices[0].delta, 'reasoning_content', None) is None + + def test_done_chunk_with_thinking(self): + """ + Test that the final chunk with done=True and thinking content works. + """ + iterator = OllamaChatCompletionResponseIterator( + streaming_response=iter([]), + sync_stream=True, + ) + + # Final chunk with thinking + done_chunk = { + "model": "deepseek-r1", + "message": {"role": "assistant", "thinking": "Final thought"}, + "done": True, + "done_reason": "stop", + } + result = iterator.chunk_parser(done_chunk) + assert result.choices[0].delta.reasoning_content == "Final thought" + assert result.choices[0].finish_reason == "stop" + + diff --git a/tests/test_litellm/llms/vertex_ai/gemini/test_vertex_ai_gemini_transformation.py b/tests/test_litellm/llms/vertex_ai/gemini/test_vertex_ai_gemini_transformation.py index a9c27e30930..c474461e0a2 100644 --- a/tests/test_litellm/llms/vertex_ai/gemini/test_vertex_ai_gemini_transformation.py +++ b/tests/test_litellm/llms/vertex_ai/gemini/test_vertex_ai_gemini_transformation.py @@ -7,6 +7,7 @@ from litellm.llms.vertex_ai.gemini.transformation import ( check_if_part_exists_in_parts, ) from litellm.types.llms.vertex_ai import BlobType +from litellm.types.utils import Message def test_check_if_part_exists_in_parts(): @@ -735,8 +736,9 @@ def test_file_data_field_order(): Related issue: Gemini API returns 400 INVALID_ARGUMENT when fields are in wrong order. """ import json + from litellm.llms.vertex_ai.gemini.transformation import _process_gemini_media - + # Test with HTTPS URL and explicit format (audio file) file_url = "https://generativelanguage.googleapis.com/v1beta/files/test123" format = "audio/mpeg" @@ -770,8 +772,9 @@ def test_file_data_field_order(): def test_file_data_field_order_gcs_urls(): """Test that GCS URLs also maintain correct field order.""" import json + from litellm.llms.vertex_ai.gemini.transformation import _process_gemini_media - + # Test with GCS URL gcs_url = "gs://bucket/audio.mp3" @@ -802,11 +805,14 @@ def test_extract_file_data_with_path_object(): Related issue: Files uploaded with wrong MIME type cause Gemini API to reject requests where the specified format doesn't match the uploaded file's MIME type. """ - from pathlib import Path - from litellm.litellm_core_utils.prompt_templates.common_utils import extract_file_data - import tempfile import os - + import tempfile + from pathlib import Path + + from litellm.litellm_core_utils.prompt_templates.common_utils import ( + extract_file_data, + ) + # Create a temporary MP3 file with tempfile.NamedTemporaryFile(suffix=".mp3", delete=False) as tmp: tmp.write(b"fake mp3 content") @@ -835,10 +841,13 @@ def test_extract_file_data_with_path_object(): def test_extract_file_data_with_string_path(): """Test that filename is correctly extracted from string paths.""" - from litellm.litellm_core_utils.prompt_templates.common_utils import extract_file_data - import tempfile import os - + import tempfile + + from litellm.litellm_core_utils.prompt_templates.common_utils import ( + extract_file_data, + ) + # Create a temporary WAV file with tempfile.NamedTemporaryFile(suffix=".wav", delete=False) as tmp: tmp.write(b"fake wav content") @@ -866,8 +875,10 @@ def test_extract_file_data_with_string_path(): def test_extract_file_data_with_tuple_format(): """Test that tuple format (with explicit content_type) still works correctly.""" - from litellm.litellm_core_utils.prompt_templates.common_utils import extract_file_data - + from litellm.litellm_core_utils.prompt_templates.common_utils import ( + extract_file_data, + ) + # Test with tuple format: (filename, content, content_type) filename = "test_audio.mp3" content = b"test audio content" @@ -883,10 +894,13 @@ def test_extract_file_data_with_tuple_format(): def test_extract_file_data_fallback_to_octet_stream(): """Test that unknown file types fall back to application/octet-stream.""" - from litellm.litellm_core_utils.prompt_templates.common_utils import extract_file_data - import tempfile import os - + import tempfile + + from litellm.litellm_core_utils.prompt_templates.common_utils import ( + extract_file_data, + ) + # Create a temporary file with unknown extension with tempfile.NamedTemporaryFile(suffix=".xyz123", delete=False) as tmp: tmp.write(b"unknown content") @@ -1086,3 +1100,227 @@ def test_convert_tool_response_with_nested_file_object(): assert "mime_type" in inline_data assert inline_data["mime_type"] == "application/pdf" assert inline_data["data"] == test_pdf_base64 + +def test_assistant_message_with_images_field(): + """ + Test that assistant messages with images field are properly converted to Gemini format. + + This handles the case where an assistant message contains generated images in the + `images` field (e.g., from image generation models like gemini-2.5-flash-image). + The images should be converted to inline_data parts in the Gemini format. + """ + # Create a small test image (1x1 red pixel PNG) + test_image_base64 = "iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAYAAAAfFcSJAAAADUlEQVR42mNk+M9QDwADhgGAWjR9awAAAABJRU5ErkJggg==" + image_data_uri = f"data:image/png;base64,{test_image_base64}" + + # Create messages with assistant message containing images field + messages = [ + { + "role": "user", + "content": "Generate an image of a banana wearing a costume that says LiteLLM" + }, + { + "role": "assistant", + "content": "Here's your banana in a LiteLLM costume!", + "images": [ + { + "image_url": { + "url": image_data_uri, + "detail": "auto" + }, + "index": 0, + "type": "image_url" + } + ] + } + ] + + # Convert messages to Gemini format + contents = _gemini_convert_messages_with_history(messages=messages) + + # Verify structure + assert len(contents) == 2, f"Expected 2 content blocks, got {len(contents)}" + + # Verify user message + assert contents[0]["role"] == "user" + assert len(contents[0]["parts"]) == 1 + assert contents[0]["parts"][0]["text"] == "Generate an image of a banana wearing a costume that says LiteLLM" + + # Verify assistant message + assert contents[1]["role"] == "model" + assert len(contents[1]["parts"]) == 2, f"Expected 2 parts (text + image), got {len(contents[1]['parts'])}" + + # Find text part and inline_data part + text_part = None + inline_data_part = None + for part in contents[1]["parts"]: + if "text" in part: + text_part = part + elif "inline_data" in part: + inline_data_part = part + + # Verify text part + assert text_part is not None, "Missing text part in assistant message" + assert text_part["text"] == "Here's your banana in a LiteLLM costume!" + + # Verify inline_data part (image) + assert inline_data_part is not None, "Missing inline_data part in assistant message" + inline_data: BlobType = inline_data_part["inline_data"] + assert "data" in inline_data + assert "mime_type" in inline_data + assert inline_data["mime_type"] == "image/png" + assert inline_data["data"] == test_image_base64 + + +def test_assistant_message_with_multiple_images(): + """Test that assistant messages with multiple images are properly converted.""" + # Create two test images + test_image1_base64 = "iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAYAAAAfFcSJAAAADUlEQVR42mNk+M9QDwADhgGAWjR9awAAAABJRU5ErkJggg==" + test_image2_base64 = "iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAYAAAAfFcSJAAAADUlEQVR42mP8z8DwHwAFBQIAX8jx0gAAAABJRU5ErkJggg==" + image1_data_uri = f"data:image/png;base64,{test_image1_base64}" + image2_data_uri = f"data:image/jpeg;base64,{test_image2_base64}" + + messages = [ + { + "role": "user", + "content": "Generate two images" + }, + { + "role": "assistant", + "content": "Here are your images:", + "images": [ + { + "image_url": { + "url": image1_data_uri, + "detail": "auto" + }, + "index": 0, + "type": "image_url" + }, + { + "image_url": { + "url": image2_data_uri, + "detail": "high" + }, + "index": 1, + "type": "image_url" + } + ] + } + ] + + # Convert messages to Gemini format + contents = _gemini_convert_messages_with_history(messages=messages) + + # Verify assistant message has 3 parts (1 text + 2 images) + assert contents[1]["role"] == "model" + assert len(contents[1]["parts"]) == 3, f"Expected 3 parts (text + 2 images), got {len(contents[1]['parts'])}" + + # Count inline_data parts + inline_data_parts = [part for part in contents[1]["parts"] if "inline_data" in part] + assert len(inline_data_parts) == 2, f"Expected 2 inline_data parts, got {len(inline_data_parts)}" + + # Verify first image + assert inline_data_parts[0]["inline_data"]["mime_type"] == "image/png" + assert inline_data_parts[0]["inline_data"]["data"] == test_image1_base64 + + # Verify second image + assert inline_data_parts[1]["inline_data"]["mime_type"] == "image/jpeg" + assert inline_data_parts[1]["inline_data"]["data"] == test_image2_base64 + + +def test_assistant_message_with_images_using_message_object(): + """Test that Message objects with images field are properly converted.""" + # Create a small test image + test_image_base64 = "iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAYAAAAfFcSJAAAADUlEQVR42mNk+M9QDwADhgGAWjR9awAAAABJRU5ErkJggg==" + image_data_uri = f"data:image/png;base64,{test_image_base64}" + + # Create messages using Message object (as returned by LiteLLM) + user_message = { + "role": "user", + "content": "Generate an image" + } + + assistant_message = Message( + content="Here's your image!", + role="assistant", + tool_calls=None, + function_call=None, + images=[ + { + "image_url": { + "url": image_data_uri, + "detail": "auto" + }, + "index": 0, + "type": "image_url" + } + ] + ) + + messages = [user_message, assistant_message] + + # Convert messages to Gemini format + contents = _gemini_convert_messages_with_history(messages=messages) + + # Verify assistant message has both text and image + assert contents[1]["role"] == "model" + assert len(contents[1]["parts"]) == 2 + + # Verify image was converted + inline_data_parts = [part for part in contents[1]["parts"] if "inline_data" in part] + assert len(inline_data_parts) == 1 + assert inline_data_parts[0]["inline_data"]["mime_type"] == "image/png" + assert inline_data_parts[0]["inline_data"]["data"] == test_image_base64 + + +def test_assistant_message_with_images_in_conversation_history(): + """ + Test multi-turn conversation where assistant message with images is in history. + + This simulates the real use case where: + 1. User asks for image generation + 2. Assistant generates image (with images field) + 3. User asks follow-up question about the image + """ + test_image_base64 = "iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAYAAAAfFcSJAAAADUlEQVR42mNk+M9QDwADhgGAWjR9awAAAABJRU5ErkJggg==" + image_data_uri = f"data:image/png;base64,{test_image_base64}" + + messages = [ + { + "role": "user", + "content": "Generate an image of a cat" + }, + { + "role": "assistant", + "content": "Here's a cat image:", + "images": [ + { + "image_url": { + "url": image_data_uri, + "detail": "auto" + }, + "index": 0, + "type": "image_url" + } + ] + }, + { + "role": "user", + "content": "Can you make it more colorful?" + } + ] + + # Convert messages to Gemini format + contents = _gemini_convert_messages_with_history(messages=messages) + + # Verify structure: user -> model (with image) -> user + assert len(contents) == 3 + assert contents[0]["role"] == "user" + assert contents[1]["role"] == "model" + assert contents[2]["role"] == "user" + + # Verify assistant message has image in history + inline_data_parts = [part for part in contents[1]["parts"] if "inline_data" in part] + assert len(inline_data_parts) == 1 + assert inline_data_parts[0]["inline_data"]["mime_type"] == "image/png" \ No newline at end of file diff --git a/tests/test_litellm/proxy/_experimental/mcp_server/test_discoverable_endpoints.py b/tests/test_litellm/proxy/_experimental/mcp_server/test_discoverable_endpoints.py index e4b4d3ba189..faabe40f2dc 100644 --- a/tests/test_litellm/proxy/_experimental/mcp_server/test_discoverable_endpoints.py +++ b/tests/test_litellm/proxy/_experimental/mcp_server/test_discoverable_endpoints.py @@ -2,6 +2,8 @@ import pytest from unittest.mock import AsyncMock, MagicMock, patch +from fastapi import HTTPException + # Fixture to mock IP address check for all MCP tests # This prevents tests from failing due to IP-based access control @@ -260,10 +262,16 @@ async def test_register_client_without_mcp_server_name_returns_dummy(): from litellm.proxy._experimental.mcp_server.discoverable_endpoints import ( register_client, ) + from litellm.proxy._experimental.mcp_server.mcp_server_manager import ( + global_mcp_server_manager, + ) from fastapi import Request except ImportError: pytest.skip("MCP discoverable endpoints not available") + # Clear registry to ensure no OAuth2 servers exist (otherwise resolver would find one) + global_mcp_server_manager.registry.clear() + mock_request = MagicMock(spec=Request) mock_request.base_url = "https://proxy.litellm.example/" mock_request.headers = {} @@ -680,10 +688,16 @@ async def test_register_client_respects_x_forwarded_proto(): from litellm.proxy._experimental.mcp_server.discoverable_endpoints import ( register_client, ) + from litellm.proxy._experimental.mcp_server.mcp_server_manager import ( + global_mcp_server_manager, + ) from fastapi import Request except ImportError: pytest.skip("MCP discoverable endpoints not available") + # Clear registry to ensure no OAuth2 servers exist (otherwise resolver would find one) + global_mcp_server_manager.registry.clear() + # Mock request with http base_url but X-Forwarded-Proto: https mock_request = MagicMock(spec=Request) mock_request.base_url = "http://proxy.litellm.example/" # HTTP @@ -1017,3 +1031,263 @@ def test_get_request_base_url_comprehensive( f"X-Forwarded-Host={x_forwarded_host}, " f"X-Forwarded-Port={x_forwarded_port}" ) + + +# ------------------------------------------------------------------- +# Tests for root-level OAuth endpoint resolution (no server name) +# ------------------------------------------------------------------- + + +def _create_oauth2_server( + server_id="test_oauth_server", + name="test_oauth", + server_name="test_oauth", + alias="test_oauth", + client_id="test_client_id", + client_secret="test_client_secret", +): + """Helper to create a mock OAuth2 MCPServer.""" + from litellm.types.mcp import MCPAuth + from litellm.types.mcp_server.mcp_server_manager import MCPServer + from litellm.proxy._types import MCPTransport + + return MCPServer( + server_id=server_id, + name=name, + server_name=server_name, + alias=alias, + transport=MCPTransport.http, + auth_type=MCPAuth.oauth2, + client_id=client_id, + client_secret=client_secret, + authorization_url="https://provider.com/oauth/authorize", + token_url="https://provider.com/oauth/token", + scopes=["read", "write"], + ) + + +@pytest.mark.asyncio +async def test_authorize_root_resolves_single_oauth2_server(): + """When /authorize is hit without server name and exactly 1 OAuth2 server exists, resolve it.""" + try: + from litellm.proxy._experimental.mcp_server.discoverable_endpoints import ( + authorize, + ) + from litellm.proxy._experimental.mcp_server.mcp_server_manager import ( + global_mcp_server_manager, + ) + from fastapi import Request + except ImportError: + pytest.skip("MCP discoverable endpoints not available") + + global_mcp_server_manager.registry.clear() + oauth2_server = _create_oauth2_server() + global_mcp_server_manager.registry[oauth2_server.server_id] = oauth2_server + + mock_request = MagicMock(spec=Request) + mock_request.base_url = "https://llm.example.com/" + mock_request.headers = {} + + try: + with patch( + "litellm.proxy._experimental.mcp_server.discoverable_endpoints.encrypt_value_helper" + ) as mock_encrypt: + mock_encrypt.return_value = "mocked_encrypted_state" + + # Call /authorize WITHOUT mcp_server_name, with dummy_client as client_id + response = await authorize( + request=mock_request, + client_id="dummy_client", + mcp_server_name=None, + redirect_uri="http://localhost:62646/callback", + state="test_state", + ) + + # Should resolve to the single OAuth2 server and redirect + assert response.status_code == 307 + location = response.headers["location"] + assert "https://provider.com/oauth/authorize" in location + assert "client_id=test_client_id" in location + finally: + global_mcp_server_manager.registry.clear() + + +@pytest.mark.asyncio +async def test_authorize_root_fails_with_multiple_oauth2_servers(): + """When /authorize is hit without server name and multiple OAuth2 servers exist, return 404.""" + try: + from litellm.proxy._experimental.mcp_server.discoverable_endpoints import ( + authorize, + ) + from litellm.proxy._experimental.mcp_server.mcp_server_manager import ( + global_mcp_server_manager, + ) + from fastapi import Request + except ImportError: + pytest.skip("MCP discoverable endpoints not available") + + global_mcp_server_manager.registry.clear() + server1 = _create_oauth2_server( + server_id="server1", name="server1", server_name="server1", alias="server1" + ) + server2 = _create_oauth2_server( + server_id="server2", name="server2", server_name="server2", alias="server2" + ) + global_mcp_server_manager.registry[server1.server_id] = server1 + global_mcp_server_manager.registry[server2.server_id] = server2 + + mock_request = MagicMock(spec=Request) + mock_request.base_url = "https://llm.example.com/" + mock_request.headers = {} + + try: + with pytest.raises(HTTPException) as exc_info: + await authorize( + request=mock_request, + client_id="dummy_client", + mcp_server_name=None, + redirect_uri="http://localhost:62646/callback", + state="test_state", + ) + assert exc_info.value.status_code == 404 + assert "MCP server not found" in str(exc_info.value.detail) + finally: + global_mcp_server_manager.registry.clear() + + +@pytest.mark.asyncio +async def test_token_root_resolves_single_oauth2_server(): + """When /token is hit without server name and exactly 1 OAuth2 server exists, resolve it.""" + try: + from litellm.proxy._experimental.mcp_server.discoverable_endpoints import ( + token_endpoint, + ) + from litellm.proxy._experimental.mcp_server.mcp_server_manager import ( + global_mcp_server_manager, + ) + from fastapi import Request + except ImportError: + pytest.skip("MCP discoverable endpoints not available") + + global_mcp_server_manager.registry.clear() + oauth2_server = _create_oauth2_server() + global_mcp_server_manager.registry[oauth2_server.server_id] = oauth2_server + + mock_request = MagicMock(spec=Request) + mock_request.base_url = "https://llm.example.com/" + mock_request.headers = {} + + mock_response = MagicMock() + mock_response.json.return_value = { + "access_token": "ya29.test_token", + "token_type": "Bearer", + "expires_in": 3599, + } + mock_response.raise_for_status = MagicMock() + + mock_async_client = MagicMock() + mock_async_client.post = AsyncMock(return_value=mock_response) + + try: + with patch( + "litellm.proxy._experimental.mcp_server.discoverable_endpoints.get_async_httpx_client" + ) as mock_get_client: + mock_get_client.return_value = mock_async_client + + # Call /token WITHOUT mcp_server_name + response = await token_endpoint( + request=mock_request, + grant_type="authorization_code", + code="test_auth_code", + redirect_uri="http://localhost:62646/callback", + client_id="dummy_client", + mcp_server_name=None, + client_secret=None, + code_verifier="test_verifier", + ) + + # Should resolve and exchange token with the upstream server + import json + + token_data = json.loads(response.body) + assert token_data["access_token"] == "ya29.test_token" + + # Verify it called the correct upstream token URL + call_args = mock_async_client.post.call_args + assert call_args.args[0] == "https://provider.com/oauth/token" + finally: + global_mcp_server_manager.registry.clear() + + +@pytest.mark.asyncio +async def test_register_root_resolves_single_oauth2_server(): + """When /register is hit without