v1.84.8 - Database Resilience & Grace-Period Key Rotation

Deploy this version

Docker

docker run \
-e STORE_MODEL_IN_DB=True \
-p 4000:4000 \
docker.litellm.ai/berriai/litellm:1.84.8

Pip

pip install litellm==1.84.8

v1.84.8 is a patch release on top of v1.84.7. It backports a database-resilience set (Prisma reconnection, prepared-statement and timeout controls, 5xx on DB infra errors during auth) together with the complete grace-period key rotation fix, plus routing and streaming corrections.

What's Changed

• fix(anthropic): correct streaming reasoning token usage - PR #27319
• fix(proxy): grace-period key rotation deprecated-key lookup - PR #27756
• fix(router): use forwarded model_id for native Azure container IDs - PR #27921
• fix(proxy): recover from cached-plan errors by reconnecting the Prisma client - PR #29983
• fix(proxy): expose Prisma idle/connect timeout and extra DB URL params - PR #28395
• feat(proxy): add option to disable server-side prepared statements for DB lookups - PR #29984
• fix(proxy): return 5xx on DB infra errors during auth - PR #29986
• fix(passthrough): resolve costing model when body model is unknown - PR #30160
• fix(proxy): return deprecated-key lookup result directly in get_data combined view - PR #30327

Full Changelog

https://github.com/BerriAI/litellm/compare/v1.84.7...v1.84.8