prompt: file issue for xla-cpu-eigen-pool-deadlock (#1530) - #349
Merged
Conversation
The research follow-up to the shipped jax-compile-stall epic. #1528 delivered a workaround: both test workspaces now run every JAX script with single-threaded Eigen, ~15% slower on the heaviest paths, behind a flag whose removal silently brings back seven quarantines. The prompt carries what is already settled so the next session does not re-run it -- the hang locus (materialising a vmap result, not compiling it) and three refuted hypotheses (compilation cache, jax/jaxlib version, the 3.12-vs-3.13 split) -- plus the method note that the hang rate wanders, so any claim needs ABAB against a re-measured control rather than before/after. The question worth the most: whether XLA's pool is merely mis-sized against the runner's cgroup quota rather than genuinely deadlocked. If so the fix is sizing it and the 15% comes back. Branch restarted from origin/main, since the previous PR on this branch (PyAutoMind#348) is merged and merged history is not stacked on. Co-Authored-By: Claude <noreply@anthropic.com>
Jammy2211
pushed a commit
that referenced
this pull request
Aug 27, 2026
The branch was behind main by PyAutoMind#348/#349/#350. Merged, never rebased. All three conflicts were in generated files — complete/index.md, dashboard.md, dashboard.html — and were resolved by regenerating from the merged sources (lifecycle.py index --apply, pyauto-brain intake --apply dashboard), never by hand-editing. No source-of-truth file conflicted. All four PR gates re-run green after the merge: lifecycle check, lifecycle index --check, registry_toc --check, intake dashboard --check, plus the spawn privacy suite (240 passed). Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01KmWbYJw4cJJnt4eK6NoZjL
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
2 participants
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.
Issues the research follow-up to the shipped
jax-compile-stallepic: PyAutoFit#1530.Registry write only — prompt
draft/research/ci/→active/,active.mdentry, dashboard regenerated. No code.Why it exists
PyAutoFit#1528 delivered a workaround. Both test workspaces now run every JAX script with single-threaded Eigen — ~15% slower on the heaviest paths — behind a flag whose removal silently brings back seven quarantines. We know where the hang is and what avoids it, not why the pool wedges.
What the prompt carries forward
The things that cost a day to learn, so the next session doesn't pay again:
jax.block_until_ready/try_to_blockandjax.Array._value, compile finished ~12–18s earlier).0.11.1stack on the day it hung 3/3 on both legs), the 3.12-vs-3.13 split (passed 2/2 then hung 2/2 on one commit).The question worth the most is whether XLA's pool is merely mis-sized against the runner's cgroup quota rather than genuinely deadlocked. If so the fix is sizing it, and the 15% comes back.
Branch note
Restarted from
origin/mainafter PyAutoMind#348 merged — merged history is not stacked on.lifecycle.py checkOK ·intake dashboard --checkcurrent.Generated by Claude Code