server name and exactly 1 OAuth2 server exists, resolve it.""" + try: + from litellm.proxy._experimental.mcp_server.discoverable_endpoints import ( + register_client, + ) + from litellm.proxy._experimental.mcp_server.mcp_server_manager import ( + global_mcp_server_manager, + ) + from fastapi import Request + except ImportError: + pytest.skip("MCP discoverable endpoints not available") + + global_mcp_server_manager.registry.clear() + oauth2_server = _create_oauth2_server() + global_mcp_server_manager.registry[oauth2_server.server_id] = oauth2_server + + mock_request = MagicMock(spec=Request) + mock_request.base_url = "https://llm.example.com/" + mock_request.headers = {} + + try: + with patch( + "litellm.proxy._experimental.mcp_server.discoverable_endpoints._read_request_body", + new=AsyncMock(return_value={}), + ): + result = await register_client(request=mock_request, mcp_server_name=None) + + # Should resolve to the single server and return its name as client_id + assert result["client_id"] == "test_oauth" + assert "redirect_uris" in result + finally: + global_mcp_server_manager.registry.clear() + + +@pytest.mark.asyncio +async def test_discovery_root_includes_server_name_prefix(): + """When root discovery is hit and exactly 1 OAuth2 server exists, include server name in URLs.""" + try: + from litellm.proxy._experimental.mcp_server.discoverable_endpoints import ( + _build_oauth_authorization_server_response, + ) + from litellm.proxy._experimental.mcp_server.mcp_server_manager import ( + global_mcp_server_manager, + ) + from fastapi import Request + except ImportError: + pytest.skip("MCP discoverable endpoints not available") + + global_mcp_server_manager.registry.clear() + oauth2_server = _create_oauth2_server() + global_mcp_server_manager.registry[oauth2_server.server_id] = oauth2_server + + mock_request = MagicMock(spec=Request) + mock_request.base_url = "https://llm.example.com/" + mock_request.headers = {} + + try: + # Call with mcp_server_name=None (root discovery) + response = _build_oauth_authorization_server_response( + request=mock_request, + mcp_server_name=None, + ) + + # Should resolve to the single server and include its name in endpoint URLs + assert "/test_oauth/authorize" in response["authorization_endpoint"] + assert "/test_oauth/token" in response["token_endpoint"] + assert "/test_oauth/register" in response["registration_endpoint"] + assert response["scopes_supported"] == ["read", "write"] + finally: + global_mcp_server_manager.registry.clear() diff --git a/tests/test_litellm/proxy/_experimental/mcp_server/test_mcp_server_manager.py b/tests/test_litellm/proxy/_experimental/mcp_server/test_mcp_server_manager.py index abb8dd49159..b4b5811666b 100644 --- a/tests/test_litellm/proxy/_experimental/mcp_server/test_mcp_server_manager.py +++ b/tests/test_litellm/proxy/_experimental/mcp_server/test_mcp_server_manager.py @@ -1043,7 +1043,7 @@ class TestMCPServerManager: "litellm.proxy._experimental.mcp_server.tool_registry.global_mcp_tool_registry.register_tool", return_value=None, ): - manager._register_openapi_tools( + await manager._register_openapi_tools( spec_path=str(spec_path), server=server, base_url="https://example.com", diff --git a/tests/test_litellm/proxy/_experimental/mcp_server/test_rest_endpoints.py b/tests/test_litellm/proxy/_experimental/mcp_server/test_rest_endpoints.py index f6c8115f7bc..9c77edc6743 100644 --- a/tests/test_litellm/proxy/_experimental/mcp_server/test_rest_endpoints.py +++ b/tests/test_litellm/proxy/_experimental/mcp_server/test_rest_endpoints.py @@ -155,6 +155,160 @@ class TestExecuteWithMcpClient: } + @pytest.mark.asyncio + async def test_m2m_credentials_forwarded_to_server_model(self, monkeypatch): + """M2M OAuth credentials (client_id, client_secret) from the nested + ``credentials`` dict must be forwarded to the MCPServer model so that + ``has_client_credentials`` returns True and the proxy auto-fetches tokens.""" + captured: dict = {} + + def fake_build_stdio_env(server, raw_headers): + return None + + async def fake_create_client(*args, **kwargs): + captured["server"] = kwargs.get("server") + return object() + + monkeypatch.setattr( + rest_endpoints.global_mcp_server_manager, + "_build_stdio_env", + fake_build_stdio_env, + raising=False, + ) + monkeypatch.setattr( + rest_endpoints.global_mcp_server_manager, + "_create_mcp_client", + fake_create_client, + raising=False, + ) + + async def ok_operation(client): + return {"status": "ok"} + + payload = NewMCPServerRequest( + server_name="m2m-server", + url="https://example.com", + auth_type=MCPAuth.oauth2, + token_url="https://auth.example.com/token", + credentials={ + "client_id": "my-id", + "client_secret": "my-secret", + "scopes": ["read", "write"], + }, + ) + + result = await rest_endpoints._execute_with_mcp_client( + payload, ok_operation + ) + + assert result["status"] == "ok" + server = captured["server"] + assert server.client_id == "my-id" + assert server.client_secret == "my-secret" + assert server.token_url == "https://auth.example.com/token" + assert server.scopes == ["read", "write"] + assert server.has_client_credentials is True + + @pytest.mark.asyncio + async def test_m2m_drops_incoming_oauth2_headers(self, monkeypatch): + """For M2M OAuth servers the incoming Authorization header (which carries + the litellm API key) must NOT be forwarded as extra_headers — otherwise + it overwrites the auto-fetched M2M token.""" + captured: dict = {} + + def fake_build_stdio_env(server, raw_headers): + return None + + async def fake_create_client(*args, **kwargs): + captured["extra_headers"] = kwargs.get("extra_headers") + return object() + + monkeypatch.setattr( + rest_endpoints.global_mcp_server_manager, + "_build_stdio_env", + fake_build_stdio_env, + raising=False, + ) + monkeypatch.setattr( + rest_endpoints.global_mcp_server_manager, + "_create_mcp_client", + fake_create_client, + raising=False, + ) + + async def ok_operation(client): + return {"status": "ok"} + + payload = NewMCPServerRequest( + server_name="m2m-server", + url="https://example.com", + auth_type=MCPAuth.oauth2, + token_url="https://auth.example.com/token", + credentials={ + "client_id": "my-id", + "client_secret": "my-secret", + }, + ) + + incoming_oauth2 = {"Authorization": "Bearer sk-litellm-api-key"} + result = await rest_endpoints._execute_with_mcp_client( + payload, + ok_operation, + oauth2_headers=incoming_oauth2, + ) + + assert result["status"] == "ok" + # The incoming Authorization must be dropped — extra_headers should + # contain no oauth2 headers (only static_headers, which are None here). + assert captured["extra_headers"] is None or "Authorization" not in captured["extra_headers"] + + @pytest.mark.asyncio + async def test_catches_exception_group(self, monkeypatch): + """MCP SDK's anyio TaskGroup raises BaseExceptionGroup which does not + inherit from Exception. The handler must catch it and return an error + dict instead of letting a raw 500 propagate.""" + + def fake_build_stdio_env(server, raw_headers): + return None + + async def fake_create_client(*args, **kwargs): + raise BaseExceptionGroup( + "test group", [RuntimeError("Cancelled via cancel scope")] + ) + + monkeypatch.setattr( + rest_endpoints.global_mcp_server_manager, + "_build_stdio_env", + fake_build_stdio_env, + raising=False, + ) + monkeypatch.setattr( + rest_endpoints.global_mcp_server_manager, + "_create_mcp_client", + fake_create_client, + raising=False, + ) + + async def ok_operation(client): + return {"status": "ok"} + + payload = NewMCPServerRequest( + server_name="bad-server", + url="https://example.com", + auth_type=MCPAuth.none, + ) + + result = await rest_endpoints._execute_with_mcp_client( + payload, ok_operation + ) + + assert result["status"] == "error" + assert result["error"] is True + assert "Failed to connect to MCP server" in result["message"] + # Error message must not leak raw exception details + assert "cancel scope" not in result["message"] + + class TestTestConnection: def test_requires_auth_dependency(self): route = _get_route("/mcp-rest/test/connection", "POST") diff --git a/tests/test_litellm/proxy/auth/test_route_checks.py b/tests/test_litellm/proxy/auth/test_route_checks.py index a745ac3de13..65bb329e7ea 100644 --- a/tests/test_litellm/proxy/auth/test_route_checks.py +++ b/tests/test_litellm/proxy/auth/test_route_checks.py @@ -108,6 +108,22 @@ def test_virtual_key_allowed_routes_with_litellm_routes_member_name_allowed(): assert result is True +def test_virtual_key_mcp_routes_allows_v1_mcp_server(): + """Regression test for #20325: allow virtual keys to list MCP servers.""" + + valid_token = UserAPIKeyAuth( + user_id="test_user", + allowed_routes=["mcp_routes"], + ) + + result = RouteChecks.is_virtual_key_allowed_to_call_route( + route="/v1/mcp/server", + valid_token=valid_token, + ) + + assert result is True + + def test_virtual_key_allowed_routes_with_litellm_routes_member_name_denied(): """Test that virtual key is denied when route is not in the allowed LiteLLMRoutes group""" @@ -199,14 +215,14 @@ def test_virtual_key_llm_api_routes_allows_google_routes(route): def test_google_routes_with_dynamic_model_names_recognized_as_llm_api_route(route): """ Test that Google routes with dynamic model names (including custom names) are recognized as LLM API routes. - + This test verifies the fix for the issue where routes like: /v1beta/models/google-gemini-2-5-pro-code-reviewer-k8s:generateContent were incorrectly classified as "custom admin only route" instead of LLM API routes. - + The fix adds pattern matching for Google routes with placeholders like {model_name}. """ - + # Test that the route is recognized as an LLM API route assert RouteChecks.is_llm_api_route(route) is True @@ -214,28 +230,28 @@ def test_google_routes_with_dynamic_model_names_recognized_as_llm_api_route(rout def test_google_routes_with_dynamic_model_names_accessible_to_internal_users(): """ Test that internal users can access Google routes with dynamic model names. - + This ensures that routes like /v1beta/models/{model_name}:generateContent are properly accessible to internal users and not blocked as admin-only routes. """ - + # Create an internal user object user_obj = LiteLLM_UserTable( user_id="test_user", user_email="test@example.com", user_role=LitellmUserRoles.INTERNAL_USER.value, ) - + # Create an internal user API key auth valid_token = UserAPIKeyAuth( user_id="test_user", user_role=LitellmUserRoles.INTERNAL_USER.value, ) - + # Create a mock request request = MagicMock(spec=Request) request.query_params = {} - + # Test that calling Google route with dynamic model name does NOT raise an exception try: RouteChecks.non_proxy_admin_allowed_routes_check( @@ -263,11 +279,13 @@ def test_virtual_key_allowed_routes_with_multiple_litellm_routes_member_names(): # Test that routes from both groups are allowed result1 = RouteChecks.is_virtual_key_allowed_to_call_route( - route="/chat/completions", valid_token=valid_token # This is in openai_routes + route="/chat/completions", + valid_token=valid_token, # This is in openai_routes ) result2 = RouteChecks.is_virtual_key_allowed_to_call_route( - route="/user/info", valid_token=valid_token # This is in info_routes + route="/user/info", + valid_token=valid_token, # This is in info_routes ) assert result1 is True @@ -288,11 +306,13 @@ def test_virtual_key_allowed_routes_with_mixed_member_names_and_explicit_routes( # Test that both info routes and explicit custom route are allowed result1 = RouteChecks.is_virtual_key_allowed_to_call_route( - route="/user/info", valid_token=valid_token # This is in info_routes + route="/user/info", + valid_token=valid_token, # This is in info_routes ) result2 = RouteChecks.is_virtual_key_allowed_to_call_route( - route="/custom/route", valid_token=valid_token # This is explicitly listed + route="/custom/route", + valid_token=valid_token, # This is explicitly listed ) assert result1 is True @@ -323,7 +343,8 @@ def test_virtual_key_allowed_routes_with_no_member_names_only_explicit(): # Test that non-allowed route raises HTTPException with pytest.raises(HTTPException) as exc_info: RouteChecks.is_virtual_key_allowed_to_call_route( - route="/user/info", valid_token=valid_token # Not in allowed routes + route="/user/info", + valid_token=valid_token, # Not in allowed routes ) assert exc_info.value.status_code == 403 @@ -372,12 +393,15 @@ def test_virtual_key_llm_api_routes_allows_registered_pass_through_endpoints(): }, } - with patch( - "litellm.proxy.pass_through_endpoints.pass_through_endpoints._registered_pass_through_routes", - mock_registered_routes, - ), patch( - "litellm.proxy.pass_through_endpoints.pass_through_endpoints.get_server_root_path", - return_value="/", + with ( + patch( + "litellm.proxy.pass_through_endpoints.pass_through_endpoints._registered_pass_through_routes", + mock_registered_routes, + ), + patch( + "litellm.proxy.pass_through_endpoints.pass_through_endpoints.get_server_root_path", + return_value="/", + ), ): # Create a virtual key with llm_api_routes permission valid_token = UserAPIKeyAuth( @@ -421,12 +445,15 @@ def test_virtual_key_without_llm_api_routes_cannot_access_pass_through(): }, } - with patch( - "litellm.proxy.pass_through_endpoints.pass_through_endpoints._registered_pass_through_routes", - mock_registered_routes, - ), patch( - "litellm.proxy.pass_through_endpoints.pass_through_endpoints.get_server_root_path", - return_value="/", + with ( + patch( + "litellm.proxy.pass_through_endpoints.pass_through_endpoints._registered_pass_through_routes", + mock_registered_routes, + ), + patch( + "litellm.proxy.pass_through_endpoints.pass_through_endpoints.get_server_root_path", + return_value="/", + ), ): # Create a virtual key without llm_api_routes permission valid_token = UserAPIKeyAuth( @@ -442,7 +469,9 @@ def test_virtual_key_without_llm_api_routes_cannot_access_pass_through(): ) assert exc_info.value.status_code == 403 - assert "Virtual key is not allowed to call this route" in str(exc_info.value.detail) + assert "Virtual key is not allowed to call this route" in str( + exc_info.value.detail + ) def test_check_passthrough_route_access_key_metadata_exact_match(): @@ -833,6 +862,7 @@ def test_videos_route_with_virtual_key_llm_api_routes(): result is True ), f"Virtual key with llm_api_routes should be able to access {route}" + def test_non_proxy_admin_wildcard_allowed_routes(): """Test that nonproxy admin users can still use wildcard routes""" @@ -847,7 +877,7 @@ def test_non_proxy_admin_wildcard_allowed_routes(): user_role=LitellmUserRoles.INTERNAL_USER.value, allowed_routes=["/scim/*"], ) - + request = MagicMock(spec=Request) request.query_params = {} @@ -864,14 +894,14 @@ def test_non_proxy_admin_wildcard_allowed_routes(): def test_proxy_admin_viewer_can_access_global_spend_tags(): """ Test that proxy_admin_viewer can access /global/spend/tags endpoint. - + This test verifies the fix for the issue where proxy_admin_viewer was getting 403 errors when trying to access /global/spend/tags endpoint. - + Related: Slack thread from 10/9/2025 - Erik Kristensen reported this issue. proxy_admin_viewer role should have access to "view all spend" endpoints. """ - + # Create a proxy admin viewer user object user_obj = LiteLLM_UserTable( user_id="viewer_user", @@ -912,8 +942,12 @@ def test_route_in_additional_public_routes_wildcard_match(): """ from litellm.proxy.auth.auth_utils import route_in_additonal_public_routes - with patch("litellm.proxy.proxy_server.general_settings", {"public_routes": ["/api/*"]}), \ - patch("litellm.proxy.proxy_server.premium_user", True): + with ( + patch( + "litellm.proxy.proxy_server.general_settings", {"public_routes": ["/api/*"]} + ), + patch("litellm.proxy.proxy_server.premium_user", True), + ): # Wildcard should match subpaths assert route_in_additonal_public_routes("/api/users") is True assert route_in_additonal_public_routes("/api/users/123") is True @@ -927,11 +961,15 @@ def test_route_in_additional_public_routes_exact_match(): """ from litellm.proxy.auth.auth_utils import route_in_additonal_public_routes - with patch("litellm.proxy.proxy_server.general_settings", {"public_routes": ["/health", "/status"]}), \ - patch("litellm.proxy.proxy_server.premium_user", True): + with ( + patch( + "litellm.proxy.proxy_server.general_settings", + {"public_routes": ["/health", "/status"]}, + ), + patch("litellm.proxy.proxy_server.premium_user", True), + ): # Exact matches should work assert route_in_additonal_public_routes("/health") is True assert route_in_additonal_public_routes("/status") is True # Non-matching routes should fail assert route_in_additonal_public_routes("/other") is False - diff --git a/tests/test_litellm/proxy/auth/test_user_api_key_auth.py b/tests/test_litellm/proxy/auth/test_user_api_key_auth.py index 1f379f4371e..9b7b7f46155 100644 --- a/tests/test_litellm/proxy/auth/test_user_api_key_auth.py +++ b/tests/test_litellm/proxy/auth/test_user_api_key_auth.py @@ -41,6 +41,12 @@ def test_get_api_key(): ("Basic sk-12345678", "sk-12345678", "Basic sk-12345678"), ("bearer sk-12345678", "sk-12345678", "bearer sk-12345678"), ("sk-12345678", "sk-12345678", "sk-12345678"), + # AWS Signature V4 format (LangChain AWS SDK) + ( + "AWS4-HMAC-SHA256 Credential=Bearer sk-12345678/20260210/us-east-1/bedrock/aws4_request, SignedHeaders=host, Signature=abc123", + "sk-12345678", + "AWS4-HMAC-SHA256 Credential=Bearer sk-12345678/20260210/us-east-1/bedrock/aws4_request, SignedHeaders=host, Signature=abc123", + ), ], ) def test_get_api_key_with_custom_litellm_key_header( @@ -243,10 +249,10 @@ async def test_proxy_admin_expired_key_from_cache(): Regression test for issue where PROXY_ADMIN keys from cache skipped expiration check. """ from datetime import datetime, timedelta, timezone - + from fastapi import Request from starlette.datastructures import URL - + from litellm.proxy._types import ( LitellmUserRoles, ProxyErrorTypes, @@ -255,7 +261,7 @@ async def test_proxy_admin_expired_key_from_cache(): ) from litellm.proxy.auth.user_api_key_auth import _user_api_key_auth_builder from litellm.proxy.proxy_server import hash_token - + # Create an expired PROXY_ADMIN key api_key = "sk-test-proxy-admin-key" hashed_key = hash_token(api_key) @@ -368,7 +374,7 @@ async def test_return_user_api_key_auth_obj_user_spend_and_budget(): from user_obj attributes. """ from datetime import datetime - + from litellm.proxy._types import UserAPIKeyAuth from litellm.proxy.auth.user_api_key_auth import _return_user_api_key_auth_obj diff --git a/tests/test_litellm/proxy/guardrails/guardrail_hooks/test_presidio.py b/tests/test_litellm/proxy/guardrails/guardrail_hooks/test_presidio.py index 5ec9b13408e..f01c23f7116 100644 --- a/tests/test_litellm/proxy/guardrails/guardrail_hooks/test_presidio.py +++ b/tests/test_litellm/proxy/guardrails/guardrail_hooks/test_presidio.py @@ -6,6 +6,7 @@ Tests PII detection and masking for different message formats import asyncio import os import sys +from contextlib import asynccontextmanager from unittest.mock import MagicMock, patch import pytest @@ -18,10 +19,41 @@ from litellm.proxy._types import UserAPIKeyAuth from litellm.proxy.guardrails.guardrail_hooks.presidio import ( _OPTIONAL_PresidioPIIMasking, ) +from litellm.exceptions import GuardrailRaisedException from litellm.types.guardrails import LitellmParams, PiiAction, PiiEntityType from litellm.types.utils import Choices, Message, ModelResponse +def _make_mock_session_iterator(json_response): + """Create a mock _get_session_iterator that yields a session returning json_response.""" + + @asynccontextmanager + async def mock_iterator(): + class MockResponse: + async def json(self): + return json_response + + async def __aenter__(self): + return self + + async def __aexit__(self, *args): + pass + + class MockSession: + def post(self, *args, **kwargs): + return MockResponse() + + async def __aenter__(self): + return