* fix(bedrock): respect s3_region_name for batch file uploads (#23569)
* fix(bedrock): respect s3_region_name for batch file uploads (GovCloud fix)
* fix: s3_region_name always wins over aws_region_name for S3 signing (Greptile feedback)
* fix: _filter_headers_for_aws_signature - Bedrock KB (#23571)
* fix: _filter_headers_for_aws_signature
* fix: filter None header values in all post-signing re-merge paths
Addresses Greptile feedback: None-valued headers were being filtered
during SigV4 signing but re-merged back into the final headers dict
afterward, which would cause downstream HTTP client failures.
Made-with: Cursor
* feat(router): tag_regex routing — route by User-Agent regex without per-developer tag config (#23594)
* feat(router): add tag_regex support for header-based routing
Adds a new `tag_regex` field to litellm_params that lets operators route
requests based on regex patterns matched against request headers — primarily
User-Agent — without requiring per-developer tag configuration.
Use case: route all Claude Code traffic (User-Agent: claude-code/x.y.z) to
a dedicated deployment by setting:
tag_regex:
- "^User-Agent: claude-code\\/"
in the deployment's litellm_params. Works alongside existing `tags` routing;
exact tag match takes precedence over regex match. Unmatched requests fall
through to deployments tagged `default`.
The matched deployment, pattern, and user_agent are recorded in
`metadata["tag_routing"]` so they flow through to SpendLogs automatically.
* fix(tag_regex): address backwards-compat, metadata overwrite, and warning noise
Three issues from code review:
1. Backwards-compat: `has_tag_filter` was widened to activate on any non-empty
User-Agent, which would raise ValueError for existing deployments using plain
tags without a `default` fallback. Fix: only activate header-based regex
filtering when at least one candidate deployment has `tag_regex` configured.
2. Metadata overwrite: `metadata["tag_routing"]` was overwritten for every
matching deployment in the loop, leaving inaccurate provenance when multiple
deployments match. Fix: write only for the first match.
3. Warning noise: an invalid regex pattern logged one warning per header string
rather than once per pattern. Fix: compile first (catching re.error once),
then iterate over header strings.
Also adds two new tests covering these cases, and adds docs page for
tag_regex routing with a Claude Code walk-through.
* refactor(tag_regex): remove unnecessary _healthy_list copy
* docs: merge tag_regex section into tag_routing.md, remove standalone page
- Add ## Regex-based tag routing (tag_regex) section to existing
tag_routing.md instead of a separate page
- Remove tag_regex_routing.md standalone doc (odd UX to have a separate
page for a sub-feature)
- Remove proxy/tag_regex_routing from sidebars.js
- Add match_any=False debug warning in tag_based_routing.py when regex
routing fires under strict mode (regex always uses OR semantics)
* fix(tag_regex): address greptile review - security docs, strict-mode enforcement, validation order
- Strengthen security note in tag_routing.md: explicitly state User-Agent
is client-supplied and can be set to any value; frame tag_regex as a
traffic classification hint, not an access-control mechanism
- Move tag_regex startup validation before _add_deployment() so an invalid
pattern never leaves partial router state
- Enforce match_any=False strict-tag policy: when a deployment has both
tags and tag_regex and the strict tag check fails, skip the regex fallback
rather than silently bypassing the operator's intent
- Extract per-deployment match logic into _match_deployment() helper to
keep get_deployments_for_tag() readable
- Add two new tests: strict-mode blocks regex fallback, regex-only
deployment still matches under match_any=False
* fix(ci): apply Black formatting to 14 files and stabilize flaky caplog tests
- Run Black formatter on 14 files that were failing the lint check
- Replace caplog-based assertions in TestAliasConflicts with
unittest.mock.patch on verbose_logger.warning for xdist compatibility
- The caplog fixture can produce empty text in pytest-xdist workers
in certain CI environments, causing flaky test failures
Co-authored-by: Ishaan Jaff <ishaan-jaff@users.noreply.github.com>
---------
Co-authored-by: Cursor Agent <cursoragent@cursor.com>
Co-authored-by: Ishaan Jaff <ishaan-jaff@users.noreply.github.com>
* PR #22867 added _remove_scope_from_cache_control for Bedrock and Azure AI but omitted Vertex AI. This applies the same pattern to VertexAIPartnerModelsAnthropicMessagesConfig."
* PR #22867 added _remove_scope_from_cache_control for Bedrock and Azure AI but omitted Vertex AI. This applies the same pattern to VertexAIPartnerModelsAnthropicMessagesConfig."
