The gate at three call sites was calling _supports_factory with
custom_llm_provider=None, which relies on get_llm_provider inferring the
provider from the model string. For the invoke path, the model still
carries an 'invoke/' routing prefix (e.g. 'invoke/us.anthropic.claude-opus-4-6-v1')
that is not a known provider, so inference raises BadRequestError,
_supports_factory swallows it and returns False, and the user's
output_config.effort gets silently dropped before the Bedrock request.
Strip the routing prefix with the existing strip_bedrock_routing_prefix
helper and pass custom_llm_provider='bedrock' explicitly so the
declarative 'supports_output_config' flag in
model_prices_and_context_window.json is the actual source of truth.
Also adds a regression test that exercises the full
'invoke/us.anthropic.claude-opus-4-6-v1' path and asserts output_config
survives.
The partner/gemma/model-garden handlers now call self._ensure_access_token
(inherited from VertexBase) instead of instantiating VertexLLM, so tests
must patch VertexBase._ensure_access_token for the mock to take effect.
* Add Rubrik as officially-supported guardrail plugin
Adds tool blocking and batch logging integration with an external Rubrik
webhook service. The plugin validates LLM tool calls against a policy
service (fail-open on errors) and batch-logs all requests/responses.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* Update Rubrik docs: config.yaml as primary, env vars as fallback
Restructures the Quick Start to present config.yaml as the recommended
approach with tabbed UI, and environment variables as an alternative
fallback.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* Add Rubrik env vars to config_settings reference
Fixes documentation validation by adding RUBRIK_API_KEY,
RUBRIK_BATCH_SIZE, RUBRIK_SAMPLING_RATE, and RUBRIK_WEBHOOK_URL
to the environment settings reference table.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* Add fallback message when blocking service returns empty explanation
Prevents whitespace-only violation message when the tool blocking
service blocks tools but returns an empty content field.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
When litellm migrated from Poetry to uv (PR #24905, v1.83.1), the core
dependency specifications in pyproject.toml changed from Poetry bare-version
strings (e.g. openai = "2.30.0") to PEP 621 exact pins (openai==2.24.0).
Poetry bare-version strings are actually caret ranges (^X.Y.Z == >=X.Y.Z,<X+1),
but PEP 621 == is exact. This means every downstream package that installs
litellm as a library dependency is now forced to downgrade aiohttp, pydantic,
openai, click, and 8 other common packages to exact old versions.
Fix: restore range specifiers for the 12 core runtime dependencies. The
optional extras (proxy, proxy-runtime, etc.) are consumed primarily by
Docker images where exact pins are appropriate and are left unchanged.
The uv.lock file continues to provide exact reproducibility for Docker
builds and CI.
Fixes: #26154
- preserve existing shared backend `mode` when router deployment registration
reuses a provider/model key already in `litellm.model_cost` (prevents alias
with `mode: chat` from downgrading shared `chatgpt/gpt-5.4` from `responses`
to `chat` and triggering 403s on /v1/chat/completions)
- teach the ChatGPT Responses parser to recover `response.output_item.done`
entries when `response.completed.output` is empty
- add defensive /responses -> /chat/completions bridge fallback that
reconstructs output items from raw SSE when `raw_response.output` is empty
- regression coverage for shared alias routing, empty completed.output
parsing, and SSE bridge recovery
Closes#25403
Co-authored-by: afoninsky <andrey.afoninsky@gmail.com>
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* fix: reuse cached credentials in VertexAIPartnerModels instead of creating new VertexLLM per request
VertexAIPartnerModels.completion() was creating a throwaway VertexLLM()
instance on every call to get an access token, bypassing the credential
cache inherited from VertexBase. This caused a fresh token fetch for
every single request, adding significant latency overhead.
Fix: call super().__init__() to initialize VertexBase's credential cache,
and use self._ensure_access_token() instead of a new VertexLLM instance.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* fix: apply same credential caching fix to VertexAIGemmaModels and VertexAIModelGardenModels
Same bug as VertexAIPartnerModels: both classes had `pass` in __init__
instead of `super().__init__()`, and created throwaway VertexLLM()
instances per request instead of using self._ensure_access_token().
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* fix(vertex_ai): single-flight credential refresh to prevent thundering herd
When GCP credentials expire under high concurrency, all requests
simultaneously call credentials.refresh() via asyncify, saturating the
40-thread anyio pool and blocking the proxy for 20+ seconds.
This adds:
- Per-credential asyncio.Lock in get_access_token_async for single-flight
refresh (1 coroutine refreshes, others wait on the lock)
- Background refresh when token_state is STALE (usable but near expiry),
returning the current token immediately with zero added latency
- threading.Lock on the sync get_access_token path
- Uses google-auth's TokenState enum (FRESH/STALE/INVALID) instead of
reimplementing expiry logic
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* fix: address PR review comments
- Use asyncio.create_task() instead of deprecated get_event_loop().create_task()
- Track in-flight background refresh tasks to prevent duplicate refreshes
when multiple STALE-path callers pass through the lock before the first
background task completes
- Add token validation in the STALE branch (consistent with FRESH/INVALID)
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* fix: lazy-import TokenState to avoid breaking when google-auth is not installed
Also extract helper methods to bring get_access_token_async under the
PLR0915 statement limit (50).
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* chore: apply Black formatting to test file and update uv.lock
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* fix: remove user-provided project_id from log messages (CodeQL log injection)
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* fix: avoid leaking token value in error message, log type instead
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* chore: restore uv.lock to match litellm_oss_branch
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* fix: remove project_id from remaining log message (CodeQL log injection)
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* fix: remove remaining project_id from log and error messages
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Replace hardcoded _is_claude_4_6_model() string matching with
supports_output_config flag in model_prices_and_context_window.json,
accessed via _supports_factory(). This follows the project's established
pattern for model capability checks (per AGENTS.md rule #8).
Bedrock Invoke now conditionally preserves output_config for models
that declare supports_output_config=true (currently Claude 4.6 models),
while stripping it for older models to avoid request rejection.
Ref: https://github.com/BerriAI/litellm/issues/22797
Members tab column reads this field; dropping it from the type in the
previous revert broke the type check without affecting the reverted
render logic.
MCP server CRUD endpoints (/v1/mcp/server*) were bundled with MCP
tool-call / passthrough endpoints under llm_api_routes, so setting
DISABLE_LLM_API_ENDPOINTS=true on admin-only nodes also blocked the
Admin UI from listing, adding, or attaching MCP servers.
Separate mcp_inference_routes (data-plane, gated by
DISABLE_LLM_API_ENDPOINTS) from mcp_management_routes (control-plane,
gated by DISABLE_ADMIN_ENDPOINTS). Keep mcp_routes as a union for
backward compat with allowed_routes=["mcp_routes"] virtual key configs.
Upgrade is_management_route to pattern-aware matching so
/v1/mcp/server/{path:path} resolves for concrete IDs.
The `_test-unit-services-base.yml` reusable workflow attached every job
to the `integration-postgres` GHA environment to read three "secrets":
DATABASE_URL, POSTGRES_USER, POSTGRES_PASSWORD. These are not secrets —
the postgres service container is spawned per-job on localhost and
destroyed with the job, so the user/password are bootstrap values for a
throwaway container and the URL is always `postgresql://…@localhost:…`.
Each environment attachment produces a "temporarily deployed to
integration-postgres" deployment record, which the PR timeline renders
as a message per matrix shard per push. With 14 proxy-db shards that's
~14 notifications per push, drowning the PR conversation.
Changes:
* Hardcode POSTGRES_USER/POSTGRES_PASSWORD/POSTGRES_DB and the derived
DATABASE_URL in `_test-unit-services-base.yml`.
* Delete the `environment: integration-postgres` attachment.
* Delete the `secrets:` declarations on the reusable workflow and on
the two callers (test-unit-proxy-db.yml, test-unit-security.yml).
* The `services:` container still starts a fresh postgres per job;
the connection string now matches what the container boots up with.