self + + async def __aexit__(self, *args): + pass + + yield MockSession() + + return mock_iterator + + @pytest.fixture def presidio_guardrail(): """Create a Presidio guardrail instance for testing""" @@ -889,39 +921,134 @@ async def test_analyze_text_error_dict_handling(): output_parse_pii=False, ) - # Mock the HTTP response to return error dict - class MockResponse: - async def json(self): - return {"error": "No text provided"} - - async def __aenter__(self): - return self - - async def __aexit__(self, *args): - pass - - class MockSession: - def post(self, *args, **kwargs): - return MockResponse() - - async def __aenter__(self): - return self - - async def __aexit__(self, *args): - pass - - with patch("aiohttp.ClientSession", return_value=MockSession()): + with patch.object( + presidio, + "_get_session_iterator", + _make_mock_session_iterator({"error": "No text provided"}), + ): result = await presidio.analyze_text( text="some text", presidio_config=None, request_data={}, ) - # Should return empty list when error dict is received - assert result == [], "Error dict should be handled gracefully" + assert result == [], "Error dict should be handled gracefully" print("✓ analyze_text error dict handling test passed") +@pytest.mark.asyncio +async def test_analyze_text_string_response_handling(): + """ + Test that analyze_text handles string responses from Presidio API. + + When Presidio returns a string (e.g. error message from websearch/hosted models), + should handle gracefully instead of crashing with TypeError about mapping vs str. + """ + presidio = _OPTIONAL_PresidioPIIMasking( + presidio_analyzer_api_base="http://mock-presidio:5002/", + presidio_anonymizer_api_base="http://mock-presidio:5001/", + output_parse_pii=False, + ) + + with patch.object( + presidio, + "_get_session_iterator", + _make_mock_session_iterator("Internal Server Error"), + ): + result = await presidio.analyze_text( + text="some text", + presidio_config=None, + request_data={}, + ) + assert result == [], "String response should be handled gracefully" + + +@pytest.mark.asyncio +async def test_analyze_text_invalid_response_raises_when_block_configured(): + """ + When pii_entities_config has BLOCK and Presidio returns invalid response, + should raise GuardrailRaisedException (fail-closed) rather than silently allowing content. + """ + presidio = _OPTIONAL_PresidioPIIMasking( + presidio_analyzer_api_base="http://mock-presidio:5002/", + presidio_anonymizer_api_base="http://mock-presidio:5001/", + output_parse_pii=False, + pii_entities_config={PiiEntityType.CREDIT_CARD: PiiAction.BLOCK}, + ) + + with patch.object( + presidio, + "_get_session_iterator", + _make_mock_session_iterator("Internal Server Error"), + ): + with pytest.raises(GuardrailRaisedException) as exc_info: + await presidio.analyze_text( + text="some text", + presidio_config=None, + request_data={}, + ) + assert "BLOCK" in str(exc_info.value) or "Presidio" in str(exc_info.value) + + +@pytest.mark.asyncio +async def test_analyze_text_invalid_response_raises_when_mask_configured(): + """ + When pii_entities_config has MASK and Presidio returns invalid response, + should raise GuardrailRaisedException (fail-closed) because PII masking is expected. + """ + presidio = _OPTIONAL_PresidioPIIMasking( + presidio_analyzer_api_base="http://mock-presidio:5002/", + presidio_anonymizer_api_base="http://mock-presidio:5001/", + output_parse_pii=False, + pii_entities_config={PiiEntityType.CREDIT_CARD: PiiAction.MASK}, + ) + + with patch.object( + presidio, + "_get_session_iterator", + _make_mock_session_iterator("Internal Server Error"), + ): + with pytest.raises(GuardrailRaisedException) as exc_info: + await presidio.analyze_text( + text="some text", + presidio_config=None, + request_data={}, + ) + assert "PII protection is configured" in str(exc_info.value) + + +@pytest.mark.asyncio +async def test_analyze_text_list_with_non_dict_items(): + """ + Test that analyze_text skips non-dict items in the result list. + + When Presidio returns a list containing strings (malformed response), + should skip invalid items and return parsed valid ones. + """ + presidio = _OPTIONAL_PresidioPIIMasking( + presidio_analyzer_api_base="http://mock-presidio:5002/", + presidio_anonymizer_api_base="http://mock-presidio:5001/", + output_parse_pii=False, + ) + + json_response = [ + {"entity_type": "PERSON", "start": 0, "end": 5, "score": 0.9}, + "invalid_string_item", + {"entity_type": "EMAIL", "start": 10, "end": 25, "score": 0.85}, + ] + with patch.object( + presidio, "_get_session_iterator", _make_mock_session_iterator(json_response) + ): + result = await presidio.analyze_text( + text="some text", + presidio_config=None, + request_data={}, + ) + assert len(result) == 2, "Should parse 2 valid dict items and skip the string" + assert result[0].get("entity_type") == "PERSON" + assert result[1].get("entity_type") == "EMAIL" + + @pytest.mark.asyncio async def test_tool_calling_complete_scenario( presidio_guardrail, mock_user_api_key, mock_cache diff --git a/tests/test_litellm/proxy/guardrails/test_guardrail_endpoints.py b/tests/test_litellm/proxy/guardrails/test_guardrail_endpoints.py index 88f56c24067..c0f16c8b953 100644 --- a/tests/test_litellm/proxy/guardrails/test_guardrail_endpoints.py +++ b/tests/test_litellm/proxy/guardrails/test_guardrail_endpoints.py @@ -149,6 +149,111 @@ async def test_list_guardrails_v2_with_db_and_config( assert isinstance(config_guardrail.litellm_params, BaseLitellmParams) +@pytest.mark.asyncio +async def test_list_guardrails_v2_masks_sensitive_data_in_db_guardrails(mocker): + """Test that sensitive litellm_params are masked for DB guardrails in list response""" + db_guardrail_with_secrets = { + "guardrail_id": "secret-db-guardrail", + "guardrail_name": "DB Guardrail with Secrets", + "litellm_params": { + "guardrail": "azure/text_moderations", + "mode": "pre_call", + "api_key": "sk-1234567890abcdef", + "api_base": "https://api.secret.example.com", + }, + "guardrail_info": {"description": "Test guardrail"}, + "created_at": datetime.now(), + "updated_at": datetime.now(), + } + + mock_prisma_client = mocker.Mock() + mock_prisma_client.db = mocker.Mock() + mock_prisma_client.db.litellm_guardrailstable = mocker.Mock() + mock_prisma_client.db.litellm_guardrailstable.find_many = AsyncMock( + return_value=[db_guardrail_with_secrets] + ) + + mock_in_memory_handler = mocker.Mock() + mock_in_memory_handler.list_in_memory_guardrails.return_value = [] + + mocker.patch("litellm.proxy.proxy_server.prisma_client", mock_prisma_client) + mocker.patch( + "litellm.proxy.guardrails.guardrail_registry.IN_MEMORY_GUARDRAIL_HANDLER", + mock_in_memory_handler, + ) + + response = await list_guardrails_v2() + + assert len(response.guardrails) == 1 + guardrail = response.guardrails[0] + litellm_params = guardrail.litellm_params + if isinstance(litellm_params, dict): + params = litellm_params + else: + params = litellm_params.model_dump() if hasattr(litellm_params, "model_dump") else dict(litellm_params) + + # Sensitive keys (containing "key", "secret", "token", etc.) should be masked + assert params["api_key"] != "sk-1234567890abcdef" + assert "****" in str(params["api_key"]) + # Non-sensitive keys should remain unchanged + assert params["guardrail"] == "azure/text_moderations" + assert params["mode"] == "pre_call" + assert params["api_base"] == "https://api.secret.example.com" + + +@pytest.mark.asyncio +async def test_list_guardrails_v2_masks_sensitive_data_in_config_guardrails(mocker): + """Test that sensitive litellm_params are masked for in-memory/config guardrails in list response""" + config_guardrail_with_secrets = { + "guardrail_id": "secret-config-guardrail", + "guardrail_name": "Config Guardrail with Secrets", + "litellm_params": { + "guardrail": "bedrock", + "mode": "during_call", + "api_key": "my-secret-bedrock-key", + "vertex_credentials": "{sensitive_creds}", + }, + "guardrail_info": {"description": "Test guardrail from config"}, + } + + mock_prisma_client = mocker.Mock() + mock_prisma_client.db = mocker.Mock() + mock_prisma_client.db.litellm_guardrailstable = mocker.Mock() + mock_prisma_client.db.litellm_guardrailstable.find_many = AsyncMock( + return_value=[] + ) + + mock_in_memory_handler = mocker.Mock() + mock_in_memory_handler.list_in_memory_guardrails.return_value = [ + config_guardrail_with_secrets + ] + + mocker.patch("litellm.proxy.proxy_server.prisma_client", mock_prisma_client) + mocker.patch( + "litellm.proxy.guardrails.guardrail_registry.IN_MEMORY_GUARDRAIL_HANDLER", + mock_in_memory_handler, + ) + + response = await list_guardrails_v2() + + assert len(response.guardrails) == 1 + guardrail = response.guardrails[0] + litellm_params = guardrail.litellm_params + if isinstance(litellm_params, dict): + params = litellm_params + else: + params = litellm_params.model_dump() if hasattr(litellm_params, "model_dump") else dict(litellm_params) + + # Sensitive keys should be masked + assert params["api_key"] != "my-secret-bedrock-key" + assert "****" in str(params["api_key"]) + assert params["vertex_credentials"] != "{sensitive_creds}" + assert "****" in str(params["vertex_credentials"]) + # Non-sensitive keys should remain unchanged + assert params["guardrail"] == "bedrock" + assert params["mode"] == "during_call" + + @pytest.mark.asyncio async def test_get_guardrail_info_from_db(mocker, mock_prisma_client): """Test getting guardrail info from DB""" diff --git a/tests/test_litellm/proxy/management_endpoints/test_budget_endpoints.py b/tests/test_litellm/proxy/management_endpoints/test_budget_endpoints.py index d8c505223d9..b15b9d622e4 100644 --- a/tests/test_litellm/proxy/management_endpoints/test_budget_endpoints.py +++ b/tests/test_litellm/proxy/management_endpoints/test_budget_endpoints.py @@ -21,6 +21,7 @@ sys.path.insert( def client_and_mocks(monkeypatch): # Setup MagicMock Prisma mock_prisma = MagicMock() + mock_table = MagicMock() mock_table.create = AsyncMock(side_effect=lambda *, data: data) mock_table.update = AsyncMock(side_effect=lambda *, where, data: {**where, **data}) diff --git a/tests/test_litellm/proxy/management_endpoints/test_mcp_management_endpoints.py b/tests/test_litellm/proxy/management_endpoints/test_mcp_management_endpoints.py index 83914e30354..b4c8d5f07cc 100644 --- a/tests/test_litellm/proxy/management_endpoints/test_mcp_management_endpoints.py +++ b/tests/test_litellm/proxy/management_endpoints/test_mcp_management_endpoints.py @@ -1,4 +1,3 @@ -import json import os import sys import types @@ -204,25 +203,28 @@ class TestListMCPServers: transport="http", ), ] - mock_manager.get_all_allowed_mcp_servers = AsyncMock( - return_value=mock_servers - ) + mock_manager.get_all_allowed_mcp_servers = AsyncMock(return_value=mock_servers) for idx, server in enumerate(mock_servers): server.credentials = {"auth_value": f"secret_{idx}"} - with patch( - "litellm.proxy.management_endpoints.mcp_management_endpoints.global_mcp_server_manager", - mock_manager, - ), patch( - "litellm.proxy.management_endpoints.mcp_management_endpoints._user_has_admin_view", - return_value=True, - ), patch( - "litellm.proxy.management_endpoints.mcp_management_endpoints.get_prisma_client_or_throw", - return_value=mock_prisma_client, - ), patch( - "litellm.proxy.management_endpoints.mcp_management_endpoints.build_effective_auth_contexts", - AsyncMock(return_value=[mock_user_auth]), + with ( + patch( + "litellm.proxy.management_endpoints.mcp_management_endpoints.global_mcp_server_manager", + mock_manager, + ), + patch( + "litellm.proxy.management_endpoints.mcp_management_endpoints._user_has_admin_view", + return_value=True, + ), + patch( + "litellm.proxy.management_endpoints.mcp_management_endpoints.get_prisma_client_or_throw", + return_value=mock_prisma_client, + ), + patch( + "litellm.proxy.management_endpoints.mcp_management_endpoints.build_effective_auth_contexts", + AsyncMock(return_value=[mock_user_auth]), + ), ): # Import and call the function from litellm.proxy.management_endpoints.mcp_management_endpoints import ( @@ -269,12 +271,15 @@ class TestListMCPServers: return_value=mock_servers ) - with patch( - "litellm.proxy.management_endpoints.mcp_management_endpoints._get_user_mcp_management_mode", - return_value="view_all", - ), patch( - "litellm.proxy.management_endpoints.mcp_management_endpoints.global_mcp_server_manager", - mock_manager, + with ( + patch( + "litellm.proxy.management_endpoints.mcp_management_endpoints._get_user_mcp_management_mode", + return_value="view_all", + ), + patch( + "litellm.proxy.management_endpoints.mcp_management_endpoints.global_mcp_server_manager", + mock_manager, + ), ): from litellm.proxy.management_endpoints.mcp_management_endpoints import ( fetch_all_mcp_servers, @@ -285,6 +290,79 @@ class TestListMCPServers: assert len(result) == 2 assert {server.server_id for server in result} == {"server-1", "server-2"} + @pytest.mark.asyncio + async def test_list_mcp_servers_view_all_mode_virtual_key_is_sanitized(self): + """Issue #20325: virtual keys should get a safe discovery view.""" + + mock_user_auth = UserAPIKeyAuth( + user_role=LitellmUserRoles.INTERNAL_USER, + user_id="test_user_id", + api_key="test_api_key", + allowed_routes=["mcp_routes"], + ) + + mock_servers = [ + generate_mock_mcp_server_db_record(server_id="server-1", alias="One"), + generate_mock_mcp_server_db_record(server_id="server-2", alias="Two"), + ] + for idx, server in enumerate(mock_servers): + server.credentials = {"auth_value": f"secret_{idx}"} + server.env = {"API_KEY": "super-secret"} + server.static_headers = {"Authorization": "Bearer super-secret"} + server.mcp_access_groups = ["group-a"] + server.teams = [{"team_id": "team-1", "team_alias": "Team 1"}] + server.command = "bash" + server.args = ["-lc", "echo hi"] + server.extra_headers = ["Authorization"] + + mock_manager = MagicMock() + mock_manager.get_all_mcp_servers_unfiltered = AsyncMock( + return_value=mock_servers + ) + mock_manager.get_all_allowed_mcp_servers = AsyncMock(return_value=mock_servers) + + with ( + patch( + "litellm.proxy.management_endpoints.mcp_management_endpoints._get_user_mcp_management_mode", + return_value="view_all", + ), + patch( + "litellm.proxy.management_endpoints.mcp_management_endpoints.global_mcp_server_manager", + mock_manager, + ), + patch( + "litellm.proxy.management_endpoints.mcp_management_endpoints.get_prisma_client_or_throw", + return_value=MagicMock(), + ), + patch( + "litellm.proxy.management_endpoints.mcp_management_endpoints.build_effective_auth_contexts", + AsyncMock(return_value=[mock_user_auth]), + ), + ): + from litellm.proxy.management_endpoints.mcp_management_endpoints import ( + fetch_all_mcp_servers, + ) + + result = await fetch_all_mcp_servers(user_api_key_dict=mock_user_auth) + + # Ensure we did not bypass filtering via view_all for restricted virtual keys. + mock_manager.get_all_mcp_servers_unfiltered.assert_not_called() + + assert len(result) == 2 + assert {server.server_id for server in result} == {"server-1", "server-2"} + + for server in result: + assert server.credentials is None + assert server.url is None + assert server.static_headers is None + assert server.env == {} + assert server.command is None + assert server.args == [] + assert server.extra_headers == [] + assert server.allowed_tools == [] + assert server.mcp_access_groups == [] + assert server.teams == [] + @pytest.mark.asyncio async def test_list_mcp_servers_combined_config_and_db(self): """ @@ -374,25 +452,28 @@ class TestListMCPServers: transport="http", ), ] - mock_manager.get_all_allowed_mcp_servers = AsyncMock( - return_value=mock_servers - ) + mock_manager.get_all_allowed_mcp_servers = AsyncMock(return_value=mock_servers) for idx, server in enumerate(mock_servers): server.credentials = {"auth_value": f"secret_{idx}"} - with patch( - "litellm.proxy.management_endpoints.mcp_management_endpoints.global_mcp_server_manager", - mock_manager, - ), patch( - "litellm.proxy.management_endpoints.mcp_management_endpoints._user_has_admin_view", - return_value=True, - ), patch( - "litellm.proxy.management_endpoints.mcp_management_endpoints.get_prisma_client_or_throw", - return_value=mock_prisma_client, - ), patch( - "litellm.proxy.management_endpoints.mcp_management_endpoints.build_effective_auth_contexts", - AsyncMock(return_value=[mock_user_auth]), + with ( + patch( + "litellm.proxy.management_endpoints.mcp_management_endpoints.global_mcp_server_manager", + mock_manager, + ), + patch( + "litellm.proxy.management_endpoints.mcp_management_endpoints._user_has_admin_view", + return_value=True, + ), + patch( + "litellm.proxy.management_endpoints.mcp_management_endpoints.get_prisma_client_or_throw", + return_value=mock_prisma_client, + ), + patch( + "litellm.proxy.management_endpoints.mcp_management_endpoints.build_effective_auth_contexts", + AsyncMock(return_value=[mock_user_auth]), + ), ): # Import and call the function from litellm.proxy.management_endpoints.mcp_management_endpoints import ( @@ -494,25 +575,28 @@ class TestListMCPServers: url="https://actions.zapier.com/mcp/sse", ), ] - mock_manager.get_all_allowed_mcp_servers = AsyncMock( - return_value=mock_servers - ) + mock_manager.get_all_allowed_mcp_servers = AsyncMock(return_value=mock_servers) for idx, server in enumerate(mock_servers): server.credentials = {"auth_value": f"secret_{idx}"} - with patch( - "litellm.proxy.management_endpoints.mcp_management_endpoints.global_mcp_server_manager", - mock_manager, - ), patch( - "litellm.proxy.management_endpoints.mcp_management_endpoints._user_has_admin_view", - return_value=False, - ), patch( - "litellm.proxy.management_endpoints.mcp_management_endpoints.get_prisma_client_or_throw", - return_value=mock_prisma_client, - ), patch( - "litellm.proxy.management_endpoints.mcp_management_endpoints.build_effective_auth_contexts", - AsyncMock(return_value=[mock_user_auth]), + with ( + patch( + "litellm.proxy.management_endpoints.mcp_management_endpoints.global_mcp_server_manager", + mock_manager, + ), + patch( + "litellm.proxy.management_endpoints.mcp_management_endpoints._user_has_admin_view", + return_value=False, + ), + patch( + "litellm.proxy.management_endpoints.mcp_management_endpoints.get_prisma_client_or_throw", + return_value=mock_prisma_client, + ), + patch( + "litellm.proxy.management_endpoints.mcp_management_endpoints.build_effective_auth_contexts", + AsyncMock(return_value=[mock_user_auth]), + ), ): # Import and call the function from litellm.proxy.management_endpoints.mcp_management_endpoints import ( @@ -540,7 +624,6 @@ class TestListMCPServers: assert server.alias == "Allowed Zapier MCP" assert server.url == "https://actions.zapier.com/mcp/sse" - @pytest.mark.asyncio async def test_fetch_single_mcp_server_redacts_credentials(self): mock_server = generate_mock_mcp_server_db_record( @@ -562,18 +645,23 @@ class TestListMCPServers: user_role=LitellmUserRoles.PROXY_ADMIN ) - with patch( - "litellm.proxy.management_endpoints.mcp_management_endpoints.get_prisma_client_or_throw", - return_value=mock_prisma_client, - ), patch( - "litellm.proxy.management_endpoints.mcp_management_endpoints.get_mcp_server", - AsyncMock(return_value=mock_server), - ), patch( - "litellm.proxy.management_endpoints.mcp_management_endpoints.global_mcp_server_manager.health_check_server", - AsyncMock(return_value=mock_health_result), - ), patch( - "litellm.proxy.management_endpoints.mcp_management_endpoints._user_has_admin_view", - return_value=True, + with ( + patch( + "litellm.proxy.management_endpoints.mcp_management_endpoints.get_prisma_client_or_throw", + return_value=mock_prisma_client, + ), + patch( + "litellm.proxy.management_endpoints.mcp_management_endpoints.get_mcp_server", + AsyncMock(return_value=mock_server), + ), + patch( + "litellm.proxy.management_endpoints.mcp_management_endpoints.global_mcp_server_manager.health_check_server", + AsyncMock(return_value=mock_health_result), + ), + patch( + "litellm.proxy.management_endpoints.mcp_management_endpoints._user_has_admin_view", + return_value=True, + ), ): from