* PR #22867 added _remove_scope_from_cache_control to AzureAnthropicMessagesConfig
but missed VertexAIPartnerModelsAnthropicMessagesConfi Rather than duplicating the method again, moved it up to the base AnthropicMessagesConfig so all providers
inherit it, and removed the now-redundant copy from the Azure AI subclass.
* PR #22867 added _remove_scope_from_cache_control to AzureAnthropicMessagesConfig
but missed VertexAIPartnerModelsAnthropicMessagesConfi Rather than duplicating the method again, moved it up to the base AnthropicMessagesConfig so all providers
inherit it, and removed the now-redundant copy from the Azure AI subclass.
---------
Co-authored-by: Krish Dholakia <krrishdholakia@gmail.com>
Tests were outdated after #23472 added pagination (take/order) to find_many
and stale-row cleanup via update_many. Updated assertions to match new call
signatures.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The self-exclusion filter compared raw key strings against SHA-256
hashed tokens from the DB, so keys were never excluded and
double-counting persisted. Now hash data.key before comparison.
Also add tpm_limit_type/rpm_limit_type to _throughput_fields_changed
guard, fall back to existing_key_row.team_id for team limit checks
(matching the org pattern), and add team self-exclusion test.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Only run org validation (get_org_object + _check_org_key_limits) when
the update actually touches throughput-related fields (tpm_limit,
rpm_limit, or organization_id). Previously, any update to a key
belonging to an org would trigger the check, which would fail with a
400 if the org had been deleted — blocking unrelated field changes.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
When updating a key, _check_org_key_limits and _check_team_key_limits
would include the key being updated in the find_many results, causing
its current limits to be counted twice (once from the DB query, once
from the new requested limits). This caused false 400 errors on valid
limit adjustments.
Fix: exclude the key being updated (by matching token) from the
allocated totals before checking limits.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The terminal-state DB shortcut in retrieve_batch returned a LiteLLMBatch
with empty _hidden_params, causing the managed_files hook to skip encoding
output_file_id into a unified ID. This adds the same model_id extraction
from unified_batch_id that the non-terminal path already has.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The terminal-state DB shortcut in retrieve_batch returned a LiteLLMBatch
with empty _hidden_params, causing the managed_files hook to skip encoding
output_file_id into a unified ID. This adds the same model_id extraction
from unified_batch_id that the non-terminal path already has.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The previous fix (124b44ec) only updated 3 tests but missed 10 more
that still patched the old `ui_sso.httpx.AsyncClient` path. Also
updated credential assertions to check Authorization header instead
of httpx.BasicAuth kwargs, matching the production code change.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The test was creating a real AsyncHTTPHandler instance and patching its
post method, but the internal code creates its own handler, bypassing
the mock. This caused real API calls to Vertex AI, resulting in 401
auth errors in CI. Switched to patching AsyncHTTPHandler at the class
level, matching the pattern used by the passing GPT-OSS test.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The /key/update endpoint was missing support for organization_id, which
was already available on /key/generate. This adds the field to
UpdateKeyRequest and validates org key limits during updates.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The test_chat_completion_low_budget test was flaky because async spend
tracking couldn't reliably catch up within 50 calls with 0.5s sleeps.
Increased to 200 calls with 0.1s sleeps (same total time budget) to
give more opportunities for budget enforcement to trigger.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The recent commit 2a997993d4 replaced httpx.AsyncClient() with
get_async_httpx_client() in ui_sso.py, but the PKCE tests still
patched the old httpx.AsyncClient path. Updated all 10 affected
tests to mock get_async_httpx_client and removed unnecessary
context manager setup since AsyncHTTPHandler is returned directly.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Use dependency_overrides for user_api_key_auth instead of relying on
uninitialized proxy globals. The auth dependency was crashing with 500
(instead of 401) and returning MagicMock user_id/team_id values that
broke json.dumps in _encode_realtime_token_payload.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The existing AttributeError detection in proxy error handling only
checked one level deep in the exception chain (__cause__, __context__,
original_exception). In practice, the AttributeError from malformed
messages gets wrapped in multiple layers (AttributeError ->
OpenAIException -> APIConnectionError), so the check never found it.
Extracted the check into _has_attribute_error_in_chain() which walks
the full exception chain recursively (depth-capped at 10 to prevent
infinite loops from circular references).
