Skip to content

Fix resume skipping an interrupted Hessian fit - #449

Open
luciaquirke wants to merge 1 commit into
mainfrom
fix-hessian-step-skip
Open

Fix resume skipping an interrupted Hessian fit#449
luciaquirke wants to merge 1 commit into
mainfrom
fix-hessian-step-skip

Conversation

@luciaquirke

Copy link
Copy Markdown
Collaborator

The step-2 resume check tests _step_complete(hessian_path), but the fit runs with run_path = {hessian_path}/{method} and promotes {hessian_path}/{method}.part -> {hessian_path}/{method}. Creating the .part directory also creates the parent hessian/, so after any interruption the parent exists, step_state reports complete, and a resumed run skips the unfinished fit — step 3 then crashes on missing factors (or, with a stale partial method dir, silently applies an incomplete Hessian).

Hit in practice when a shared volume filled mid-fit: every resumed pipeline jumped to apply_hessian and died with apply_hessian child exited with code 1; the only recovery was deleting hessian/ by hand and refitting from scratch.

Fix: check the path the fit actually promotes, {hessian_path}/{method} — the same path step 3 reads.

🤖 Generated with Claude Code

https://claude.ai/code/session_0167XKnE2yoCE162n3jHvxC1

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