mirror of
https://github.com/BerriAI/litellm.git
synced 2026-10-05 02:41:56 +00:00
Three of the review findings, each restoring a pre-PR behavior the v2 work changed: Provider credentials are merged in memory again on a partial PATCH. _sync_in_memory_credential now always _merge_credential_info, so a PATCH of only description keeps custom_llm_provider in litellm.credential_list rather than replacing it (which 401'd the credential until the reload tick). The phoenix, arize and agentops presets append their global exporter unconditionally again, so a zero-credential deployment (local Phoenix on localhost:6006, no-auth self-hosted arize) keeps exporting exactly as before instead of dropping the exporter when no env credentials are set. LITELLM_OTEL_V2 is the sole activation gate again. Registering an admin-owned destination no longer flips a v1 deployment onto v2 with the flag off; the resolver no-ops and _maybe_construct_otel_v2 returns None when the flag is off, so the flag-off + destination orphaned-trace configuration can't arise and an existing v1 deployment is unaffected by merely registering a credential. |
||
|---|---|---|
| .. | ||
| test_endpoints.py | ||