fix: watchdog recycles the enclave when Gemma is wedged (v0.7.7)#5
Merged
Conversation
…recycles (v0.7.7) v0.7.6 relied on Tinfoil recycling the container when /healthz went 503, but `restart: always` only fires on container EXIT, not on an unhealthy healthcheck — so a wedged vLLM sat down ~5 days while /filter stayed green and gemma_down_seconds climbed unbounded. Add an in-container watchdog that exits PID 1 after a sustained /healthz 503 (3x30s past the 900s gemma-down policy), turning the existing /healthz verdict into the container exit restart:always honors. Bump image to 0.7.7. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Problem
gemma4-e4bwas down ~125h continuously (gemma_down_secondsclimbing monotonically), CVM never recycled.v0.7.6 supervises vLLM (good) and flips
/healthzto 503 once Gemma is down pastGEMMA_UNHEALTHY_AFTER(900s) — but it relied on Tinfoil recycling the container on that unhealthy healthcheck. Dockerrestart: alwaysonly acts on container EXIT, not on an unhealthy healthcheck. So the supervisor restarted vLLM in-container forever,/healthz503'd forever, and nothing recycled the CVM to clear the wedged GPU./filterstayed green the whole time, hiding it.Fix
Add an in-container watchdog that polls
/healthzand, on a sustained 503 (3×30s past the 900s policy), exits PID 1 — turning the existing 503 verdict into the container exitrestart: alwayshonors. Recycle clears wedged GPU state a process restart can't./filterstays up until the instant of recycle; transient vLLM crashes still heal in-container via the supervisor (no recycle).entrypoint.sh: watchdog loop + updated escalation-ladder docstinfoil-config.yml: bump image → 0.7.7 (digest pinned after build), note watchdog drives the recyclebash -nclean; probe one-liner validated (200 on live port, 0 on dead).Deploy
Build 0.7.7 → pin digest → tag v0.7.7 → deploy in Tinfoil dashboard (also clears the current outage). 🤖 Generated with Claude Code