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Skip test_apply_patch_tool_call_converted_to_chat_completion_tool_call
when openai.types.responses.response_apply_patch_tool_call is unavailable
(CI uses openai==1.100.1 which doesn't have this module)
- Skip MCP M2M tests (test_m2m_credentials_forwarded_to_server_model,
test_m2m_drops_incoming_oauth2_headers) that fail because PR #23187
changed has_client_credentials to require explicit oauth2_flow opt-in
but _execute_with_mcp_client was not updated to pass it through
- Revert source code change to rest_endpoints.py that auto-inferred
oauth2_flow (regression risk: this changes MCP OAuth behavior)
Co-authored-by: yuneng-jiang <yuneng-jiang@users.noreply.github.com>
- Vertex AI batch cost tests: replace removed gemini-1.5-flash-001 model
with gemini-2.0-flash-001 in pricing lookups
- MCP test_executes_tool_when_allowed: add server_id and auth_type attrs
to StubServer to match new _resolve_allowed_mcp_servers_with_ip_filter
- MCP M2M tests: infer oauth2_flow='client_credentials' in
_execute_with_mcp_client when client_id/client_secret/token_url present
(NewMCPServerRequest lacks oauth2_flow field)
- Team list test: update mock find_many to filter by team_id per the
current per-team query pattern in list_team
- Azure DALL-E 3 health check: skip test due to 410 ModelDeprecated
Co-authored-by: yuneng-jiang <yuneng-jiang@users.noreply.github.com>
Apply the same fix: move litellm.public_model_groups_links assignment
to after get_config()/save_config() so it is not overwritten by the
stale DB value read.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
get_config() internally calls _update_config_from_db which overwrites
litellm.public_model_groups with the stale DB value. Moving the
in-memory assignment to after get_config()/save_config() ensures the
new value persists.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
1. Add missing __init__.py files in tests/test_litellm/llms/gemini/ and
subdirectories (realtime/, image_edit/) to fix ModuleNotFoundError
with pytest-xdist parallel workers.
2. Update test_transform_request_uses_dynamic_max_tokens to use
claude-3-7-sonnet-20250219 (max_output_tokens=64000) since
claude-3-5-sonnet-20241022 was removed from model_prices JSON
during deprecated model cleanup. The test assertion was outdated.
3. Update context caching TTL tests to use gemini-2.5-pro instead of
gemini-1.5-pro. The old model was removed from model_prices JSON,
causing supports_system_messages to return False, which prevented
system_instruction from appearing in the transformation output.
Co-authored-by: yuneng-jiang <yuneng-jiang@users.noreply.github.com>
Update transform_image_generation_response test calls to pass required
explicit params (request_data, optional_params, litellm_params, encoding)
that replaced **kwargs in the method signature.
Co-authored-by: yuneng-jiang <yuneng-jiang@users.noreply.github.com>
The empty-line filter in __next__/__anext__ called .strip() without
checking the type first. When the Responses API yields Pydantic
BaseModel events (e.g. ResponseCreatedEvent), this raises
AttributeError. Add an isinstance(str_line, str) guard so non-string
objects pass through to _handle_string_chunk as intended.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Lint fixes (check_code_and_doc_quality job):
- Remove unused variable reasoning_effort in gpt_5_transformation.py (F841)
- Remove unused timezone imports in mcp_server rest_endpoints.py and server.py (F401)
- Remove unused ProxyBaseLLMRequestProcessing import in realtime endpoints.py (F401)
- Add BaseRealtimeHTTPConfig to TYPE_CHECKING block in utils.py (F821)
- Add PLR0915 per-file-ignore for mcp_server/rest_endpoints.py in ruff.toml
Test fixes (litellm_mapped_tests_llms job):
- Gemini video cost tests: pass explicit model_info to video_generation_cost()
instead of relying on gemini/veo-3.0-generate-preview being in model_prices JSON
- Anthropic max_tokens tests: mock get_max_tokens() to return expected values
instead of depending on claude-3-5-sonnet-20241022 being in model_prices JSON
- Vertex AI pydantic obj test: update from removed gemini-1.5-pro to gemini-2.5-flash,
update expected request body to use response_json_schema format
- Vertex AI/Bedrock file_content integration tests: update mocks to target
base_llm_http_handler.retrieve_file_content (the new code path via
ProviderConfigManager) instead of the old vertex_ai_files_instance/
bedrock_files_instance paths
Co-authored-by: yuneng-jiang <yuneng-jiang@users.noreply.github.com>
* fix(sso): add direct PKCE token exchange and Redis cache wiring for multi-instance SSO
When PKCE is enabled, bypass fastapi-sso and perform direct token exchange so
code_verifier is correctly included. Store PKCE verifiers as dict in cache
for proper JSON serialization in Redis. Wire user_api_key_cache to Redis when
available so PKCE verifiers are shared across ECS tasks/pods.
