Add DEFAULT_MANAGEMENT_OBJECT_IN_MEMORY_CACHE_TTL to both async_set_cache
calls in sync_user_role_and_teams for consistency with all other user cache
writes. Add 3 tests covering cache invalidation on role change, team change,
and no-op when nothing changes.
sync_user_role_and_teams updates the DB when a user's JWT role changes,
but the in-memory cache retained the stale role until TTL expiry. This
caused subsequent requests to see the old role for up to 60 seconds.
Fix: accept user_api_key_cache parameter and re-cache the updated user
object after both role and team membership DB writes.
- Migrate budget CRUD from manual state to React Query hooks (useBudgets, useCreateBudget, useUpdateBudget, useDeleteBudget)
- Fix crash when budget list contains null entries by filtering in query hook
- Fix max_budget type from string to number to match DB schema (double precision)
- Disable budget_id field in edit modal to prevent accidental changes
- Use budget_id as React key instead of array index
- Update tests to mock hooks instead of networking functions
/user/bulk_update was missing from the management_routes list in _types.py,
causing it to fall through to a 403 in non_proxy_admin_allowed_routes_check
even for proxy admin users. Also added it to the PROXY_ADMIN_VIEW_ONLY
blocked write operations list in route_checks.py to prevent view-only
admins from using it.
Add None-token test cases to both proxy_unit_tests and test_litellm
to cover the guard added in the previous commit. Also add -> bool
return type annotation to is_jwt().
When JWT auth is enabled and a request arrives without an Authorization
header (e.g. health checks, monitoring), api_key is None due to
APIKeyHeader(auto_error=False). The is_jwt() call crashes with
AttributeError: 'NoneType' object has no attribute 'split'.
Return False for None tokens since they are not JWTs.
The schema sync adopted the proxy version which includes source_url,
approval_status, and other BYOM fields. These were previously dropped
in migration 20260311180521 due to schema drift. This migration
restores them to match the now-unified schema.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Add persist-credentials: false to check-schema-sync (read-only, no push needed).
Explicitly set persist-credentials: true on sync-schema (required for git push).
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Sync all 3 schema.prisma copies and add GHA workflows to keep them in sync automatically.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
During SSO login, bearer tokens are stripped from the OAuth response
before role mapping runs. Custom role claims encoded inside the JWT
access token are lost, so map_jwt_role_to_litellm_role() returns None
and the user falls back to internal_user_viewer.
process_sso_jwt_access_token() now returns the decoded JWT payload, and
a new _sync_user_role_from_jwt_role_map() receives it so
jwt_litellm_role_map works correctly during SSO login.
Budget checks on API keys, teams, and team members were not enforced in
multi-pod deployments because user_api_key_cache is intentionally
in-memory-only. Each pod tracked spend independently, so with N pods
the effective budget was N × max_budget.
Introduces a separate spend_counter_cache (DualCache wired to
redis_usage_cache) with atomic increment/read helpers:
- increment_spend_counters(): awaited in cost callback (not create_task)
to update both in-memory and Redis before the next auth check
- get_current_spend(): reads Redis first (cross-pod authoritative),
falls back to in-memory, then to cached object .spend from DB
Budget check functions (_virtual_key_max_budget_check,
_team_max_budget_check, _check_team_member_budget) now read spend via
get_current_spend() instead of cached object .spend fields.
When Redis is not configured, falls back to in-memory-only counters
(same as current single-instance behavior).
Fixes#23714
Switch query suite from security-extended to security-and-quality to
match the default GitHub Advanced Security setup. Run scheduled scans
daily instead of weekly. Remove paths-ignore for _experimental/out so
build artifacts are also scanned.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
prisma generate internally runs `npm install prisma@5.4.2` against the
npm registry at runtime. Without a bundled Node.js, this causes
ECONNRESET failures on flaky GitHub Actions network and leaves the
npm transitive dependency tree unpinned.
Pre-install nodejs-wheel-binaries==24.13.1 (matching the Dockerfiles)
so prisma uses the bundled Node/npm instead of fetching from the
registry.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
These models include the full 1M token context at standard pricing with no 2x surcharge above 200k tokens.
Co-Authored-By: Claude Sonnet 4.6 (1M context) <noreply@anthropic.com>
The isort hook picked up a stale rename from the working directory.
Both router.py and proxy_server.py need litellm._uuid, not _litellm_uuid.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- model_id is now added after _clean_endpoint_data() so it survives
health_check_details: False (MINIMAL_DISPLAY_PARAMS filtering)
- Health state write failures logged at warning instead of debug
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Background health checks now feed deployment health state into the
router candidate-filtering pipeline. Unhealthy deployments are excluded
proactively instead of waiting for request failures to trigger cooldown.
Gated by `enable_health_check_routing: true` in general_settings.
Off by default — zero behavior change for existing users.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