Security review: no regression. The environment wasn't gating anything
real — no protection rules configured, no approval gates, and the
branch restriction is already enforced by `on: push: branches: [...]`
on both caller workflows. Zizmor pedantic-mode findings are identical
before and after (same 6 pre-existing findings, zero new ones).
The `integration-postgres` environment and its three "secrets" in repo
settings are now unreferenced and can be deleted from repo admin.
test_db_schema_migration.py has exactly one test, and that test is mostly
waiting on prisma subprocesses (~170s: prisma migrate deploy + prisma
migrate diff). No CPU-bound Python work inside the test body, and only
one test in the file means xdist's parallelism is unused regardless.
Previous run on commit 5df9f397e6: 10.0m wall-clock for the shard, of
which 4:56 was silence between step start and pytest banner — the cost
of 4 xdist workers each cold-starting (pytest plugin load + litellm
import + pytest-cov instrumentation) so that exactly one of them could
pick up the single test.
Switching to workers: 0 takes the serial pytest branch in the base
workflow, which already handles this case correctly (no -n, no --dist).
Single-process startup instead of 4. Expected wall-clock: ~6m.
Temporary MCP OAuth sessions were kept in process-local memory, so on
multi-instance/LB proxy deployments a session created on instance A could
not be found when the follow-up /server/oauth/{server_id}/... request
landed on instance B.
Persist temporary session records to Redis (encrypted with the existing
proxy encryption helpers) as a best-effort L2 cache alongside the current
in-memory L1. Convert get_cached_temporary_mcp_server to async and await
it from the authorize/token/register OAuth endpoints.
Made-with: Cursor
Vertex multi-region endpoints (e.g. us, eu) use the rep host pattern, not
{geo}-aiplatform.googleapis.com. Regional IDs still contain a hyphen.
common_utils.get_vertex_base_url centralizes the rule for SDK/API URL building.
Proxy pass-through duplicates the same branching in a local get_vertex_base_url
(with trailing slashes) to avoid importing from common_utils there; live
WebSocket passthrough uses the same multi-region host logic for wss://.
Tests cover us/eu for the common_utils helper.
Made-with: Cursor
Relative labels ("today", "in 2 days", "on May 12, 2026") mixed three
shapes in one column, breaking scannability. Always render MMM D, YYYY
for consistency and easier at-a-glance comparison across members.
Two changes:
1. workers: 8 -> 4 on every non-serial proxy-db shard. ubuntu-latest is a
4-core runner; -n 8 oversubscribes 2x and workers block each other
during their cold-start imports (pytest-cov instruments every litellm
module per worker). Measured ~441% CPU locally with -n 8 on 8 cores
(i.e. ~55% effective). Matching -n to physical cores should give
~2x faster worker startup, which is where most of the ~9m wall-clock
per shard goes (7+ minutes is plugin load + xdist imports before any
test runs).
2. Revert the -k split on test_proxy_utils.py. It was split into
proxy-utils-a-h / proxy-utils-i-z as a semantic-adjacent hack; merge
back to a single proxy-utils shard. Still uses --dist=worksteal so
xdist can balance the 188 parametrized cases across workers.
Also drops the now-unused `keyword` input from _test-unit-services-base.yml
and its matching matrix field across all proxy-db entries.
Shard count: 14 -> 13 (+ the assert-shard-coverage guard).
Default GHA matrix job names join every matrix field, producing unreadable
check labels like:
'proxy-db (logging-misc, tests/proxy_unit_tests/test_proxy_reject_logging.py
tests/proxy_unit_tests/test_audit_logs_proxy.py ..., 8, loadscope, "", 15)'
Set the job's display name to '${{ matrix.test-group }}' so each check
shows just 'logging-misc', 'proxy-utils-a-h', etc.
Previous run (13.8m total) was bottlenecked by shards with 9-12m wall-clock.
Setup + xdist spawn + coverage teardown is ~3m per shard, so each shard's
pytest runtime must stay under ~4m to fit inside 7m total.
Observed per-shard pytest times (before split):
db-and-spend 9:08 (170s outlier: test_aaaasschema_migration_check)
proxy-server 7:15
logging-and-callbacks 6:45
guardrails-budget-hooks 6:37
proxy-utils 6:23
auth-and-jwt 6:54
Split 6 shards into 12, keeping key-generation and endpoints-and-responses
(already <7m). Adds a `keyword` input to _test-unit-services-base.yml so
test_proxy_utils.py can be split by -k expression (same file, two runners).
New matrix entries:
auth-and-jwt -> auth-checks + jwt-and-keys
proxy-server -> proxy-server-core + proxy-runtime
logging-and-callbacks -> custom-logging + logging-misc
db-and-spend -> schema-migration (isolated 170s test) + db-and-spend
guardrails-budget-hooks-> guardrails-hooks + budgets
proxy-utils -> proxy-utils-a-h + proxy-utils-i-z (-k split)
The -k expression split is verified to cover every one of the 64 test
functions in test_proxy_utils.py exactly once. The assert-shard-coverage
guard still catches any file not in any shard.
The create-branch job in create-release.yml calls the reusable
create-release-branch.yml workflow, which requires contents: write.
The top-level permissions: {} blocks the inherited default, and only
the release job overrode it, so the nested call failed with:
The nested job 'create-branch' is requesting 'contents: write',
but is only allowed 'contents: none'.
Add the permission at the calling job level so the reusable
workflow is granted what it needs.
The previous `while not self._queue.empty(): await self._queue.join()`
pattern skipped the join entirely when the worker had already dequeued a
task but not yet called task_done(). asyncio.Queue.join() tracks
_unfinished_tasks (incremented by put, decremented by task_done), not
queue depth, so it already handles that case on its own.
Two fixes to proxy-db CI:
1. test_realtime_webrtc_endpoints.py's `proxy_app` fixture mutated the
module-global `proxy_server.master_key` without restoring it, leaking
state into any test that shared the same xdist worker. Under
--dist=loadscope with 2 workers (GHA proxy-endpoints), this caused the
google_endpoints tests to fail with "No api key passed in." because
user_api_key_auth saw a set master_key and a missing API key on the
test request. The fixture now saves and restores the original value.
2. Address the Greptile note that the semantic shard design has no
catch-all, so a new test file added to tests/proxy_unit_tests/ without
a matrix entry would silently skip CI. Adds an assert-shard-coverage
job that enumerates test_*.py files and fails the workflow if any are
not referenced by a matrix entry, with a clear message telling the
author which semantic shard to place it in. All proxy-db shards now
depend on this guard.
Split into two related cleanups:
1. Delete CCI jobs that duplicate GHA coverage:
- mcp_testing (tests/mcp_tests) — already run by test-mcp.yml
- litellm_mapped_tests_proxy_part1/part2 (tests/test_litellm/proxy) —
already run across test-unit-proxy-auth.yml, test-unit-proxy-endpoints.yml,
and test-unit-proxy-infra.yml
Add rag_endpoints and realtime_endpoints to test-unit-proxy-endpoints.yml
(they were only covered by the deleted CCI part2 job).
Remove the corresponding workflow wiring, coverage combine entries, and
upload-coverage dependencies in .circleci/config.yml.
2. Re-shard test-unit-proxy-db.yml from 4 alphabetic buckets to 8 semantic
ones (auth-and-jwt, proxy-server, logging-and-callbacks, db-and-spend,
guardrails-budget-hooks, endpoints-and-responses, plus the existing
serial key-generation and test_proxy_utils.py shards). New test files are
placed in whichever group they belong to instead of reshuffling slices.
Add a dist input to _test-unit-services-base.yml so the test_proxy_utils.py
shard can use --dist=worksteal to spread its ~64 (many parametrized)
functions across workers; the default --dist=loadscope pins a single file
to a single worker, which was the root cause of that shard running 10m+.
The mocked async_increment_cache_pipeline is invoked from Router's
deployment_callback_on_success, registered as an async success callback.
Those callbacks are enqueued to GLOBAL_LOGGING_WORKER and run on a
background task, so the mock may not have been called yet when the test
asserts on it. Flush the worker before asserting.