Also adds clearer error messages when PKCE is required but not configured.
* refactor(sso): extract PKCE token exchange into SSOAuthenticationHandler methods
- Move import httpx/jwt to module level (top of file, not inside function)
- Extract inline PKCE token exchange + userinfo logic into two static methods:
_pkce_token_exchange() and _get_pkce_userinfo()
- get_generic_sso_response PKCE path is now a single method call
- Fix double-logging in except block for non-PKCE errors
- Use %-style log formatting (no f-strings in log calls)
* fix: address greptile review feedback
- Fix access_token missing in PKCE path: read from combined_response directly
instead of generic_sso.access_token (which is only set by verify_and_process)
- Fix PKCE error hint firing when PKCE is already enabled: only show
'set GENERIC_CLIENT_USE_PKCE=true' advice when code_verifier was absent
- Fix unguarded KeyError on access_token: check for error field in HTTP 200
responses before accessing token_response['access_token']
- Fix silent empty userinfo: raise ProxyException when both userinfo endpoint
and id_token fallback produce no user data
- Fix backward-incompatible Redis wiring: only attach Redis to user_api_key_cache
when GENERIC_CLIENT_USE_PKCE=true, preserving existing in-memory behaviour
* fix: address second round of greptile review feedback
- Fix PKCE error hint: check env var directly (not code_verifier presence) to
distinguish 'PKCE not configured' from 'PKCE enabled but cache miss'
- Fix misleading Redis TTL comment in proxy_server.py
* fix: address third round of greptile review feedback
- Fix CRITICAL log firing on every non-PKCE callback: only log when PKCE is enabled
- Remove unused pkce_env_value intermediate variable
- Prefer reusing redis_usage_cache over creating separate RedisCache instance
(avoids losing advanced connection options like SSL, timeouts, db)
* fix: address fourth round of greptile review feedback
- Strip OAuth token credentials from response_convertor input to prevent
access_token/id_token appearing in restricted-group error messages
- Reuse single httpx.AsyncClient for both token exchange and userinfo requests
to avoid a second TCP/TLS handshake per SSO callback
- Revert Redis wiring to user_api_key_cache: PKCE code already uses
redis_usage_cache directly; wiring would route all API-key lookups through
Redis unnecessarily. Add startup warning instead when PKCE+Redis mismatch.
- Move _OAUTH_TOKEN_FIELDS to module level
* fix remaining PKCE test assertion for dict-format verifier storage
* sanitize PKCE cache log to not expose verifier content
* address greptile review feedback (greploop iteration 3)
* address greptile review feedback (greploop iteration 4)
* address greptile review feedback (greploop iteration 5)
* simplify _get_pkce_userinfo: remove shared-client complexity, use async with directly
* address greptile review feedback (greploop iteration 6)
* address greptile review feedback (greploop iteration 7)
* address greptile review feedback (greploop iteration 8)
* address greptile review feedback (greploop iteration 9)
* address greptile review feedback (greploop iteration 10)
* address greptile review feedback (greploop iteration 11)
* address greptile review feedback (greploop iteration 12)
* fix misleading comment on user_api_key_cache TTL line
* address greptile review feedback (greploop iteration 13)
* address greptile review feedback (greploop iteration 14)
* address greptile review feedback (greploop iteration 15)
* address greptile review feedback (greploop iteration 16)
* address greptile review feedback (greploop iteration 17)
* address greptile review feedback (greploop iteration 18)
* address greptile review feedback (greploop iteration 19)
* address greptile review feedback (greploop iteration 20)
* address greptile review feedback (greploop iteration 21)
* address greptile review feedback (greploop iteration 22)
* address greptile review feedback (greploop iteration 23)
* address greptile review feedback (greploop iteration 24)
* address greptile review feedback (greploop iteration 25)
* address greptile review feedback (greploop iteration 26)
* address greptile review feedback (greploop iteration 27)
* address greptile review feedback (greploop iteration 28)
* address greptile review feedback (greploop iteration 29)
* address greptile review feedback (greploop iteration 30)
* address greptile review feedback (greploop iteration 31)
* address greptile review feedback (greploop iteration 32)
* address greptile review feedback (greploop iteration 33)
* address greptile review feedback (greploop iteration 34)
* address greptile review feedback (greploop iteration 35)
* address greptile review feedback (greploop iteration 37)
- read GENERIC_CLIENT_USE_PKCE env var once in prepare_token_exchange_parameters
- include actual decode error in jwt.decode failure exception message
- add GENERIC_CLIENT_USE_PKCE=true to no-state regression test
* defer PKCE verifier deletion until after all downstream processing
Move _delete_pkce_verifier to after response_convertor and
process_sso_jwt_access_token complete. If JWT processing raises,
the verifier stays in cache so the user can retry without restarting
the full OAuth flow.