litellm.proxy.management_endpoints.mcp_management_endpoints import ( fetch_mcp_server, @@ -610,18 +698,23 @@ class TestListMCPServers: user_role=LitellmUserRoles.PROXY_ADMIN ) - with patch( - "litellm.proxy.management_endpoints.mcp_management_endpoints.get_prisma_client_or_throw", - return_value=mock_prisma_client, - ), patch( - "litellm.proxy.management_endpoints.mcp_management_endpoints.get_mcp_server", - AsyncMock(return_value=mock_server), - ), patch( - "litellm.proxy.management_endpoints.mcp_management_endpoints.global_mcp_server_manager.health_check_server", - AsyncMock(return_value=mock_health_result), - ), patch( - "litellm.proxy.management_endpoints.mcp_management_endpoints._user_has_admin_view", - return_value=True, + with ( + patch( + "litellm.proxy.management_endpoints.mcp_management_endpoints.get_prisma_client_or_throw", + return_value=mock_prisma_client, + ), + patch( + "litellm.proxy.management_endpoints.mcp_management_endpoints.get_mcp_server", + AsyncMock(return_value=mock_server), + ), + patch( + "litellm.proxy.management_endpoints.mcp_management_endpoints.global_mcp_server_manager.health_check_server", + AsyncMock(return_value=mock_health_result), + ), + patch( + "litellm.proxy.management_endpoints.mcp_management_endpoints._user_has_admin_view", + return_value=True, + ), ): from litellm.proxy.management_endpoints.mcp_management_endpoints import ( fetch_mcp_server, @@ -763,16 +856,20 @@ class TestTemporaryMCPSessionEndpoints: mock_manager.get_mcp_server_by_id.return_value = inherited_server mock_manager.build_mcp_server_from_table = AsyncMock(return_value=built_server) - with patch( - "litellm.proxy.management_endpoints.mcp_management_endpoints.validate_and_normalize_mcp_server_payload", - MagicMock(), - ) as validate_mock, patch( - "litellm.proxy.management_endpoints.mcp_management_endpoints.global_mcp_server_manager", - mock_manager, - ), patch( - "litellm.proxy.management_endpoints.mcp_management_endpoints._cache_temporary_mcp_server", - MagicMock(), - ) as cache_mock: + with ( + patch( + "litellm.proxy.management_endpoints.mcp_management_endpoints.validate_and_normalize_mcp_server_payload", + MagicMock(), + ) as validate_mock, + patch( + "litellm.proxy.management_endpoints.mcp_management_endpoints.global_mcp_server_manager", + mock_manager, + ), + patch( + "litellm.proxy.management_endpoints.mcp_management_endpoints._cache_temporary_mcp_server", + MagicMock(), + ) as cache_mock, + ): response = await add_session_mcp_server( payload=payload, user_api_key_dict=user_auth, @@ -832,13 +929,16 @@ class TestTemporaryMCPSessionEndpoints: server = generate_mock_mcp_server_config_record(server_id="server-1") authorize_response = MagicMock() - with patch( - "litellm.proxy.management_endpoints.mcp_management_endpoints._get_cached_temporary_mcp_server_or_404", - return_value=server, - ) as get_server, patch( - "litellm.proxy.management_endpoints.mcp_management_endpoints.authorize_with_server", - AsyncMock(return_value=authorize_response), - ) as authorize_mock: + with ( + patch( + "litellm.proxy.management_endpoints.mcp_management_endpoints._get_cached_temporary_mcp_server_or_404", + return_value=server, + ) as get_server, + patch( + "litellm.proxy.management_endpoints.mcp_management_endpoints.authorize_with_server", + AsyncMock(return_value=authorize_response), + ) as authorize_mock, + ): result = await mcp_authorize( request=request, server_id="server-1", @@ -875,13 +975,16 @@ class TestTemporaryMCPSessionEndpoints: server = generate_mock_mcp_server_config_record(server_id="server-1") exchange_response = {"access_token": "token"} - with patch( - "litellm.proxy.management_endpoints.mcp_management_endpoints._get_cached_temporary_mcp_server_or_404", - return_value=server, - ) as get_server, patch( - "litellm.proxy.management_endpoints.mcp_management_endpoints.exchange_token_with_server", - AsyncMock(return_value=exchange_response), - ) as exchange_mock: + with ( + patch( + "litellm.proxy.management_endpoints.mcp_management_endpoints._get_cached_temporary_mcp_server_or_404", + return_value=server, + ) as get_server, + patch( + "litellm.proxy.management_endpoints.mcp_management_endpoints.exchange_token_with_server", + AsyncMock(return_value=exchange_response), + ) as exchange_mock, + ): result = await mcp_token( request=request, server_id="server-1", @@ -922,16 +1025,20 @@ class TestTemporaryMCPSessionEndpoints: "token_endpoint_auth_method": "client_secret_basic", } - with patch( - "litellm.proxy.management_endpoints.mcp_management_endpoints._get_cached_temporary_mcp_server_or_404", - return_value=server, - ) as get_server, patch( - "litellm.proxy.management_endpoints.mcp_management_endpoints._read_request_body", - AsyncMock(return_value=request_body), - ) as read_body, patch( - "litellm.proxy.management_endpoints.mcp_management_endpoints.register_client_with_server", - AsyncMock(return_value=register_response), - ) as register_mock: + with ( + patch( + "litellm.proxy.management_endpoints.mcp_management_endpoints._get_cached_temporary_mcp_server_or_404", + return_value=server, + ) as get_server, + patch( + "litellm.proxy.management_endpoints.mcp_management_endpoints._read_request_body", + AsyncMock(return_value=request_body), + ) as read_body, + patch( + "litellm.proxy.management_endpoints.mcp_management_endpoints.register_client_with_server", + AsyncMock(return_value=register_response), + ) as register_mock, + ): result = await mcp_register(request=request, server_id="server-1") assert result is register_response @@ -947,6 +1054,7 @@ class TestTemporaryMCPSessionEndpoints: fallback_client_id="server-1", ) + class TestUpdateMCPServer: """Test suite for update MCP server functionality""" @@ -954,7 +1062,7 @@ class TestUpdateMCPServer: async def test_update_mcp_server_respects_extra_headers(self): """ Test that updating an MCP server with extra_headers properly saves the field. - + This test ensures that extra_headers field in UpdateMCPServerRequest is properly handled and persisted when updating an MCP server. """ @@ -999,21 +1107,27 @@ class TestUpdateMCPServer: ) # Mock the update_mcp_server function to capture the call - with patch( - "litellm.proxy.management_endpoints.mcp_management_endpoints.get_prisma_client_or_throw", - return_value=mock_prisma_client, - ), patch( - "litellm.proxy.management_endpoints.mcp_management_endpoints.validate_and_normalize_mcp_server_payload", - MagicMock(), - ), patch( - "litellm.proxy.management_endpoints.mcp_management_endpoints.update_mcp_server", - AsyncMock(return_value=updated_server), - ) as update_mock, patch( - "litellm.proxy.management_endpoints.mcp_management_endpoints.global_mcp_server_manager.add_server", - AsyncMock(), - ), patch( - "litellm.proxy.management_endpoints.mcp_management_endpoints.global_mcp_server_manager.reload_servers_from_database", - AsyncMock(), + with ( + patch( + "litellm.proxy.management_endpoints.mcp_management_endpoints.get_prisma_client_or_throw", + return_value=mock_prisma_client, + ), + patch( + "litellm.proxy.management_endpoints.mcp_management_endpoints.validate_and_normalize_mcp_server_payload", + MagicMock(), + ), + patch( + "litellm.proxy.management_endpoints.mcp_management_endpoints.update_mcp_server", + AsyncMock(return_value=updated_server), + ) as update_mock, + patch( + "litellm.proxy.management_endpoints.mcp_management_endpoints.global_mcp_server_manager.add_server", + AsyncMock(), + ), + patch( + "litellm.proxy.management_endpoints.mcp_management_endpoints.global_mcp_server_manager.reload_servers_from_database", + AsyncMock(), + ), ): # Import and call the function from litellm.proxy.management_endpoints.mcp_management_endpoints import ( @@ -1030,7 +1144,10 @@ class TestUpdateMCPServer: # First arg is prisma_client, second is the payload (UpdateMCPServerRequest) called_payload = call_args[0][1] assert called_payload.server_id == "test-server-1" - assert called_payload.extra_headers == ["X-Custom-Header", "X-Another-Header"] + assert called_payload.extra_headers == [ + "X-Custom-Header", + "X-Another-Header", + ] assert called_payload.alias == "Updated Test Server" # Verify the result includes extra_headers @@ -1081,12 +1198,15 @@ class TestHealthCheckServers: return_value=[mock_health_result_1, mock_health_result_2] ) - with patch( - "litellm.proxy.management_endpoints.mcp_management_endpoints.global_mcp_server_manager", - mock_manager, - ), patch( - "litellm.proxy.management_endpoints.mcp_management_endpoints.build_effective_auth_contexts", - AsyncMock(return_value=[mock_user_auth]), + with ( + patch( + "litellm.proxy.management_endpoints.mcp_management_endpoints.global_mcp_server_manager", + mock_manager, + ), + patch( + "litellm.proxy.management_endpoints.mcp_management_endpoints.build_effective_auth_contexts", + AsyncMock(return_value=[mock_user_auth]), + ), ): result = await health_check_servers( server_ids=None, @@ -1133,9 +1253,12 @@ class TestMCPRegistryEndpoint: # The registry endpoint uses get_filtered_registry (filters by client IP) mock_manager.get_filtered_registry.return_value = {mock_server.server_id: mock_server} - with patch_proxy_general_settings({"enable_mcp_registry": True}), patch( - "litellm.proxy.management_endpoints.mcp_management_endpoints.global_mcp_server_manager", - mock_manager, + with ( + patch_proxy_general_settings({"enable_mcp_registry": True}), + patch( + "litellm.proxy.management_endpoints.mcp_management_endpoints.global_mcp_server_manager", + mock_manager, + ), ): response = client.get("/v1/mcp/registry.json") @@ -1182,12 +1305,15 @@ class TestMCPRegistryEndpoint: return_value=[mock_health_result] ) - with patch( - "litellm.proxy.management_endpoints.mcp_management_endpoints.global_mcp_server_manager", - mock_manager, - ), patch( - "litellm.proxy.management_endpoints.mcp_management_endpoints.build_effective_auth_contexts", - AsyncMock(return_value=[mock_user_auth]), + with ( + patch( + "litellm.proxy.management_endpoints.mcp_management_endpoints.global_mcp_server_manager", + mock_manager, + ), + patch( + "litellm.proxy.management_endpoints.mcp_management_endpoints.build_effective_auth_contexts", + AsyncMock(return_value=[mock_user_auth]), + ), ): result = await health_check_servers( server_ids=["server-1"], @@ -1245,12 +1371,15 @@ class TestManagementPayloadValidation: return_value=[health_result_one, health_result_two] ) - with patch( - "litellm.proxy.management_endpoints.mcp_management_endpoints._get_user_mcp_management_mode", - return_value="view_all", - ), patch( - "litellm.proxy.management_endpoints.mcp_management_endpoints.global_mcp_server_manager", - mock_manager, + with ( + patch( + "litellm.proxy.management_endpoints.mcp_management_endpoints._get_user_mcp_management_mode", + return_value="view_all", + ), + patch( + "litellm.proxy.management_endpoints.mcp_management_endpoints.global_mcp_server_manager", + mock_manager, + ), ): result = await health_check_servers( server_ids=None, @@ -1295,12 +1424,15 @@ class TestManagementPayloadValidation: return_value=[mock_health_result] # Only server-1 is returned (accessible) ) - with patch( - "litellm.proxy.management_endpoints.mcp_management_endpoints.global_mcp_server_manager", - mock_manager, - ), patch( - "litellm.proxy.management_endpoints.mcp_management_endpoints.build_effective_auth_contexts", - AsyncMock(return_value=[mock_user_auth]), + with ( + patch( + "litellm.proxy.management_endpoints.mcp_management_endpoints.global_mcp_server_manager", + mock_manager, + ), + patch( + "litellm.proxy.management_endpoints.mcp_management_endpoints.build_effective_auth_contexts", + AsyncMock(return_value=[mock_user_auth]), + ), ): result = await health_check_servers( server_ids=["server-1", "server-unauthorized"], diff --git a/tests/test_litellm/proxy/management_endpoints/test_ui_sso.py b/tests/test_litellm/proxy/management_endpoints/test_ui_sso.py index 16f80826798..74d36c0acac 100644 --- a/tests/test_litellm/proxy/management_endpoints/test_ui_sso.py +++ b/tests/test_litellm/proxy/management_endpoints/test_ui_sso.py @@ -2374,47 +2374,6 @@ class TestProcessSSOJWTAccessToken: "groups": ["team1", "team2", "team3"], } - def test_process_sso_jwt_access_token_with_valid_token( - self, mock_jwt_handler, sample_jwt_token, sample_jwt_payload - ): - """Test processing a valid JWT access token with team extraction""" - from litellm.proxy.management_endpoints.ui_sso import ( - process_sso_jwt_access_token, - ) - - # Create a result object without team_ids - result = CustomOpenID( - id="test_user", - email="test@example.com", - first_name="Test", - last_name="User", - display_name="Test User", - provider="generic", - team_ids=[], - ) - - with patch("jwt.decode", return_value=sample_jwt_payload) as mock_jwt_decode: - # Act - process_sso_jwt_access_token( - access_token_str=sample_jwt_token, - sso_jwt_handler=mock_jwt_handler, - result=result, - ) - - # Assert - # Verify JWT was decoded correctly - mock_jwt_decode.assert_called_once_with( - sample_jwt_token, options={"verify_signature": False} - ) - - # Verify team IDs were extracted from JWT - mock_jwt_handler.get_team_ids_from_jwt.assert_called_once_with( - sample_jwt_payload - ) - - # Verify team IDs were set on the result object - assert result.team_ids == ["team1", "team2", "team3"] - def test_process_sso_jwt_access_token_with_existing_team_ids( self, mock_jwt_handler, sample_jwt_token ): @@ -2549,27 +2508,6 @@ class TestProcessSSOJWTAccessToken: mock_jwt_handler.get_team_ids_from_jwt.assert_not_called() assert result.team_ids == [] - def test_process_sso_jwt_access_token_no_sso_jwt_handler(self, sample_jwt_token): - """Test that JWT is decoded for role extraction even when sso_jwt_handler is None, - but team_ids are not extracted (team extraction requires sso_jwt_handler).""" - from litellm.proxy.management_endpoints.ui_sso import ( - process_sso_jwt_access_token, - ) - - result = CustomOpenID(id="test_user", email="test@example.com", team_ids=[]) - - mock_payload = {"sub": "test_user", "email": "test@example.com"} - with patch("jwt.decode", return_value=mock_payload) as mock_jwt_decode: - # Act - process_sso_jwt_access_token( - access_token_str=sample_jwt_token, sso_jwt_handler=None, result=result - ) - - # JWT is decoded (for role extraction) but team_ids are not extracted - mock_jwt_decode.assert_called_once() - assert result.team_ids == [] - assert result.user_role is None - def test_process_sso_jwt_access_token_no_result( self, mock_jwt_handler, sample_jwt_token ): @@ -2590,10 +2528,12 @@ class TestProcessSSOJWTAccessToken: mock_jwt_decode.assert_not_called() mock_jwt_handler.get_team_ids_from_jwt.assert_not_called() - def test_process_sso_jwt_access_token_jwt_decode_exception( + def test_process_sso_jwt_access_token_non_decode_exception_propagates( self, mock_jwt_handler, sample_jwt_token ): - """Test that JWT decode exceptions are not caught (should propagate up)""" + """Test that non-DecodeError JWT exceptions still propagate up.""" + import jwt as pyjwt + from litellm.proxy.management_endpoints.ui_sso import ( process_sso_jwt_access_token, ) @@ -2601,19 +2541,16 @@ class TestProcessSSOJWTAccessToken: result = CustomOpenID(id="test_user", email="test@example.com", team_ids=[]) with patch( - "jwt.decode", side_effect=Exception("JWT decode error") + "jwt.decode", side_effect=pyjwt.exceptions.InvalidKeyError("Invalid key") ) as mock_jwt_decode: - # Act & Assert - with pytest.raises(Exception, match="JWT decode error"): + with pytest.raises(pyjwt.exceptions.InvalidKeyError, match="Invalid key"): process_sso_jwt_access_token( access_token_str=sample_jwt_token, sso_jwt_handler=mock_jwt_handler, result=result, ) - # Verify JWT decode was attempted mock_jwt_decode.assert_called_once() - # But team extraction should not have been called mock_jwt_handler.get_team_ids_from_jwt.assert_not_called() def test_process_sso_jwt_access_token_empty_team_ids_from_jwt( @@ -2646,6 +2583,124 @@ class TestProcessSSOJWTAccessToken: # Even empty team IDs should be set assert result.team_ids == [] + def test_process_sso_jwt_access_token_with_opaque_token(self, mock_jwt_handler): + """Test that opaque (non-JWT) access tokens are handled gracefully without raising.""" + from litellm.proxy.management_endpoints.ui_sso import ( + process_sso_jwt_access_token, + ) + + result = CustomOpenID( + id="test_user", + email="test@example.com", + first_name="Test", + last_name="User", + display_name="Test User", + provider="generic", + team_ids=["existing_team"], + user_role=None, + ) + + # Opaque tokens like those from Logto are short random strings, not JWTs + opaque_token = "uTxyjXbS_random_opaque_token_string" + + # Should NOT raise - opaque tokens should be silently skipped + process_sso_jwt_access_token( + access_token_str=opaque_token, + sso_jwt_handler=mock_jwt_handler, + result=result, + ) + + # Result should be untouched + mock_jwt_handler.get_team_ids_from_jwt.assert_not_called() + assert result.team_ids == ["existing_team"] + assert result.user_role is None + + def test_process_sso_jwt_access_token_real_jwt_with_role_and_teams( + self, mock_jwt_handler + ): + """Test that a real JWT containing role and team fields is correctly processed.""" + import jwt as pyjwt + + from litellm.proxy.management_endpoints.ui_sso import ( + process_sso_jwt_access_token, + ) + + payload = { + "sub": "user123", + "email": "admin@example.com", + "role": "proxy_admin", + "groups": ["team_alpha", "team_beta"], + } + real_jwt_token = pyjwt.encode(payload, "test-secret", algorithm="HS256") + + mock_jwt_handler.get_team_ids_from_jwt.return_value = [ + "team_alpha", + "team_beta", + ] + + result = CustomOpenID( + id="user123", + email="admin@example.com", + first_name="Admin", + last_name="User", + display_name="Admin User", + provider="generic", + team_ids=[], + user_role=None, + ) + + process_sso_jwt_access_token( + access_token_str=real_jwt_token, + sso_jwt_handler=mock_jwt_handler, + result=result, + ) + + # Team IDs should be extracted via sso_jwt_handler + mock_jwt_handler.get_team_ids_from_jwt.assert_called_once_with(payload) + assert result.team_ids == ["team_alpha", "team_beta"] + + # Role should be extracted from the "role" field in the JWT + from litellm.proxy._types import LitellmUserRoles + + assert result.user_role == LitellmUserRoles.PROXY_ADMIN + + def test_process_sso_jwt_access_token_real_jwt_without_role_and_teams(self): + """Test that a real JWT without role/team fields leaves result unchanged.""" + import jwt as pyjwt + + from litellm.proxy.management_endpoints.ui_sso import ( + process_sso_jwt_access_token, + ) + + payload = { + "sub": "user456", + "email": "plain@example.com", + "iat": 1700000000, + } + real_jwt_token = pyjwt.encode(payload, "test-secret", algorithm="HS256") + + result = CustomOpenID( + id="user456", + email="plain@example.com", + first_name="Plain", + last_name="User", + display_name="Plain User", + provider="generic", + team_ids=[], + user_role=None, + ) + + # No sso_jwt_handler, no role/team fields in JWT + process_sso_jwt_access_token( + access_token_str=real_jwt_token, + sso_jwt_handler=None, + result=result, + ) + + # Nothing should be modified + assert result.team_ids == [] + assert result.user_role is None + @pytest.mark.asyncio async def test_get_ui_settings_includes_api_doc_base_url(): @@ -4071,3 +4126,123 @@ def test_process_sso_jwt_access_token_with_role_mappings(): # Should get highest privilege role assert result.user_role == LitellmUserRoles.PROXY_ADMIN + +def test_generic_response_convertor_with_extra_attributes(monkeypatch): + """Test that extra attributes are extracted when GENERIC_USER_EXTRA_ATTRIBUTES is set""" + from litellm.proxy.management_endpoints.ui_sso import generic_response_convertor + + monkeypatch.setenv("GENERIC_CLIENT_ID", "test_client") + monkeypatch.setenv("GENERIC_USER_EXTRA_ATTRIBUTES", "custom_field1,custom_field2,custom_field3") + + mock_response = { + "sub": "user-id-123", + "email": "user@example.com", + "given_name": "John", + "family_name": "Doe", + "name": "John Doe", + "provider": "generic", + "custom_field1": "value1", + "custom_field2": ["item1", "item2"], + "custom_field3": {"nested": "data"}, + } + + mock_jwt_handler = MagicMock(spec=JWTHandler) + mock_jwt_handler.get_team_ids_from_jwt.return_value = [] + + result = generic_response_convertor( + response=mock_response, + jwt_handler=mock_jwt_handler, + sso_jwt_handler=None, + role_mappings=None, + ) + + assert result.extra_fields is not None + assert result.extra_fields["custom_field1"] == "value1" + assert result.extra_fields["custom_field2"] == ["item1", "item2"] + assert result.extra_fields["custom_field3"] == {"nested": "data"} + +def test_generic_response_convertor_without_extra_attributes(monkeypatch): + """Test backward compatibility - extra_fields is None when env var not set""" + from litellm.proxy.management_endpoints.ui_sso import generic_response_convertor + + monkeypatch.setenv("GENERIC_CLIENT_ID", "test_client") + # Don't set GENERIC_USER_EXTRA_ATTRIBUTES + + mock_response = { + "sub": "user-id-123", + "email": "user@example.com", + "given_name": "John", + "family_name": "Doe", + "name": "John Doe", + "provider": "generic", + "custom_field1": "value1", + "custom_field2": "value2", + } + + mock_jwt_handler = MagicMock(spec=JWTHandler) + mock_jwt_handler.get_team_ids_from_jwt.return_value = [] + + result = generic_response_convertor( + response=mock_response, + jwt_handler=mock_jwt_handler, + sso_jwt_handler=None, + role_mappings=None, + ) + + assert result.extra_fields is None + +def test_generic_response_convertor_extra_attributes_with_nested_paths(monkeypatch): + """Test that nested paths work with dot notation""" + from litellm.proxy.management_endpoints.ui_sso import generic_response_convertor + + monkeypatch.setenv("GENERIC_CLIENT_ID", "test_client") + monkeypatch.setenv("GENERIC_USER_EXTRA_ATTRIBUTES", "org_info.department,org_info.manager") + + mock_response = { + "sub": "user-id-123", + "email": "user@example.com", + "org_info": { + "department": "Engineering", + "manager": "Jane Smith" + } + } + + mock_jwt_handler = MagicMock(spec=JWTHandler) + mock_jwt_handler.get_team_ids_from_jwt.return_value = [] + + result = generic_response_convertor( + response=mock_response, + jwt_handler=mock_jwt_handler, + sso_jwt_handler=None, + role_mappings=None, + ) + + assert result.extra_fields is not None + assert result.extra_fields["org_info.department"] == "Engineering" + assert result.extra_fields["org_info.manager"] == "Jane Smith" + +def test_generic_response_convertor_extra_attributes_missing_field(monkeypatch): + """Test that missing fields return None""" + from litellm.proxy.management_endpoints.ui_sso import generic_response_convertor + + monkeypatch.setenv("GENERIC_CLIENT_ID", "test_client") + monkeypatch.setenv("GENERIC_USER_EXTRA_ATTRIBUTES", "missing_field,another_missing") + + mock_response = { + "sub": "user-id-123", + "email": "user@example.com", + } + + mock_jwt_handler = MagicMock(spec=JWTHandler) + mock_jwt_handler.get_team_ids_from_jwt.return_value = [] + + result = generic_response_convertor( + response=mock_response, + jwt_handler=mock_jwt_handler, + sso_jwt_handler=None, + role_mappings=None, + ) + + assert result.extra_fields is not None + assert result.extra_fields["missing_field"] is None + assert result.extra_fields["another_missing"] is None \ No newline at end of file diff --git a/tests/test_litellm/proxy/pass_through_endpoints/test_pass_through_endpoints.py b/tests/test_litellm/proxy/pass_through_endpoints/test_pass_through_endpoints.py index daae6d465a7..e50e10352e2 100644 --- a/tests/test_litellm/proxy/pass_through_endpoints/test_pass_through_endpoints.py +++ b/tests/test_litellm/proxy/pass_through_endpoints/test_pass_through_endpoints.py @@ -1316,6 +1316,133 @@ async def test_delete_pass_through_endpoint_not_found(): assert "not found" in str(exc_info.value.detail).lower() +@pytest.mark.asyncio +async def test_get_pass_through_endpoints_includes_config_and_db(): + """ + Test that get_pass_through_endpoints returns both config-defined and DB endpoints, + with correct is_from_config flag. Config-only endpoints have is_from_config=True, + DB endpoints have is_from_config=False. When same path exists in both, DB overrides. + """ + from litellm.proxy._types import ( + PassThroughEndpointResponse, + PassThroughGenericEndpoint, + UserAPIKeyAuth, + ) + from litellm.proxy.pass_through_endpoints.pass_through_endpoints import ( + get_pass_through_endpoints, + ) + + # Config-defined endpoints (from config file) + config_endpoints = [ + { + "path": "/v1/rerank", + "target": "https://api.cohere.com/v1/rerank", + "headers": {"content-type": "application/json"}, + }, + { + "path": "/v1/config-only", + "target": "https://config.example.com/api", + "headers": {}, + }, + ] + + # DB endpoints (one overlaps with config path, one is DB-only) + db_endpoints = [ + { + "id": "db-endpoint-1", + "path": "/v1/rerank", # Same as config - DB should override + "target": "https://db-override.com/v1/rerank", + "headers": {}, + "include_subpath": False, + }, + { + "id": "db-endpoint-2", + "path": "/db/only", + "target": "https://db-only.example.com/api", + "headers": {}, + "include_subpath": False, + }, + ] + + with patch( + "litellm.proxy.proxy_server.prisma_client", + MagicMock(), + ): + with patch( + "litellm.proxy.pass_through_endpoints.pass_through_endpoints._get_pass_through_endpoints_from_db", + new_callable=AsyncMock, + ) as mock_get_db: + with patch( + "litellm.proxy.pass_through_endpoints.pass_through_endpoints._get_pass_through_endpoints_from_config" + ) as mock_get_config: + db_objects = [ + PassThroughGenericEndpoint(**ep, is_from_config=False) + for ep in db_endpoints + ] + config_objects = [ + PassThroughGenericEndpoint(**ep, is_from_config=True) + for ep in config_endpoints + ] + mock_get_db.return_value = db_objects + mock_get_config.return_value = config_objects + + mock_user = MagicMock(spec=UserAPIKeyAuth) + + result = await get_pass_through_endpoints( + endpoint_id=None, + user_api_key_dict=mock_user, + team_id=None, + ) + + assert isinstance(result, PassThroughEndpointResponse) + # config_only: /v1/config-only (not in db_paths) + # db: /v1/rerank (overrides config), /db/only + # So we should have: /v1/config-only (from config) + /v1/rerank + /db/only (from db) + assert len(result.endpoints) == 3 + + # Check is_from_config values + by_path = {ep.path: ep for ep in result.endpoints} + assert by_path["/v1/config-only"].is_from_config is True + assert by_path["/v1/rerank"].is_from_config is False # DB overrides + assert by_path["/db/only"].is_from_config is False + + # Verify DB override: /v1/rerank should have DB target + assert by_path["/v1/rerank"].target == "https://db-override.com/v1/rerank" + + +def test_get_pass_through_endpoints_from_config_skips_malformed(): + """ + Test that _get_pass_through_endpoints_from_config skips malformed endpoints + and returns only valid ones, without raising. + """ + from litellm.proxy.pass_through_endpoints.pass_through_endpoints import ( + _get_pass_through_endpoints_from_config, + ) + + # Mix of valid and malformed config endpoints + config_passthrough_endpoints = [ + {"path": "/valid/1", "target": "https://valid1.example.com"}, + {}, # Missing required path and target + {"path": "/missing-target"}, # Missing required target + {"target": "https://example.com"}, # Missing required path + {"path": "/valid/2", "target": "https://valid2.example.com", "headers": {}}, + ] + + with patch( + "litellm.proxy.proxy_server.config_passthrough_endpoints", + config_passthrough_endpoints, + ): + result = _get_pass_through_endpoints_from_config() + + # Only the 2 valid endpoints should be returned + assert len(result) == 2 + paths = {ep.path for ep in result} + assert "/valid/1" in paths + assert "/valid/2" in paths + for ep in result: + assert ep.is_from_config is True + + @pytest.mark.asyncio async def test_delete_pass_through_endpoint_empty_list(): """ diff --git a/tests/test_litellm/proxy/spend_tracking/test_spend_management_endpoints.py b/tests/test_litellm/proxy/spend_tracking/test_spend_management_endpoints.py index 54a276bc97f..08205cd2d9d 100644 --- a/tests/test_litellm/proxy/spend_tracking/test_spend_management_endpoints.py +++ b/tests/test_litellm/proxy/spend_tracking/test_spend_management_endpoints.py @@ -206,6 +206,7 @@ ignored_keys = [ "metadata.additional_usage_values.cache_creation_input_tokens", "metadata.additional_usage_values.cache_read_input_tokens", "metadata.additional_usage_values.inference_geo", + "metadata.additional_usage_values.speed", "metadata.litellm_overhead_time_ms", "metadata.cost_breakdown", ] @@ -1025,6 +1026,85 @@ async def test_ui_view_spend_logs_with_model(client, monkeypatch): assert data["data"][0]["model"] == "gpt-3.5-turbo" +@pytest.mark.asyncio +async def test_ui_view_spend_logs_with_model_id(client, monkeypatch): + """Test that the model_id query param filters spend logs by litellm model deployment id.""" + mock_spend_logs = [ + { + "id": "log1", + "request_id": "req1", + "api_key": "sk-test-key", + "user": "test_user_1", + "team_id": "team1", + "spend": 0.05, + "startTime": datetime.datetime.now(timezone.utc).isoformat(), + "model": "gpt-3.5-turbo", + "model_id": "deployment-id-1", + "status": "success", + }, + { + "id": "log2", + "request_id": "req2", + "api_key": "sk-test-key", + "user": "test_user_2", + "team_id": "team1", + "spend": 0.10, + "startTime": datetime.datetime.now(timezone.utc).isoformat(), + "model": "gpt-4", + "model_id": "deployment-id-2", + "status": "success", + }, + ] + + class MockDB: + async def find_many(self, *args, **kwargs): + if ( + "where" in kwargs + and "model_id" in kwargs["where"] + and kwargs["where"]["model_id"] == "deployment-id-1" + ): + return [mock_spend_logs[0]] + return mock_spend_logs + + async def count(self, *args, **kwargs): + if ( + "where" in kwargs + and "model_id" in kwargs["where"] + and kwargs["where"]["model_id"] == "deployment-id-1" + ): + return 1 + return len(mock_spend_logs) + + class MockPrismaClient: + def __init__(self): + self.db = MockDB() + self.db.litellm_spendlogs = self.db + + mock_prisma_client = MockPrismaClient() + monkeypatch.setattr("litellm.proxy.proxy_server.prisma_client", mock_prisma_client) + + start_date = ( + datetime.datetime.now(timezone.utc) - datetime.timedelta(days=7) + ).strftime("%Y-%m-%d %H:%M:%S") + end_date = datetime.datetime.now(timezone.utc).strftime("%Y-%m-%d %H:%M:%S") + + response = client.get( + "/spend/logs/ui", + params={ + "model_id": "deployment-id-1", + "start_date": start_date, + "end_date": end_date, + }, + headers={"Authorization": "Bearer sk-test"}, + ) + + assert response.status_code == 200 + data = response.json() + assert data["total"] == 1 + assert len(data["data"]) == 1 + assert data["data"][0]["model_id"] == "deployment-id-1" + + @pytest.mark.asyncio async def test_ui_view_spend_logs_with_key_hash(client, monkeypatch): # Mock data for the test diff --git a/tests/test_litellm/responses/litellm_completion_transformation/test_litellm_completion_responses.py b/tests/test_litellm/responses/litellm_completion_transformation/test_litellm_completion_responses.py index 5074bbf4397..6d6162437c4 100644 --- a/tests/test_litellm/responses/litellm_completion_transformation/test_litellm_completion_responses.py +++ b/tests/test_litellm/responses/litellm_completion_transformation/test_litellm_completion_responses.py @@ -7,6 +7,7 @@ sys.path.insert( from litellm.responses.litellm_completion_transformation.transformation import ( LiteLLMCompletionResponsesConfig, + TOOL_CALLS_CACHE, ) from litellm.types.llms.openai import ( ChatCompletionResponseMessage, @@ -17,6 +18,8 @@ from litellm.types.utils import ( CompletionTokensDetailsWrapper, Message, ModelResponse, + Function, + ChatCompletionMessageToolCall, PromptTokensDetailsWrapper, Usage, ) @@ -755,6 +758,98 @@ class TestFunctionCallTransformation: tool_call = tool_calls[0] assert tool_call.get("id") == "fallback_id" + def test_ensure_tool_results_preserves_cached_openai_object_tool_call(self): + """ + Test cached ChatCompletionMessageToolCall objects are normalized correctly. + """ + tool_call_id = "call_cached_openai_object" + TOOL_CALLS_CACHE.set_cache( + key=tool_call_id, + value=ChatCompletionMessageToolCall( + id=tool_call_id, + type="function", + function=Function( + name="search_web", + arguments='{"query": "python bugs"}', + ), + ), + ) + + messages_missing_tool_calls = [ + {"role": "user", "content": "Search for python bugs"}, + {"role": "assistant", "content": None, "tool_calls": []}, + {"role": "tool", "content": "Found 5 results", "tool_call_id": tool_call_id}, + ] + + try: + fixed_messages = LiteLLMCompletionResponsesConfig._ensure_tool_results_have_corresponding_tool_calls( + messages=messages_missing_tool_calls, + tools=None, + ) + finally: + TOOL_CALLS_CACHE.delete_cache(key=tool_call_id) + + assistant_msg = fixed_messages[1] + tool_calls = assistant_msg.get("tool_calls", []) + assert len(tool_calls) == 1 + + tool_call = tool_calls[0] + function = tool_call.get("function", {}) + assert function.get("name") == "search_web" + assert function.get("arguments") == '{"query": "python bugs"}' + + def test_ensure_tool_results_preserves_cached_attr_object_tool_call(self): + """ + Test cached attribute-only tool call objects are normalized correctly. + """ + + class AttrOnlyFunction: + def __init__(self, name: str, arguments: str): + self.name = name + self.arguments = arguments + + class AttrOnlyToolCall: + def __init__(self, id: str, type: str, function: AttrOnlyFunction): + self.id = id + self.type = type + self.function = function + + tool_call_id = "call_cached_attr_object" + TOOL_CALLS_CACHE.set_cache( + key=tool_call_id, + value=AttrOnlyToolCall( + id=tool_call_id, + type="function", + function=AttrOnlyFunction( + name="search_web", + arguments='{"query": "attribute objects"}', + ), + ), + ) + + messages_missing_tool_calls = [ + {"role": "user", "content": "Search using attr object"}, + {"role": "assistant", "content": None, "tool_calls": []}, + {"role": "tool", "content": "Found 3 results", "tool_call_id": tool_call_id}, + ] + + try: + fixed_messages = LiteLLMCompletionResponsesConfig._ensure_tool_results_have_corresponding_tool_calls( + messages=messages_missing_tool_calls, + tools=None, + ) + finally: + TOOL_CALLS_CACHE.delete_cache(key=tool_call_id) + + assistant_msg = fixed_messages[1] + tool_calls = assistant_msg.get("tool_calls", []) + assert len(tool_calls) == 1 + + tool_call = tool_calls[0] + function = tool_call.get("function", {}) + assert function.get("name") == "search_web" + assert function.get("arguments") == '{"query": "attribute objects"}' + class TestToolChoiceTransformation: """Test the tool_choice transformation fix for Cursor IDE bug""" @@ -1424,6 +1519,47 @@ class TestUsageTransformation: assert response_usage.input_tokens_details is None assert response_usage.output_tokens_details is None + def test_transform_usage_with_image_tokens(self): + """Test that image_tokens from Vertex AI/Gemini are properly transformed to output_tokens_details""" + # Setup: Simulate Vertex AI/Gemini usage with image_tokens in completion_tokens_details + usage = Usage( + prompt_tokens=10, + completion_tokens=150, + total_tokens=160, + completion_tokens_details=CompletionTokensDetailsWrapper( + reasoning_tokens=0, + text_tokens=50, + image_tokens=100, # From Vertex AI candidatesTokensDetails with modality="IMAGE" + ), + ) + + chat_completion_response = ModelResponse( + id="test-response-id", + created=1234567890, + model="gemini-2.0-flash", + object="chat.completion", + usage=usage, + choices=[ + Choices( + finish_reason="stop", + index=0, + message=Message(content="Here is the generated image.", role="assistant"), + ) + ], + ) + + # Execute + response_usage = LiteLLMCompletionResponsesConfig._transform_chat_completion_usage_to_responses_usage( + chat_completion_response=chat_completion_response + ) + + # Assert + assert response_usage.output_tokens == 150 + assert response_usage.output_tokens_details is not None + assert response_usage.output_tokens_details.reasoning_tokens == 0 + assert response_usage.output_tokens_details.text_tokens == 50 + assert response_usage.output_tokens_details.image_tokens == 100 + class TestStreamingIDConsistency: """Test cases for consistent IDs across streaming events (issue #14962)""" @@ -1637,4 +1773,4 @@ class TestStreamingIDConsistency: # Verify it matches the cached ID assert iterator._cached_item_id is not None - assert iterator._cached_item_id == text_done_id \ No newline at end of file + assert iterator._cached_item_id == text_done_id diff --git a/tests/test_litellm/secret_managers/test_secret_managers_main.py b/tests/test_litellm/secret_managers/test_secret_managers_main.py index eaef6956cd5..4a6e303586a 100644 --- a/tests/test_litellm/secret_managers/test_secret_managers_main.py +++ b/tests/test_litellm/secret_managers/test_secret_managers_main.py @@ -46,15 +46,24 @@ def mock_env(): yield os.environ -@patch("litellm.secret_managers.main.oidc_cache") -@patch("litellm.secret_managers.main._get_oidc_http_handler") -@patch("httpx.Client") # Prevent any real HTTP connections -def test_oidc_google_success(mock_httpx_client, mock_get_http_handler, mock_oidc_cache): - mock_oidc_cache.get_cache.return_value = None - mock_handler = MockHTTPHandler(timeout=600.0) - mock_get_http_handler.return_value = mock_handler +def test_oidc_google_success(): + """Test Google OIDC token fetch with mocked handler (no real network calls).""" secret_name = "oidc/google/[invalid url, do not cite]" - result = get_secret(secret_name) + mock_handler = MockHTTPHandler(timeout=600.0) + mock_get_http_handler = Mock(return_value=mock_handler) + mock_oidc_cache = Mock() + mock_oidc_cache.get_cache.return_value = None + + with patch("litellm.secret_managers.main.oidc_cache", mock_oidc_cache): + with patch( + "litellm.secret_managers.main._get_oidc_http_handler", + mock_get_http_handler, + ): + with patch( + "litellm.secret_managers.main.HTTPHandler", + side_effect=lambda timeout=None: mock_handler, + ): + result = get_secret(secret_name) assert result == "mocked_token" assert mock_handler.last_params == {"audience": "[invalid url, do not cite]"} @@ -63,32 +72,49 @@ def test_oidc_google_success(mock_httpx_client, mock_get_http_handler, mock_oidc ) -@patch("litellm.secret_managers.main.oidc_cache") -@patch("litellm.secret_managers.main._get_oidc_http_handler") -def test_oidc_google_cached(mock_get_http_handler, mock_oidc_cache): +def test_oidc_google_cached(): + """Test Google OIDC uses cache and does not call HTTP (no real network calls).""" + secret_name = "oidc/google/[invalid url, do not cite]" + mock_get_http_handler = Mock() + mock_oidc_cache = Mock() mock_oidc_cache.get_cache.return_value = "cached_token" - secret_name = "oidc/google/[invalid url, do not cite]" - result = get_secret(secret_name) + with patch("litellm.secret_managers.main.oidc_cache", mock_oidc_cache): + with patch( + "litellm.secret_managers.main._get_oidc_http_handler", + mock_get_http_handler, + ): + with patch( + "litellm.secret_managers.main.HTTPHandler", + Mock(side_effect=AssertionError("HTTPHandler should not be used")), + ): + result = get_secret(secret_name) assert result == "cached_token", f"Expected cached token, got {result}" mock_oidc_cache.get_cache.assert_called_with(key=secret_name) - # Verify HTTP handler was never called since we had a cached token mock_get_http_handler.assert_not_called() -@patch("litellm.secret_managers.main.oidc_cache") -@patch("litellm.secret_managers.main._get_oidc_http_handler") -def test_oidc_google_failure(mock_get_http_handler, mock_oidc_cache): +def test_oidc_google_failure(): + """Test Google OIDC raises when provider returns error (no real network calls).""" + secret_name = "oidc/google/https://example.com/api" mock_handler = MockHTTPHandler(timeout=600.0) mock_handler.status_code = 400 - mock_get_http_handler.return_value = mock_handler + mock_get_http_handler = Mock(return_value=mock_handler) + mock_oidc_cache = Mock() mock_oidc_cache.get_cache.return_value = None - - secret_name = "oidc/google/https://example.com/api" - with pytest.raises(ValueError, match="Google OIDC provider failed"): - get_secret(secret_name) + with patch("litellm.secret_managers.main.oidc_cache", mock_oidc_cache): + with patch( + "litellm.secret_managers.main._get_oidc_http_handler", + mock_get_http_handler, + ): + with patch( + "litellm.secret_managers.main.HTTPHandler", + side_effect=lambda timeout=None: mock_handler, + ): + with pytest.raises(ValueError, match="Google OIDC provider failed"): + get_secret(secret_name) def test_oidc_circleci_success(monkeypatch): @@ -151,20 +177,18 @@ def test_oidc_azure_file_success(mock_env, tmp_path): @patch("litellm.secret_managers.main.get_azure_ad_token_provider") -@patch.dict(os.environ, {}, clear=False) # Ensure AZURE_FEDERATED_TOKEN_FILE is not set -def test_oidc_azure_ad_token_success(mock_get_azure_ad_token_provider): - # Ensure the env var is not set so it falls through to Azure AD token provider - if "AZURE_FEDERATED_TOKEN_FILE" in os.environ: - del os.environ["AZURE_FEDERATED_TOKEN_FILE"] - +def test_oidc_azure_ad_token_success(mock_get_azure_ad_token_provider, monkeypatch): + # Force-unset so we always hit the Azure AD token provider path (CI may set AZURE_FEDERATED_TOKEN_FILE) + monkeypatch.delenv("AZURE_FEDERATED_TOKEN_FILE", raising=False) + # Mock the token provider function that gets returned and called mock_token_provider = Mock(return_value="azure_ad_token") mock_get_azure_ad_token_provider.return_value = mock_token_provider - + # Also mock the Azure Identity SDK to prevent any real Azure calls with patch("azure.identity.get_bearer_token_provider") as mock_bearer: mock_bearer.return_value = mock_token_provider - + secret_name = "oidc/azure/api://azure-audience" result = get_secret(secret_name) diff --git a/tests/test_litellm/test_constants.py b/tests/test_litellm/test_constants.py index 77f2f308f88..23447a02e04 100644 --- a/tests/test_litellm/test_constants.py +++ b/tests/test_litellm/test_constants.py @@ -38,6 +38,11 @@ def test_all_numeric_constants_can_be_overridden(): print("all numeric constants", json.dumps(numeric_constants, indent=4)) + # Constants that use a different env var name than the constant name + constant_to_env_var = { + "MAX_CALLBACKS": "LITELLM_MAX_CALLBACKS", + } + # Verify all numeric constants have environment variable support for name, value in numeric_constants: # Skip constants that are not meant to be overridden (if any) @@ -47,8 +52,11 @@ def test_all_numeric_constants_can_be_overridden(): # Create a test value that's different from the default test_value = value + 1 if isinstance(value, int) else value + 0.1 + # Use the env var name that the constants module actually reads + env_var_name = constant_to_env_var.get(name, name) + # Set the environment variable - with mock.patch.dict(os.environ, {name: str(test_value)}): + with mock.patch.dict(os.environ, {env_var_name: str(test_value)}): print("overriding", name, "with", test_value) importlib.reload(constants) diff --git a/tests/test_litellm/test_logging.py b/tests/test_litellm/test_logging.py index 7e5931d8c0f..6f65ada7459 100644 --- a/tests/test_litellm/test_logging.py +++ b/tests/test_litellm/test_logging.py @@ -1,27 +1,21 @@ import asyncio -import datetime import json import os import sys -import unittest -from typing import List, Optional, Tuple -from unittest.mock import ANY, MagicMock, Mock, patch +from typing import List -import httpx import pytest sys.path.insert( 0, os.path.abspath("../../..") ) # Adds the parent directory to the system-path -import io import logging import sys -import unittest -from contextlib import redirect_stdout import litellm from litellm._logging import ( ALL_LOGGERS, + JsonFormatter, _initialize_loggers_with_handler, _turn_on_json, verbose_logger, @@ -72,6 +66,117 @@ def test_json_mode_emits_one_record_per_logger(capfd): assert "timestamp" in obj, "`timestamp` key missing" +def test_json_formatter_parses_embedded_json_message(): + """ + Test that JsonFormatter parses embedded JSON in the message field and promotes + sub-fields to first-class JSON properties for downstream querying. + """ + formatter = JsonFormatter() + record = logging.LogRecord( + name="LiteLLM", + level=logging.DEBUG, + pathname="", + lineno=0, + msg='{"event": "giveup", "exception": "Connection failed", "model_name": "gpt-4"}', + args=(), + exc_info=None, + ) + output = formatter.format(record) + obj = json.loads(output) + # Standard fields preserved + assert "message" in obj + assert obj["level"] == "DEBUG" + assert "timestamp" in obj + # Embedded JSON fields promoted to top-level for querying + assert obj["event"] == "giveup" + assert obj["exception"] == "Connection failed" + assert obj["model_name"] == "gpt-4" + + +def test_json_formatter_includes_extra_attributes(): + """ + Test that JsonFormatter includes extra attributes from logger.debug("msg", extra={...}). + """ + formatter = JsonFormatter() + record = logging.LogRecord( + name="LiteLLM", + level=logging.DEBUG, + pathname="", + lineno=0, + msg="POST Request Sent from LiteLLM", + args=(), + exc_info=None, + ) + record.api_base = "https://api.openai.com" + record.authorization = "Bearer sk-***" + output = formatter.format(record) + obj = json.loads(output) + assert obj["message"] == "POST Request Sent from LiteLLM" + assert obj["api_base"] == "https://api.openai.com" + assert obj["authorization"] == "Bearer sk-***" + + +def test_json_formatter_plain_message_unchanged(): + """ + Test that non-JSON messages are passed through as-is in the message field. + """ + formatter = JsonFormatter() + record = logging.LogRecord( + name="LiteLLM", + level=logging.INFO, + pathname="", + lineno=0, + msg="Cache hit!", + args=(), + exc_info=None, + ) + output = formatter.format(record) + obj = json.loads(output) + assert obj["message"] == "Cache hit!" + assert "event" not in obj + assert "exception" not in obj + + +def test_json_formatter_parses_embedded_python_dict_repr(): + """ + Test that JsonFormatter parses Python dict repr (str/deployment) embedded in + plain text, e.g. from get_available_deployment logs. + Reproduces Roni's reported case. + """ + formatter = JsonFormatter() + msg = ( + "get_available_deployment for model: text-embedding-3-large, " + "Selected deployment: {'model_name': 'text-embedding-3-large', " + "'litellm_params': {'api_key': 'sk**********', 'tpm': 1000000, 'rpm': 2000, " + "'use_in_pass_through': False, 'use_litellm_proxy': False, " + "'merge_reasoning_content_in_choices': False, 'model': 'text-embedding-3-large'}, " + "'model_info': {'id': 'a624b057aec64ada48311', 'db_model': False}} " + "for model: text-embedding-3-large" + ) + record = logging.LogRecord( + name="LiteLLM Router", + level=logging.INFO, + pathname="", + lineno=0, + msg=msg, + args=(), + exc_info=None, + ) + output = formatter.format(record) + obj = json.loads(output) + assert "message" in obj + assert obj["level"] == "INFO" + # Python dict parsed and promoted to first-class properties + assert obj["model_name"] == "text-embedding-3-large" + assert "litellm_params" in obj + assert obj["litellm_params"]["api_key"] == "sk**********" + assert obj["litellm_params"]["tpm"] == 1000000 + assert obj["litellm_params"]["use_in_pass_through"] is False + assert "model_info" in obj + assert obj["model_info"]["id"] == "a624b057aec64ada48311" + assert obj["model_info"]["db_model"] is False + + def test_initialize_loggers_with_handler_sets_propagate_false(): """ Test that the initialize_loggers_with_handler function sets propagate to False for all loggers @@ -96,7 +201,7 @@ async def test_cache_hit_includes_custom_llm_provider(): test_custom_logger = CacheHitCustomLogger() original_callbacks = litellm.callbacks.copy() if litellm.callbacks else [] litellm.callbacks = [test_custom_logger] - + try: # First call - should be a cache miss response1 = await litellm.acompletion( @@ -105,10 +210,10 @@ async def test_cache_hit_includes_custom_llm_provider(): mock_response="test response", caching=True, ) - + # Wait for logging to complete await asyncio.sleep(0.5) - + # Second identical call - should be a cache hit response2 = await litellm.acompletion( model="gpt-3.5-turbo", @@ -116,38 +221,43 @@ async def test_cache_hit_includes_custom_llm_provider(): mock_response="test response", caching=True, ) - + # Wait for logging to complete await asyncio.sleep(0.5) - + # Verify we have logged events - assert len(test_custom_logger.logged_standard_logging_payloads) >= 2, \ - f"Expected at least 2 logged events, got {len(test_custom_logger.logged_standard_logging_payloads)}" - + assert ( + len(test_custom_logger.logged_standard_logging_payloads) >= 2 + ), f"Expected at least 2 logged events, got {len(test_custom_logger.logged_standard_logging_payloads)}" + # Find the cache hit event (should be the second call) cache_hit_payload = None for payload in test_custom_logger.logged_standard_logging_payloads: if payload.get("cache_hit") is True: cache_hit_payload = payload break - + # Verify cache hit event was found - assert cache_hit_payload is not None, "No cache hit event found in logged payloads" - + assert ( + cache_hit_payload is not None + ), "No cache hit event found in logged payloads" + # Verify custom_llm_provider is included in the cache hit payload - assert "custom_llm_provider" in cache_hit_payload, \ - "custom_llm_provider missing from cache hit standard logging payload" - + assert ( + "custom_llm_provider" in cache_hit_payload + ), "custom_llm_provider missing from cache hit standard logging payload" + # Verify custom_llm_provider has a valid value (should be "openai" for gpt-3.5-turbo) custom_llm_provider = cache_hit_payload["custom_llm_provider"] - assert custom_llm_provider is not None and custom_llm_provider != "", \ - f"custom_llm_provider should not be None or empty, got: {custom_llm_provider}" - + assert ( + custom_llm_provider is not None and custom_llm_provider != "" + ), f"custom_llm_provider should not be None or empty, got: {custom_llm_provider}" + print( f"Cache hit standard logging payload with custom_llm_provider: {custom_llm_provider}", json.dumps(cache_hit_payload, indent=2), ) - + finally: # Clean up litellm.callbacks = original_callbacks diff --git a/tests/test_litellm/test_router.py b/tests/test_litellm/test_router.py index 75ec806ee17..9dcb16b545e 100644 --- a/tests/test_litellm/test_router.py +++ b/tests/test_litellm/test_router.py @@ -1990,3 +1990,94 @@ async def test_anthropic_messages_call_type_is_cached(): # This assertion will FAIL if anthropic_messages is filtered out assert cached_result is not None, "Model ID should be cached for anthropic_messages call type" assert cached_result["model_id"] == test_model_id, f"Expected {test_model_id}, got {cached_result['model_id']}" + + +def test_update_kwargs_with_deployment_propagates_model_tags(): + """ + Test that deployment-level tags from litellm_params are merged into + kwargs metadata when _update_kwargs_with_deployment is called. + + This ensures model-level tags defined in config.yaml appear in SpendLogs. + See: https://github.com/BerriAI/litellm/issues/XXXX + """ + router = litellm.Router( + model_list=[ + { + "model_name": "gpt-4o-mini", + "litellm_params": { + "model": "openai/gpt-4o-mini", + "api_key": "fake-key", + "tags": ["openai-account", "production"], + }, + }, + ], + ) + + kwargs: dict = {"metadata": {}} + deployment = router.get_deployment_by_model_group_name( + model_group_name="gpt-4o-mini" + ) + router._update_kwargs_with_deployment(deployment=deployment, kwargs=kwargs) + + # Deployment tags should be propagated to kwargs metadata + assert "tags" in kwargs["metadata"] + assert "openai-account" in kwargs["metadata"]["tags"] + assert "production" in kwargs["metadata"]["tags"] + + +def test_update_kwargs_with_deployment_merges_tags_without_duplicates(): + """ + Test that when both request-level and deployment-level tags exist, + they are merged without duplicates. + """ + router = litellm.Router( + model_list=[ + { + "model_name": "gpt-4o-mini", + "litellm_params": { + "model": "openai/gpt-4o-mini", + "api_key": "fake-key", + "tags": ["openai-account", "shared-tag"], + }, + }, + ], + ) + + # Simulate request that already has tags (from request body or key/team level) + kwargs: dict = {"metadata": {"tags": ["user-tag", "shared-tag"]}} + deployment = router.get_deployment_by_model_group_name( + model_group_name="gpt-4o-mini" + ) + router._update_kwargs_with_deployment(deployment=deployment, kwargs=kwargs) + + # Both sources should be merged, no duplicates + assert "user-tag" in kwargs["metadata"]["tags"] + assert "openai-account" in kwargs["metadata"]["tags"] + assert "shared-tag" in kwargs["metadata"]["tags"] + assert kwargs["metadata"]["tags"].count("shared-tag") == 1 + + +def test_update_kwargs_with_deployment_no_tags(): + """ + Test that when deployment has no tags, kwargs metadata is not affected. + """ + router = litellm.Router( + model_list=[ + { + "model_name": "gpt-4o-mini", + "litellm_params": { + "model": "openai/gpt-4o-mini", + "api_key": "fake-key", + }, + }, + ], + ) + + kwargs: dict = {"metadata": {}} + deployment = router.get_deployment_by_model_group_name( + model_group_name="gpt-4o-mini" + ) + router._update_kwargs_with_deployment(deployment=deployment, kwargs=kwargs) + + # No tags key should be added if deployment has no tags + assert "tags" not in kwargs["metadata"] diff --git a/tests/test_litellm/test_router_model_cost_isolation.py b/tests/test_litellm/test_router_model_cost_isolation.py new file mode 100644 index 00000000000..2112295e040 --- /dev/null +++ b/tests/test_litellm/test_router_model_cost_isolation.py @@ -0,0 +1,264 @@ +""" +Test that per-deployment custom pricing does not pollute the shared backend +model key in litellm.model_cost. + +When two deployments share the same backend model (e.g. vertex_ai/gemini-2.5-flash) +and one has explicit zero-cost pricing in model_info, the other deployment +should still use the built-in pricing. +""" + +import os +import sys + +import pytest + +sys.path.insert( + 0, os.path.abspath("../../..") +) # Adds the parent directory to the system path + +import litellm +from litellm import Router + + +def test_should_not_pollute_shared_key_with_zero_cost_pricing(): + """ + When deployment A has input_cost_per_token=0 and deployment B has no + custom pricing, deployment B should still report the built-in pricing + (not zero). + """ + backend_model = "vertex_ai/gemini-2.5-flash" + + # Grab built-in pricing before creating any router + builtin_info = litellm.get_model_info(model=backend_model) + builtin_input_cost = builtin_info["input_cost_per_token"] + builtin_output_cost = builtin_info["output_cost_per_token"] + + # Sanity: built-in pricing should be non-zero for this model + assert builtin_input_cost > 0, "Test requires a model with non-zero built-in pricing" + assert builtin_output_cost > 0, "Test requires a model with non-zero built-in pricing" + + router = Router( + model_list=[ + # Deployment A: explicit zero-cost pricing + { + "model_name": "custom-zero-cost-model", + "litellm_params": { + "model": backend_model, + "api_key": "fake-key-1", + }, + "model_info": { + "id": "deployment-a-zero-cost", + "input_cost_per_token": 0.0, + "output_cost_per_token": 0.0, + }, + }, + # Deployment B: no custom pricing, relies on built-in + { + "model_name": "standard-cost-model", + "litellm_params": { + "model": backend_model, + "api_key": "fake-key-2", + }, + "model_info": { + "id": "deployment-b-builtin-cost", + }, + }, + ], + ) + + # Deployment A: should report zero pricing via its unique model_id + info_a = router.get_deployment_model_info( + model_id="deployment-a-zero-cost", + model_name=backend_model, + ) + assert info_a is not None + assert info_a["input_cost_per_token"] == 0.0 + assert info_a["output_cost_per_token"] == 0.0 + + # Deployment B: should report built-in pricing, NOT zero + info_b = router.get_deployment_model_info( + model_id="deployment-b-builtin-cost", + model_name=backend_model, + ) + assert info_b is not None + assert info_b["input_cost_per_token"] == builtin_input_cost, ( + f"Deployment B should use built-in input cost {builtin_input_cost}, " + f"got {info_b['input_cost_per_token']}" + ) + assert info_b["output_cost_per_token"] == builtin_output_cost, ( + f"Deployment B should use built-in output cost {builtin_output_cost}, " + f"got {info_b['output_cost_per_token']}" + ) + + +def test_should_not_pollute_shared_key_with_custom_nonzero_pricing(): + """ + A deployment with custom (non-zero) pricing should not overwrite + the shared backend key's built-in pricing. + """ + backend_model = "vertex_ai/gemini-2.5-flash" + + builtin_info = litellm.get_model_info(model=backend_model) + builtin_input_cost = builtin_info["input_cost_per_token"] + + router = Router( + model_list=[ + # Deployment with custom high pricing + { + "model_name": "expensive-model", + "litellm_params": { + "model": backend_model, + "api_key": "fake-key-3", + }, + "model_info": { + "id": "deployment-expensive", + "input_cost_per_token": 0.99, + "output_cost_per_token": 0.99, + }, + }, + # Deployment relying on built-in pricing + { + "model_name": "standard-model", + "litellm_params": { + "model": backend_model, + "api_key": "fake-key-4", + }, + "model_info": { + "id": "deployment-standard", + }, + }, + ], + ) + + # Custom pricing deployment should see its custom values + info_expensive = router.get_deployment_model_info( + model_id="deployment-expensive", + model_name=backend_model, + ) + assert info_expensive is not None + assert info_expensive["input_cost_per_token"] == 0.99 + assert info_expensive["output_cost_per_token"] == 0.99 + + # Standard deployment should still see built-in pricing + info_standard = router.get_deployment_model_info( + model_id="deployment-standard", + model_name=backend_model, + ) + assert info_standard is not None + assert info_standard["input_cost_per_token"] == builtin_input_cost, ( + f"Standard deployment should use built-in pricing {builtin_input_cost}, " + f"got {info_standard['input_cost_per_token']}" + ) + + +def test_should_store_full_pricing_under_deployment_model_id(): + """ + Per-deployment pricing (including zero) should be stored and + retrievable via the unique model_id key in litellm.model_cost. + """ + backend_model = "vertex_ai/gemini-2.5-flash" + + router = Router( + model_list=[ + { + "model_name": "zero-cost-model", + "litellm_params": { + "model": backend_model, + "api_key": "fake-key-5", + }, + "model_info": { + "id": "deployment-zero-check", + "input_cost_per_token": 0.0, + "output_cost_per_token": 0.0, + }, + }, + ], + ) + + # The model_id entry should exist and have the zero pricing + entry = litellm.model_cost.get("deployment-zero-check") + assert entry is not None, "Deployment should be registered by model_id" + assert entry["input_cost_per_token"] == 0.0 + assert entry["output_cost_per_token"] == 0.0 + + +def test_should_preserve_builtin_pricing_regardless_of_deployment_order(): + """ + The built-in pricing should be preserved no matter which deployment + is processed first (zero-cost first, or standard first). + """ + backend_model = "vertex_ai/gemini-2.5-flash" + + builtin_info = litellm.get_model_info(model=backend_model) + builtin_input_cost = builtin_info["input_cost_per_token"] + builtin_output_cost = builtin_info["output_cost_per_token"] + + # Order 1: standard