Skip to content

hotfix: revert litellm guardrails (#540 + #544) — logging_only mode crash-looped litellm - #545

Merged
samxu01 merged 2 commits into
mainfrom
hotfix/revert-guardrails-crash
Jul 1, 2026
Merged

hotfix: revert litellm guardrails (#540 + #544) — logging_only mode crash-looped litellm#545
samxu01 merged 2 commits into
mainfrom
hotfix/revert-guardrails-crash

Conversation

@samxu01

@samxu01 samxu01 commented Jul 1, 2026

Copy link
Copy Markdown
Contributor

Incident: the guardrails config crash-looped litellm on the deploy that rolled the pod. Root cause: mode: "logging_only" (from #540) is NOT a valid guardrail event hook in LiteLLM v1.88 — only pre_call/during_call/post_call are supported. It sat in the configmap since #540 but only loaded when #544 rolled the litellm pod → ValueError: Event hook GuardrailEventHooks.logging_only is not in the supported event hooksApplication startup failed → CrashLoopBackOff (fleet LLM proxy down).

Recovery: stripped the guardrails block from the live configmap + restarted litellm (now 3/3, 0 restarts). This PR reverts #540 + #544 so main/the next deploy is safe (matches the recovered live state).

Re-do plan: re-introduce guardrails with VALID modes (drop the unsupported logging_only monitor; enforce-moderation + injection scoped to platform features via during_call), and VALIDATE the config boots in a throwaway litellm before deploying to the fleet — the lesson is an unvalidated guardrail config can take the whole proxy down. Tracked in #541 / #12.

🤖 Generated with Claude Code

samxu01 added 2 commits July 1, 2026 00:06
…n scoped to platform features (#544)"

This reverts commit 33059c1.
… for anti-pattern detection (#540)"

This reverts commit 5a646c0.
@samxu01
samxu01 merged commit 1aa7300 into main Jul 1, 2026
12 checks passed
lilyshen0722 added a commit that referenced this pull request Jul 1, 2026
…rash-looped litellm (#545)

* Revert "feat(litellm): prompt-injection guardrail + enforce-moderation scoped to platform features (#544)"

This reverts commit 33059c1.

* Revert "feat(litellm): add OpenAI moderation guardrail (monitor-only) for anti-pattern detection (#540)"

This reverts commit 5a646c0.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant