fix(eval): cluster-child per-pass telemetry + full stderr crash capture#74
Merged
Conversation
The canonical A-1 cluster child died at ~56 min under BOTH a 12GB and a 20GB heap — cap-independent death time — and the orchestrator kept only 200 chars of err.message, discarding the V8 GC dump that says which heap space exhausted and every trace of how far the loop got. Two hour-long runs were undiagnosable. - the child logs every pass to stderr AND _cluster-progress.log next to the report (pass time, maxDelta, non-finite cell, heapUsed/rss, sample size) — live-tailable, survives tmp cleanup - on crash the orchestrator keeps a 2500-char stderr tail (classifying V8 heap-dump signatures as OOM) and prints the last lines; on success it surfaces the per-pass telemetry in the orchestrator log - a 12-min/pass real-model cluster is never a black box again Diagnosis itself is the next step (probe converges in 2 passes under 6GB on a strided sample; the engine-mirrored full surface dies — trajectory unknown until an instrumented run lands). Co-Authored-By: Claude Fable 5 <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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Two canonical A-1 cluster runs died at ~56 min under 12GB AND 20GB heaps, silent, with the V8 GC dump truncated to 200 chars. The child now logs pass/delta/heap per pass (stderr +
_cluster-progress.lognext to the report, live-tailable) and the orchestrator keeps a real 2500-char stderr tail + classifies V8 heap-dump signatures as OOM. Eval suites green (10+6+5+6). Diagnosis run is next.🤖 Generated with Claude Code