first, then zero-cost + router1 = Router( + model_list=[ + { + "model_name": "standard-first", + "litellm_params": { + "model": backend_model, + "api_key": "fake-key-6", + }, + "model_info": {"id": "order1-standard"}, + }, + { + "model_name": "zero-cost-second", + "litellm_params": { + "model": backend_model, + "api_key": "fake-key-7", + }, + "model_info": { + "id": "order1-zero", + "input_cost_per_token": 0.0, + "output_cost_per_token": 0.0, + }, + }, + ], + ) + + info_std_1 = router1.get_deployment_model_info( + model_id="order1-standard", model_name=backend_model + ) + assert info_std_1["input_cost_per_token"] == builtin_input_cost + assert info_std_1["output_cost_per_token"] == builtin_output_cost + + # Order 2: zero-cost first, then standard + router2 = Router( + model_list=[ + { + "model_name": "zero-cost-first", + "litellm_params": { + "model": backend_model, + "api_key": "fake-key-8", + }, + "model_info": { + "id": "order2-zero", + "input_cost_per_token": 0.0, + "output_cost_per_token": 0.0, + }, + }, + { + "model_name": "standard-second", + "litellm_params": { + "model": backend_model, + "api_key": "fake-key-9", + }, + "model_info": {"id": "order2-standard"}, + }, + ], + ) + + info_std_2 = router2.get_deployment_model_info( + model_id="order2-standard", model_name=backend_model + ) + assert info_std_2["input_cost_per_token"] == builtin_input_cost, ( + f"Order should not matter. Expected {builtin_input_cost}, " + f"got {info_std_2['input_cost_per_token']}" + ) + assert info_std_2["output_cost_per_token"] == builtin_output_cost, ( + f"Order should not matter. Expected {builtin_output_cost}, " + f"got {info_std_2['output_cost_per_token']}" + ) diff --git a/tests/test_litellm/test_utils.py b/tests/test_litellm/test_utils.py index 352125d16c4..794b3b87187 100644 --- a/tests/test_litellm/test_utils.py +++ b/tests/test_litellm/test_utils.py @@ -661,6 +661,7 @@ def test_aaamodel_prices_and_context_window_json_is_valid(): "supports_url_context": {"type": "boolean"}, "supports_reasoning": {"type": "boolean"}, "supports_service_tier": {"type": "boolean"}, + "supports_preset": {"type": "boolean"}, "tool_use_system_prompt_tokens": {"type": "number"}, "tpm": {"type": "number"}, "supported_endpoints": { diff --git a/tests/test_litellm/test_video_generation.py b/tests/test_litellm/test_video_generation.py index 5446a0a7b3f..c8cc292519b 100644 --- a/tests/test_litellm/test_video_generation.py +++ b/tests/test_litellm/test_video_generation.py @@ -798,7 +798,7 @@ def test_openai_transform_video_content_request_empty_params(): def test_video_content_handler_uses_get_for_openai(): """HTTP handler must use GET (not POST) for OpenAI content download.""" from litellm.types.router import GenericLiteLLMParams - + handler = BaseLLMHTTPHandler() config = OpenAIVideoConfig() @@ -807,7 +807,12 @@ def test_video_content_handler_uses_get_for_openai(): mock_response.content = b"mp4-bytes" mock_client.get.return_value = mock_response + # Patch both where _get_httpx_client is used and where it is defined so the mock + # is used regardless of import order / CI environment with patch( + "litellm.llms.custom_httpx.http_handler._get_httpx_client", + return_value=mock_client, + ), patch( "litellm.llms.custom_httpx.llm_http_handler._get_httpx_client", return_value=mock_client, ): diff --git a/ui/litellm-dashboard/e2e_tests/constants.ts b/ui/litellm-dashboard/e2e_tests/constants.ts index b07bd68fcf1..58b56af0a2b 100644 --- a/ui/litellm-dashboard/e2e_tests/constants.ts +++ b/ui/litellm-dashboard/e2e_tests/constants.ts @@ -1 +1,6 @@ export const ADMIN_STORAGE_PATH = "admin.storageState.json"; + +export const E2E_UPDATE_LIMITS_KEY_ID_PREFIX = "102c"; +export const E2E_DELETE_KEY_ID_PREFIX = "94a5"; +export const E2E_DELETE_KEY_NAME = "e2eDeleteKey"; +export const E2E_REGENERATE_KEY_ID_PREFIX = "593a"; diff --git a/ui/litellm-dashboard/e2e_tests/tests/keys/deleteKey.spec.ts b/ui/litellm-dashboard/e2e_tests/tests/keys/deleteKey.spec.ts new file mode 100644 index 00000000000..a5841316251 --- /dev/null +++ b/ui/litellm-dashboard/e2e_tests/tests/keys/deleteKey.spec.ts @@ -0,0 +1,25 @@ +import { test, expect } from "@playwright/test"; +import { ADMIN_STORAGE_PATH, E2E_DELETE_KEY_ID_PREFIX, E2E_DELETE_KEY_NAME } from "../../constants"; +import { Page } from "../../fixtures/pages"; +import { navigateToPage } from "../../helpers/navigation"; + +test.describe("Delete Key", () => { + test.use({ storageState: ADMIN_STORAGE_PATH }); + + test("Able to delete a key", async ({ page }) => { + await navigateToPage(page, Page.ApiKeys); + await expect(page.getByRole("button", { name: "Next" })).toBeVisible(); + await page + .locator("button", { + hasText: E2E_DELETE_KEY_ID_PREFIX, + }) + .click(); + await page.getByRole("button", { name: "Delete Key" }).click(); + await page.getByRole("textbox", { name: E2E_DELETE_KEY_NAME }).click(); + await page.getByRole("textbox", { name: E2E_DELETE_KEY_NAME }).fill(E2E_DELETE_KEY_NAME); + const deleteButton = page.getByRole("button", { name: "Delete", exact: true }); + await expect(deleteButton).toBeEnabled(); + await deleteButton.click(); + await expect(page.getByText("Key deleted successfully")).toBeVisible(); + }); +}); diff --git a/ui/litellm-dashboard/e2e_tests/tests/keys/regenerateKey.spec.ts b/ui/litellm-dashboard/e2e_tests/tests/keys/regenerateKey.spec.ts new file mode 100644 index 00000000000..0188a4f81ce --- /dev/null +++ b/ui/litellm-dashboard/e2e_tests/tests/keys/regenerateKey.spec.ts @@ -0,0 +1,21 @@ +import { test, expect } from "@playwright/test"; +import { ADMIN_STORAGE_PATH, E2E_REGENERATE_KEY_ID_PREFIX } from "../../constants"; +import { Page } from "../../fixtures/pages"; +import { navigateToPage } from "../../helpers/navigation"; + +test.describe("Regenerate Key", () => { + test.use({ storageState: ADMIN_STORAGE_PATH }); + + test("Able to regenerate a key", async ({ page }) => { + await navigateToPage(page, Page.ApiKeys); + await expect(page.getByRole("button", { name: "Next" })).toBeVisible(); + await page + .locator("button", { + hasText: E2E_REGENERATE_KEY_ID_PREFIX, + }) + .click(); + await page.getByRole("button", { name: "Regenerate Key" }).click(); + await page.getByRole("button", { name: "Regenerate", exact: true }).click(); + await expect(page.getByText("Virtual Key regenerated")).toBeVisible(); + }); +}); diff --git a/ui/litellm-dashboard/e2e_tests/tests/keys/updateKeyLimits.spec.ts b/ui/litellm-dashboard/e2e_tests/tests/keys/updateKeyLimits.spec.ts new file mode 100644 index 00000000000..6cae36272ab --- /dev/null +++ b/ui/litellm-dashboard/e2e_tests/tests/keys/updateKeyLimits.spec.ts @@ -0,0 +1,27 @@ +import { test, expect } from "@playwright/test"; +import { ADMIN_STORAGE_PATH, E2E_UPDATE_LIMITS_KEY_ID_PREFIX } from "../../constants"; +import { Page } from "../../fixtures/pages"; +import { navigateToPage } from "../../helpers/navigation"; + +test.describe("Update Key TPM and RPM Limits", () => { + test.use({ storageState: ADMIN_STORAGE_PATH }); + + test("Able to update a key's TPM and RPM limits", async ({ page }) => { + await navigateToPage(page, Page.ApiKeys); + await expect(page.getByRole("button", { name: "Next" })).toBeVisible(); + await page + .locator("button", { + hasText: E2E_UPDATE_LIMITS_KEY_ID_PREFIX, + }) + .click(); + await page.getByRole("tab", { name: "Settings" }).click(); + await page.getByRole("button", { name: "Edit Settings" }).click(); + await page.getByRole("spinbutton", { name: "TPM Limit" }).click(); + await page.getByRole("spinbutton", { name: "TPM Limit" }).fill("123"); + await page.getByRole("spinbutton", { name: "RPM Limit" }).click(); + await page.getByRole("spinbutton", { name: "RPM Limit" }).fill("456"); + await page.getByRole("button", { name: "Save Changes" }).click(); + await expect(page.getByRole("paragraph").filter({ hasText: "TPM: 123" })).toBeVisible(); + await expect(page.getByRole("paragraph").filter({ hasText: "RPM: 456" })).toBeVisible(); + }); +}); diff --git a/ui/litellm-dashboard/package-lock.json b/ui/litellm-dashboard/package-lock.json index 4205657ca8c..3a21813fbf4 100644 --- a/ui/litellm-dashboard/package-lock.json +++ b/ui/litellm-dashboard/package-lock.json @@ -13159,6 +13159,21 @@ "type": "github", "url": "https://github.com/sponsors/wooorm" } + }, + "node_modules/@next/swc-win32-ia32-msvc": { + "version": "14.2.33", + "resolved": "https://registry.npmjs.org/@next/swc-win32-ia32-msvc/-/swc-win32-ia32-msvc-14.2.33.tgz", + "integrity": "sha512-pc9LpGNKhJ0dXQhZ5QMmYxtARwwmWLpeocFmVG5Z0DzWq5Uf0izcI8tLc+qOpqxO1PWqZ5A7J1blrUIKrIFc7Q==", + "cpu": [ + "ia32" + ], + "optional": true, + "os": [ + "win32" + ], + "engines": { + "node": ">= 10" + } } } } diff --git a/ui/litellm-dashboard/package.json b/ui/litellm-dashboard/package.json index 76ac97f008c..164368eb6ba 100644 --- a/ui/litellm-dashboard/package.json +++ b/ui/litellm-dashboard/package.json @@ -3,7 +3,7 @@ "version": "0.1.0", "private": true, "scripts": { - "dev": "next dev --webpack", + "dev": "next dev", "build": "next build", "start": "next start", "lint": "next lint", @@ -79,6 +79,7 @@ "vitest": "^3.2.4" }, "overrides": { + "diff": ">=8.0.3", "prismjs": ">=1.30.0", "webpack-dev-server": ">=5.2.1", "mermaid": ">=11.10.0", diff --git a/ui/litellm-dashboard/scripts/e2e_tests/neonHelperScripts.ts b/ui/litellm-dashboard/scripts/e2e_tests/neonHelperScripts.ts index 3078a0d90d2..089ad4e7926 100644 --- a/ui/litellm-dashboard/scripts/e2e_tests/neonHelperScripts.ts +++ b/ui/litellm-dashboard/scripts/e2e_tests/neonHelperScripts.ts @@ -1,4 +1,4 @@ -import { createApiClient } from "@neondatabase/api-client"; +import { createApiClient, EndpointType } from "@neondatabase/api-client"; import { config } from "dotenv"; import { resolve } from "path"; @@ -27,6 +27,13 @@ export async function createNeonE2ETestingBranch(projectId: string, parentBranch parent_id: parentBranchId, expires_at: expireAt ?? new Date(Date.now() + 1000 * 60 * 30).toISOString(), }, + endpoints: [ + { + type: EndpointType.ReadWrite, + autoscaling_limit_min_cu: 0.25, + autoscaling_limit_max_cu: 1, + }, + ], }); return response; } catch (error) { @@ -35,13 +42,15 @@ export async function createNeonE2ETestingBranch(projectId: string, parentBranch } export async function getNeonE2ETestingBranchConnectionString() { - await createNeonE2ETestingBranch(PROJECT_ID, PARENT_BRANCH); - + const createBranchResponse = await createNeonE2ETestingBranch(PROJECT_ID, PARENT_BRANCH); + const projectId = createBranchResponse.data.branch.project_id; const response = await apiClient.getConnectionUri({ database_name: NEON_E2E_UI_TEST_DB_NAME, role_name: "neondb_owner", - projectId: PROJECT_ID, + projectId: projectId, }); console.log("connection string:", response.data.uri); return response.data.uri; } + +getNeonE2ETestingBranchConnectionString(); diff --git a/ui/litellm-dashboard/src/app/(dashboard)/hooks/models/useModels.test.ts b/ui/litellm-dashboard/src/app/(dashboard)/hooks/models/useModels.test.ts index 4985206092f..2539cc63f95 100644 --- a/ui/litellm-dashboard/src/app/(dashboard)/hooks/models/useModels.test.ts +++ b/ui/litellm-dashboard/src/app/(dashboard)/hooks/models/useModels.test.ts @@ -3,13 +3,14 @@ import { renderHook, waitFor } from "@testing-library/react"; import React, { ReactNode } from "react"; import { beforeEach, describe, expect, it, vi } from "vitest"; import { - useModelsInfo, - useModelHub, useAllProxyModels, + useInfiniteModelInfo, + useModelHub, + useModelsInfo, useSelectedTeamModels, - type ProxyModel, type AllProxyModelsResponse, type PaginatedModelInfoResponse, + type ProxyModel, } from "./useModels"; vi.mock("@/components/networking", () => ({ @@ -23,7 +24,7 @@ vi.mock("@/app/(dashboard)/hooks/useAuthorized", () => ({ default: () => mockUseAuthorized(), })); -import { modelInfoCall, modelHubCall, modelAvailableCall } from "@/components/networking"; +import { modelAvailableCall, modelHubCall, modelInfoCall } from "@/components/networking"; const mockProxyModel: ProxyModel = { id: "model-1", @@ -106,7 +107,7 @@ describe("useModelsInfo", () => { undefined, undefined, undefined, - undefined + undefined, ); expect(modelInfoCall).toHaveBeenCalledTimes(1); }); @@ -130,7 +131,7 @@ describe("useModelsInfo", () => { undefined, undefined, undefined, - undefined + undefined, ); }); @@ -393,7 +394,7 @@ describe("useAllProxyModels", () => { null, true, false, - "expand" + "expand", ); expect(modelAvailableCall).toHaveBeenCalledTimes(1); }); @@ -531,13 +532,7 @@ describe("useSelectedTeamModels", () => { expect(result.current.data).toEqual(mockAllProxyModelsResponse); expect(result.current.error).toBeNull(); - expect(modelAvailableCall).toHaveBeenCalledWith( - "test-access-token", - "test-user-id", - "Admin", - true, - "team-1" - ); + expect(modelAvailableCall).toHaveBeenCalledWith("test-access-token", "test-user-id", "Admin", true, "team-1"); expect(modelAvailableCall).toHaveBeenCalledTimes(1); }); @@ -639,3 +634,222 @@ describe("useSelectedTeamModels", () => { expect(modelAvailableCall).not.toHaveBeenCalled(); }); }); + +describe("useInfiniteModelInfo", () => { + let queryClient: QueryClient; + + const mockPageOneResponse: PaginatedModelInfoResponse = { + data: [{ model_name: "gpt-4", model_info: { id: "model-1" } }], + total_count: 2, + current_page: 1, + total_pages: 2, + size: 50, + }; + + const mockPageTwoResponse: PaginatedModelInfoResponse = { + data: [{ model_name: "claude-3", model_info: { id: "model-2" } }], + total_count: 2, + current_page: 2, + total_pages: 2, + size: 50, + }; + + beforeEach(() => { + queryClient = new QueryClient({ + defaultOptions: { + queries: { + retry: false, + }, + }, + }); + + vi.clearAllMocks(); + + mockUseAuthorized.mockReturnValue({ + accessToken: "test-access-token", + userId: "test-user-id", + userRole: "Admin", + token: "test-token", + userEmail: "test@example.com", + premiumUser: false, + disabledPersonalKeyCreation: null, + showSSOBanner: false, + }); + }); + + const wrapper = ({ children }: { children: ReactNode }) => + React.createElement(QueryClientProvider, { client: queryClient }, children); + + it("should return defined result", () => { + (modelInfoCall as any).mockResolvedValue(mockPageOneResponse); + + const { result } = renderHook(() => useInfiniteModelInfo(), { wrapper }); + + expect(result.current).toBeDefined(); + expect(result.current).toHaveProperty("data"); + expect(result.current).toHaveProperty("fetchNextPage"); + expect(result.current).toHaveProperty("hasNextPage"); + expect(result.current).toHaveProperty("isFetchingNextPage"); + expect(result.current).toHaveProperty("isLoading"); + }); + + it("should return paginated data and call modelInfoCall with page 1 initially", async () => { + (modelInfoCall as any).mockResolvedValue(mockPageOneResponse); + + const { result } = renderHook(() => useInfiniteModelInfo(), { wrapper }); + + expect(result.current.isLoading).toBe(true); + + await waitFor(() => { + expect(result.current.isLoading).toBe(false); + expect(result.current.isSuccess).toBe(true); + }); + + expect(result.current.data?.pages).toHaveLength(1); + expect(result.current.data?.pages[0]).toEqual(mockPageOneResponse); + expect(result.current.hasNextPage).toBe(true); + expect(modelInfoCall).toHaveBeenCalledWith("test-access-token", "test-user-id", "Admin", 1, 50, undefined); + expect(modelInfoCall).toHaveBeenCalledTimes(1); + }); + + it("should use custom size parameter", async () => { + (modelInfoCall as any).mockResolvedValue(mockPageOneResponse); + + const { result } = renderHook(() => useInfiniteModelInfo(25), { wrapper }); + + await waitFor(() => { + expect(result.current.isLoading).toBe(false); + }); + + expect(modelInfoCall).toHaveBeenCalledWith("test-access-token", "test-user-id", "Admin", 1, 25, undefined); + }); + + it("should pass search parameter to modelInfoCall", async () => { + (modelInfoCall as any).mockResolvedValue(mockPageOneResponse); + + const { result } = renderHook(() => useInfiniteModelInfo(50, "gpt"), { wrapper }); + + await waitFor(() => { + expect(result.current.isLoading).toBe(false); + }); + + expect(modelInfoCall).toHaveBeenCalledWith("test-access-token", "test-user-id", "Admin", 1, 50, "gpt"); + }); + + it("should fetch next page when fetchNextPage is called", async () => { + (modelInfoCall as any).mockResolvedValueOnce(mockPageOneResponse).mockResolvedValueOnce(mockPageTwoResponse); + + const { result } = renderHook(() => useInfiniteModelInfo(), { wrapper }); + + await waitFor(() => { + expect(result.current.isSuccess).toBe(true); + expect(result.current.hasNextPage).toBe(true); + }); + + await result.current.fetchNextPage(); + + await waitFor(() => { + expect(result.current.data?.pages).toHaveLength(2); + expect(result.current.data?.pages[1]).toEqual(mockPageTwoResponse); + expect(result.current.hasNextPage).toBe(false); + }); + + expect(modelInfoCall).toHaveBeenNthCalledWith(2, "test-access-token", "test-user-id", "Admin", 2, 50, undefined); + }); + + it("should return undefined for hasNextPage when on last page", async () => { + const lastPageResponse: PaginatedModelInfoResponse = { + ...mockPageOneResponse, + current_page: 1, + total_pages: 1, + }; + (modelInfoCall as any).mockResolvedValue(lastPageResponse); + + const { result } = renderHook(() => useInfiniteModelInfo(), { wrapper }); + + await waitFor(() => { + expect(result.current.isSuccess).toBe(true); + }); + + expect(result.current.hasNextPage).toBe(false); + }); + + it("should handle error when modelInfoCall fails", async () => { + const errorMessage = "Failed to fetch models"; + const testError = new Error(errorMessage); + (modelInfoCall as any).mockRejectedValue(testError); + + const { result } = renderHook(() => useInfiniteModelInfo(), { wrapper }); + + expect(result.current.isLoading).toBe(true); + + await waitFor(() => { + expect(result.current.isLoading).toBe(false); + expect(result.current.isError).toBe(true); + }); + + expect(result.current.error).toEqual(testError); + expect(result.current.data).toBeUndefined(); + expect(modelInfoCall).toHaveBeenCalledTimes(1); + }); + + it("should not execute query when accessToken is missing", () => { + mockUseAuthorized.mockReturnValue({ + accessToken: null, + userId: "test-user-id", + userRole: "Admin", + token: null, + userEmail: "test@example.com", + premiumUser: false, + disabledPersonalKeyCreation: null, + showSSOBanner: false, + }); + + const { result } = renderHook(() => useInfiniteModelInfo(), { wrapper }); + + expect(result.current.isLoading).toBe(false); + expect(result.current.data).toBeUndefined(); + expect(result.current.isFetched).toBe(false); + expect(modelInfoCall).not.toHaveBeenCalled(); + }); + + it("should not execute query when userId is missing", () => { + mockUseAuthorized.mockReturnValue({ + accessToken: "test-access-token", + userId: null, + userRole: "Admin", + token: "test-token", + userEmail: "test@example.com", + premiumUser: false, + disabledPersonalKeyCreation: null, + showSSOBanner: false, + }); + + const { result } = renderHook(() => useInfiniteModelInfo(), { wrapper }); + + expect(result.current.isLoading).toBe(false); + expect(result.current.data).toBeUndefined(); + expect(result.current.isFetched).toBe(false); + expect(modelInfoCall).not.toHaveBeenCalled(); + }); + + it("should not execute query when userRole is missing", () => { + mockUseAuthorized.mockReturnValue({ + accessToken: "test-access-token", + userId: "test-user-id", + userRole: null, + token: "test-token", + userEmail: "test@example.com", + premiumUser: false, + disabledPersonalKeyCreation: null, + showSSOBanner: false, + }); + + const { result } = renderHook(() => useInfiniteModelInfo(), { wrapper }); + + expect(result.current.isLoading).toBe(false); + expect(result.current.data).toBeUndefined(); + expect(result.current.isFetched).toBe(false); + expect(modelInfoCall).not.toHaveBeenCalled(); + }); +}); diff --git a/ui/litellm-dashboard/src/app/(dashboard)/hooks/models/useModels.ts b/ui/litellm-dashboard/src/app/(dashboard)/hooks/models/useModels.ts index c57de675e0e..fe1afdcc39f 100644 --- a/ui/litellm-dashboard/src/app/(dashboard)/hooks/models/useModels.ts +++ b/ui/litellm-dashboard/src/app/(dashboard)/hooks/models/useModels.ts @@ -1,4 +1,4 @@ -import { useQuery } from "@tanstack/react-query"; +import { useQuery, useInfiniteQuery } from "@tanstack/react-query"; import { createQueryKeys } from "../common/queryKeysFactory"; import { modelInfoCall, modelHubCall, modelAvailableCall } from "@/components/networking"; import useAuthorized from "../useAuthorized"; @@ -26,6 +26,7 @@ const modelKeys = createQueryKeys("models"); const modelHubKeys = createQueryKeys("modelHub"); const allProxyModelsKeys = createQueryKeys("allProxyModels"); const selectedTeamModelsKeys = createQueryKeys("selectedTeamModels"); +const infiniteModelKeys = createQueryKeys("infiniteModels"); export const useModelsInfo = (page: number = 1, size: number = 50, search?: string, modelId?: string, teamId?: string, sortBy?: string, sortOrder?: string) => { const { accessToken, userId, userRole } = useAuthorized(); @@ -74,3 +75,38 @@ export const useSelectedTeamModels = (teamID: string | null) => { enabled: Boolean(accessToken && userId && userRole && teamID), }); }; + +export const useInfiniteModelInfo = ( + size: number = 50, + search?: string, +) => { + const { accessToken, userId, userRole } = useAuthorized(); + return useInfiniteQuery({ + queryKey: infiniteModelKeys.list({ + filters: { + ...