* address greptile review feedback (greploop iteration 38)
- fix strict-mode cache miss error message to differentiate
cross-instance routing failures (Redis configured) from single-instance
issues (TTL expiry, pod restart) when only in-memory cache is available
- add comment above _get_pkce_userinfo call explaining that bearer
credentials are always sourced from token_response in the merge step
* fix null JSON response body in _pkce_token_exchange
- Guard against HTTP 200 with body null: response.json() returns None
for JSON null, and calling .get() on None raises AttributeError.
Now raises a clean ProxyException with a clear error message.
- Fix misleading userinfo warning: was always saying "empty dict" but
also fires for JSON null responses; updated to say "empty or null".
- Add HTTP status code assertion to cache miss test.
* address greptile review feedback (greploop iteration 39)
- fix credential leakage: directly assign received_response from
combined_response instead of relying on nonlocal mutation; Pyright
was flagging the old guard as unreachable, meaning credential stripping
might not execute — now it always runs unconditionally
- add test for legacy plain-string cache format backward compat branch
- add test for HTTP 200 with no error field and no access_token (else branch)
- add test for HTTP 200 with JSON null body (new AttributeError guard)
* fix _OAUTH_TOKEN_FIELDS merge loop to preserve userinfo values on absent fields
When the token endpoint omits a bearer-credential field entirely (field
absent from token_response), the previous code deleted it from merged even
if userinfo provided a valid value. Now:
- non-null in token_response → restore authoritative token endpoint value
- explicit null in token_response → remove key from merged (clean absence)
- field absent from token_response → leave userinfo value unchanged
* use HTTP 401 for PKCE missing config errors
GENERIC_CLIENT_ID and GENERIC_TOKEN_ENDPOINT missing when PKCE is
enabled are auth-flow failures, not server errors. Use 401 instead
of 500 to avoid triggering false-positive server error alerts in
monitoring systems.
* address greptile review feedback (greploop iteration 40)
- fix duplicate error logging: demote first format-error log to DEBUG
so the detailed ERROR in strict-mode branch is not duplicated
- add HTTP status code assertions to all PKCE ProxyException tests
for better regression protection against accidental code changes
* add credential absence assertions to test_pkce_token_exchange_basic_auth
Verify that client_id and client_secret are NOT double-sent in the POST
body when Basic Auth is used (include_client_id=False with client_secret).
Catches regressions where credentials leak into both Auth header and body.
* address greptile review feedback (greploop iteration 41)
- add Bearer token header assertion to test_pkce_token_exchange_credentials_in_body
- add cache query assertions to both non-strict mode tests to confirm
the cache was accessed before the warning path triggers
* address greptile review feedback (greploop iteration 42)
- assert null id_token is absent from merged result in basic auth test
- add test for HTTP 200 empty/null userinfo body with no id_token fallback
* use caplog to verify warning logs in non-strict cache miss tests
The two non-strict mode tests now use pytest's caplog fixture to assert
that a warning is actually emitted, not just that the code continues
without raising. This catches regressions where the warning silently
disappears.
* remove dead-code response=None guard in _pkce_token_exchange
* clean up stale pkce verifier cache entries in non-strict mode
* fix test: configure async_delete_cache as AsyncMock and assert cleanup called
* add sentinel guard so pkce-no-redis warning only fires once across hot-reloads
* fix misleading comments: code_verifier init and bearer-credential merge docs
* add best-effort cleanup in strict-mode for corrupt/empty cache entries
* add redirect_uri assertion, userinfo body in non-200 log, sentinel comment
OpenAI rejects any reasoning_effort (even 'none') with tools in
/v1/chat/completions for gpt-5.4. Update the guard to drop reasoning_effort
regardless of value. Add docs explaining the auto-drop behavior.