(userId && { userId }), + ...(userRole && { userRole }), + size, + ...(search && { search }), + }, + }), + queryFn: async ({ pageParam }) => { + return await modelInfoCall( + accessToken!, + userId!, + userRole!, + pageParam as number, + size, + search, + ); + }, + initialPageParam: 1, + getNextPageParam: (lastPage) => { + if (lastPage.current_page < lastPage.total_pages) { + return lastPage.current_page + 1; + } + return undefined; + }, + enabled: Boolean(accessToken && userId && userRole), + }); +}; diff --git a/ui/litellm-dashboard/src/components/ModelSelect/PaginatedModelSelect/PaginatedModelSelect.test.tsx b/ui/litellm-dashboard/src/components/ModelSelect/PaginatedModelSelect/PaginatedModelSelect.test.tsx new file mode 100644 index 00000000000..b91f97885e6 --- /dev/null +++ b/ui/litellm-dashboard/src/components/ModelSelect/PaginatedModelSelect/PaginatedModelSelect.test.tsx @@ -0,0 +1,301 @@ +import { screen, waitFor } from "@testing-library/react"; +import userEvent from "@testing-library/user-event"; +import { beforeEach, describe, expect, it, vi } from "vitest"; +import { renderWithProviders } from "../../../../tests/test-utils"; +import { PaginatedModelSelect } from "./PaginatedModelSelect"; + +const mockFetchNextPage = vi.fn(); + +vi.mock("@/app/(dashboard)/hooks/models/useModels", () => ({ + useInfiniteModelInfo: vi.fn(), +})); + +vi.mock("@tanstack/react-pacer/debouncer", () => { + const React = require("react"); + return { + useDebouncedState: (initial: string) => { + const [value, setValue] = React.useState(initial); + return [value, setValue]; + }, + }; +}); + +import { useInfiniteModelInfo } from "@/app/(dashboard)/hooks/models/useModels"; + +const mockUseInfiniteModelInfo = vi.mocked(useInfiniteModelInfo); + +const mockPagesWithModels = { + pages: [ + { + data: [ + { model_name: "GPT-4", model_info: { id: "model-1" } }, + { model_name: "Claude-3", model_info: { id: "model-2" } }, + ], + total_count: 2, + current_page: 1, + total_pages: 1, + size: 50, + }, + ], +}; + +const mockEmptyPages = { + pages: [{ data: [], total_count: 0, current_page: 1, total_pages: 1, size: 50 }], +}; + +describe("PaginatedModelSelect", () => { + const mockOnChange = vi.fn(); + + const defaultHookReturn = { + data: mockPagesWithModels, + fetchNextPage: mockFetchNextPage, + hasNextPage: false, + isFetchingNextPage: false, + isLoading: false, + }; + + beforeEach(() => { + vi.clearAllMocks(); + mockUseInfiniteModelInfo.mockReturnValue(defaultHookReturn as any); + }); + + it("should render", () => { + renderWithProviders(); + + expect(screen.getByRole("combobox")).toBeInTheDocument(); + expect(screen.getByText("Select a model")).toBeInTheDocument(); + }); + + it("should display custom placeholder when provided", () => { + renderWithProviders( + , + ); + + expect(screen.getByText("Choose model")).toBeInTheDocument(); + }); + + it("should display model options when data is loaded", async () => { + renderWithProviders(); + + const combobox = screen.getByRole("combobox"); + await userEvent.click(combobox); + + await waitFor(() => { + expect(screen.getByRole("option", { name: "GPT-4 (model-1)" })).toBeInTheDocument(); + expect(screen.getByRole("option", { name: "Claude-3 (model-2)" })).toBeInTheDocument(); + }); + }); + + it("should call onChange when user selects a model", async () => { + const user = userEvent.setup({ delay: null }); + renderWithProviders(); + + const combobox = screen.getByRole("combobox"); + await user.click(combobox); + + const visibleOption = await screen.findByTitle("GPT-4 (model-1)"); + await user.click(visibleOption); + + await waitFor(() => { + expect(mockOnChange).toHaveBeenCalledWith("model-1"); + }); + }); + + it("should display selected value when value prop is provided", async () => { + renderWithProviders( + , + ); + + const combobox = screen.getByRole("combobox"); + await userEvent.click(combobox); + + await waitFor(() => { + expect(screen.getByRole("option", { name: "GPT-4 (model-1)" })).toBeInTheDocument(); + }); + }); + + it("should show loading state when isLoading is true", () => { + mockUseInfiniteModelInfo.mockReturnValue({ + ...defaultHookReturn, + isLoading: true, + } as any); + + renderWithProviders(); + + expect(screen.getByRole("combobox")).toHaveAttribute("aria-expanded", "false"); + }); + + it("should pass pageSize to useInfiniteModelInfo", () => { + renderWithProviders(); + + expect(mockUseInfiniteModelInfo).toHaveBeenCalledWith(25, undefined); + }); + + it("should pass search to useInfiniteModelInfo when user types", async () => { + const user = userEvent.setup(); + renderWithProviders(); + + const combobox = screen.getByRole("combobox"); + await user.click(combobox); + await user.keyboard("gpt"); + + await waitFor(() => { + expect(mockUseInfiniteModelInfo).toHaveBeenCalledWith(50, "gpt"); + }); + }); + + it("should have scroll container for infinite loading when hasNextPage is true", async () => { + mockUseInfiniteModelInfo.mockReturnValue({ + ...defaultHookReturn, + hasNextPage: true, + isFetchingNextPage: false, + } as any); + + renderWithProviders(); + + const combobox = screen.getByRole("combobox"); + await userEvent.click(combobox); + + await waitFor(() => { + expect(screen.getByRole("option", { name: "GPT-4 (model-1)" })).toBeInTheDocument(); + }); + + const scrollableContainer = document.querySelector( + ".ant-select-dropdown .rc-virtual-list-holder", + ); + expect(scrollableContainer).toBeInTheDocument(); + expect(scrollableContainer).toHaveAttribute("style"); + }); + + it("should deduplicate models with same id across pages", async () => { + mockUseInfiniteModelInfo.mockReturnValue({ + ...defaultHookReturn, + data: { + pages: [ + { + data: [ + { model_name: "GPT-4", model_info: { id: "model-1" } }, + { model_name: "GPT-4 Dupe", model_info: { id: "model-1" } }, + ], + total_count: 2, + current_page: 1, + total_pages: 1, + size: 50, + }, + ], + }, + fetchNextPage: mockFetchNextPage, + hasNextPage: false, + isFetchingNextPage: false, + isLoading: false, + } as any); + + renderWithProviders(); + + const combobox = screen.getByRole("combobox"); + await userEvent.click(combobox); + + await waitFor(() => { + const model1Options = screen.queryAllByRole("option", { name: /model-1/ }); + expect(model1Options.length).toBe(1); + }); + }); + + it("should skip models without model_info id", async () => { + mockUseInfiniteModelInfo.mockReturnValue({ + ...defaultHookReturn, + data: { + pages: [ + { + data: [ + { model_name: "Valid Model", model_info: { id: "valid-id" } }, + { model_name: "No ID", model_info: null }, + { model_name: "Empty ID", model_info: { id: "" } }, + ], + total_count: 3, + current_page: 1, + total_pages: 1, + size: 50, + }, + ], + }, + fetchNextPage: mockFetchNextPage, + hasNextPage: false, + isFetchingNextPage: false, + isLoading: false, + } as any); + + renderWithProviders(); + + const combobox = screen.getByRole("combobox"); + await userEvent.click(combobox); + + await waitFor(() => { + expect(screen.getByRole("option", { name: "Valid Model (valid-id)" })).toBeInTheDocument(); + expect(screen.queryByRole("option", { name: "No ID" })).not.toBeInTheDocument(); + expect(screen.queryByRole("option", { name: "Empty ID" })).not.toBeInTheDocument(); + }); + }); + + it("should show model ID only when model_name is empty", async () => { + mockUseInfiniteModelInfo.mockReturnValue({ + ...defaultHookReturn, + data: { + pages: [ + { + data: [{ model_name: "", model_info: { id: "id-only" } }], + total_count: 1, + current_page: 1, + total_pages: 1, + size: 50, + }, + ], + }, + fetchNextPage: mockFetchNextPage, + hasNextPage: false, + isFetchingNextPage: false, + isLoading: false, + } as any); + + renderWithProviders(); + + const combobox = screen.getByRole("combobox"); + await userEvent.click(combobox); + + await waitFor(() => { + expect(screen.getByRole("option", { name: "id-only" })).toBeInTheDocument(); + }); + }); + + it("should respect allowClear prop", () => { + renderWithProviders( + , + ); + + expect(screen.getByRole("combobox")).toBeInTheDocument(); + }); + + it("should respect disabled prop", () => { + renderWithProviders(); + + const combobox = screen.getByRole("combobox"); + expect(combobox.closest(".ant-select")).toHaveClass("ant-select-disabled"); + }); + + it("should not call fetchNextPage when hasNextPage is false", async () => { + mockUseInfiniteModelInfo.mockReturnValue({ + ...defaultHookReturn, + hasNextPage: false, + } as any); + + renderWithProviders(); + + await userEvent.click(screen.getByRole("combobox")); + + await waitFor(() => { + expect(screen.getByRole("option", { name: "GPT-4 (model-1)" })).toBeInTheDocument(); + }); + + expect(mockFetchNextPage).not.toHaveBeenCalled(); + }); +}); diff --git a/ui/litellm-dashboard/src/components/ModelSelect/PaginatedModelSelect/PaginatedModelSelect.tsx b/ui/litellm-dashboard/src/components/ModelSelect/PaginatedModelSelect/PaginatedModelSelect.tsx new file mode 100644 index 00000000000..9b22fd1bd87 --- /dev/null +++ b/ui/litellm-dashboard/src/components/ModelSelect/PaginatedModelSelect/PaginatedModelSelect.tsx @@ -0,0 +1,143 @@ +import { useInfiniteModelInfo } from "@/app/(dashboard)/hooks/models/useModels"; +import { LoadingOutlined } from "@ant-design/icons"; +import { useDebouncedState } from "@tanstack/react-pacer/debouncer"; +import { Select, Space, Typography } from "antd"; +import { useMemo, useState, type UIEvent } from "react"; + +const { Text } = Typography; + +export interface PaginatedModelSelectProps { + value?: string; + onChange?: (value: string) => void; + placeholder?: string; + style?: React.CSSProperties; + pageSize?: number; + allowClear?: boolean; + disabled?: boolean; +} + +const SCROLL_THRESHOLD = 0.8; +const DEBOUNCE_MS = 300; + +export const PaginatedModelSelect = ({ + value, + onChange, + placeholder = "Select a model", + style, + pageSize = 50, + allowClear = true, + disabled = false, +}: PaginatedModelSelectProps) => { + const [searchInput, setSearchInput] = useState(""); + const [debouncedSearch, setDebouncedSearch] = useDebouncedState("", { + wait: DEBOUNCE_MS, + }); + + const { + data, + fetchNextPage, + hasNextPage, + isFetchingNextPage, + isLoading, + } = useInfiniteModelInfo(pageSize, debouncedSearch || undefined); + + const options = useMemo(() => { + if (!data?.pages) return []; + + const seen = new Set(); + const result: { label: string; value: string; modelName: string; modelId: string }[] = []; + + for (const page of data.pages) { + for (const model of page.data) { + const modelId = model.model_info?.id ?? ""; + const modelName = model.model_name ?? ""; + + // Dedupe by id - skip models without id (can't uniquely identify) + if (!modelId || seen.has(modelId)) continue; + seen.add(modelId); + + result.push({ + label: modelName ? `${modelName} (${modelId})` : modelId, + value: modelId, + modelName, + modelId, + }); + } + } + + return result; + }, [data]); + + const optionRender = (option: { data: { modelName: string; modelId: string; label: string } }) => { + const { modelName, modelId } = option.data; + + return ( + <> + {modelName ? ( + + + Model name: + {modelName} + + + Model ID: {modelId} + + + ) : ( + Model ID: {modelId} + )} + + ); + }; + + const handlePopupScroll = (e: UIEvent) => { + const target = e.currentTarget; + const scrollRatio = + (target.scrollTop + target.clientHeight) / target.scrollHeight; + + if (scrollRatio >= SCROLL_THRESHOLD && hasNextPage && !isFetchingNextPage) { + fetchNextPage(); + } + }; + + const handleSearch = (value: string) => { + setSearchInput(value); + setDebouncedSearch(value); + }; + + const handleChange = (v: string | string[] | null) => { + const normalized = + typeof v === "string" ? v : Array.isArray(v) ? v[0] ?? "" : ""; + onChange?.(normalized); + }; + + return ( +
@@ -632,6 +670,27 @@ const CustomCodeModal: React.FC = ({
+ {/* Contribution CTA Banner */} +
+
+
+ +
+
+
Built a useful guardrail?
+
Share it with the community and help others build faster
+
+
+ +
+ {/* Primitives Panel */} diff --git a/ui/litellm-dashboard/src/components/mcp_tools/OAuthFormFields.tsx b/ui/litellm-dashboard/src/components/mcp_tools/OAuthFormFields.tsx new file mode 100644 index 00000000000..acad72cb21c --- /dev/null +++ b/ui/litellm-dashboard/src/components/mcp_tools/OAuthFormFields.tsx @@ -0,0 +1,166 @@ +import React from "react"; +import { Form, Select, Tooltip } from "antd"; +import { InfoCircleOutlined } from "@ant-design/icons"; +import { Button, TextInput } from "@tremor/react"; +import { OAUTH_FLOW } from "./types"; + +interface OAuthFlowStatus { + startOAuthFlow: () => void; + status: string; + error: string | null; + tokenResponse: { access_token?: string; expires_in?: number } | null; +} + +interface OAuthFormFieldsProps { + isM2M: boolean; + isEditing?: boolean; + oauthFlow?: OAuthFlowStatus; + initialFlowType?: string; +} + +const fieldClassName = "rounded-lg border-gray-300 focus:border-blue-500 focus:ring-blue-500"; + +const FieldLabel: React.FC<{ label: string; tooltip: string }> = ({ label, tooltip }) => ( + + {label} + + + + +); + +const OAuthFormFields: React.FC = ({ + isM2M, + isEditing = false, + oauthFlow, + initialFlowType, +}) => { + const placeholderSuffix = isEditing ? " (leave blank to keep existing)" : ""; + + return ( + <> + + } + name="oauth_flow_type" + {...(initialFlowType ? { initialValue: initialFlowType } : {})} + > + + + + {isM2M ? ( + <> + } + name={["credentials", "client_id"]} + rules={[{ required: true, message: "Client ID is required for M2M OAuth" }]} + > + + + } + name={["credentials", "client_secret"]} + rules={[{ required: true, message: "Client Secret is required for M2M OAuth" }]} + > + + + } + name="token_url" + rules={[{ required: true, message: "Token URL is required for M2M OAuth" }]} + > + + + } + name={["credentials", "scopes"]} + > + + + } + name="authorization_url" + > + + + } + name="token_url" + > + + + } + name="registration_url" + > + + + {oauthFlow && ( +
+

+ Use OAuth to fetch a fresh access token and temporarily save it in the session as the authentication value. +

+ + {oauthFlow.error &&

{oauthFlow.error}

} + {oauthFlow.status === "success" && oauthFlow.tokenResponse?.access_token && ( +

+ Token fetched. Expires in {oauthFlow.tokenResponse.expires_in ?? "?"} seconds. +

+ )} +
+ )} + + )} + + ); +}; + +export default OAuthFormFields; diff --git a/ui/litellm-dashboard/src/components/mcp_tools/StdioConfiguration.tsx b/ui/litellm-dashboard/src/components/mcp_tools/StdioConfiguration.tsx index 23f5f84fead..476a5b61683 100644 --- a/ui/litellm-dashboard/src/components/mcp_tools/StdioConfiguration.tsx +++ b/ui/litellm-dashboard/src/components/mcp_tools/StdioConfiguration.tsx @@ -4,9 +4,14 @@ import { InfoCircleOutlined } from "@ant-design/icons"; interface StdioConfigurationProps { isVisible: boolean; + /** + * When true, stdio_config is required + validated as JSON. + * Edit screen can set this to false when using dedicated command/args/env fields. + */ + required?: boolean; } -const StdioConfiguration: React.FC = ({ isVisible }) => { +const StdioConfiguration: React.FC = ({ isVisible, required = true }) => { if (!isVisible) return null; return ( @@ -21,7 +26,7 @@ const StdioConfiguration: React.FC = ({ isVisible }) => } name="stdio_config" rules={[ - { required: true, message: "Please enter stdio configuration" }, + ...(required ? [{ required: true, message: "Please enter stdio configuration" }] : []), { validator: (_, value) => { if (!value) return Promise.resolve(); diff --git a/ui/litellm-dashboard/src/components/mcp_tools/create_mcp_server.tsx b/ui/litellm-dashboard/src/components/mcp_tools/create_mcp_server.tsx index a56cbaf492a..ff476bd999d 100644 --- a/ui/litellm-dashboard/src/components/mcp_tools/create_mcp_server.tsx +++ b/ui/litellm-dashboard/src/components/mcp_tools/create_mcp_server.tsx @@ -3,7 +3,8 @@ import { Modal, Tooltip, Form, Select, Input } from "antd"; import { InfoCircleOutlined } from "@ant-design/icons"; import { Button, TextInput } from "@tremor/react"; import { createMCPServer } from "../networking"; -import { AUTH_TYPE, MCPServer, MCPServerCostInfo } from "./types"; +import { AUTH_TYPE, OAUTH_FLOW, MCPServer, MCPServerCostInfo } from "./types"; +import OAuthFormFields from "./OAuthFormFields"; import MCPServerCostConfig from "./mcp_server_cost_config"; import MCPConnectionStatus from "./mcp_connection_status"; import MCPToolConfiguration from "./mcp_tool_configuration"; @@ -52,6 +53,7 @@ const CreateMCPServer: React.FC = ({ const authType = formValues.auth_type as string | undefined; const shouldShowAuthValueField = authType ? AUTH_TYPES_REQUIRING_AUTH_VALUE.includes(authType) : false; const isOAuthAuthType = authType === AUTH_TYPE.OAUTH2; + const isM2MFlow = isOAuthAuthType && formValues.oauth_flow_type === OAUTH_FLOW.M2M; const persistCreateUiState = () => { if (typeof window === "undefined") { @@ -477,7 +479,7 @@ const CreateMCPServer: React.FC = ({ rules={[ { required: false, - message: "Please enter a server description!!!!!!!!!", + message: "Please enter a server description", }, ]} > @@ -561,131 +563,16 @@ const CreateMCPServer: React.FC = ({ )} {transportType !== "stdio" && isOAuthAuthType && ( - <> - - OAuth Client ID (optional) - - - - - } - name={["credentials", "client_id"]} - > - - - - OAuth Client Secret (optional) - - - - - } - name={["credentials", "client_secret"]} - > - - - - OAuth Scopes (optional) - - - - - } - name={["credentials", "scopes"]} - > - = ({ }, ]} > - setAliasManuallyEdited(true)} /> + setAliasManuallyEdited(true)} + className="rounded-lg border-gray-300 focus:border-blue-500 focus:ring-blue-500" + /> - - - validateMCPServerUrl(value) }, - ]} - > - + - Server-Sent Events (SSE) HTTP - - - - - {shouldShowAuthValueField && ( + {/* URL/Auth fields are only applicable for HTTP/SSE */} + {!isStdioTransport && ( + validateMCPServerUrl(value) }, + ]} + > + + + )} + + {!isStdioTransport && ( + + + + )} + + {isStdioTransport && ( +
+

+ Configure the stdio transport used to launch the MCP server process. You can either fill in the fields + below or paste a JSON configuration. +

+ + + + + + + + ) : option.customComponent ? ( + (() => { + const CustomComponent = option.customComponent; + return ( + handleFilterChange(option.name, value ?? "")} + placeholder={`Select ${option.label || option.name}...`} + /> + ); + })() ) : (