Skip to content

Give patch capture and grading staging 600 s instead of 120 s - #40

Open
Edward Glöckner (EdwardGlockner) wants to merge 1 commit into
microsoft:mainfrom
EdwardGlockner:patch-capture-timeout
Open

Edward Glöckner (EdwardGlockner) wants to merge 1 commit into
microsoft:mainfrom
EdwardGlockner:patch-capture-timeout

Conversation

@EdwardGlockner

Copy link
Copy Markdown

get_patch() and the pre-verifier git add -A stage the whole worktree after the agent has finished but use the 120 s default command timeout. When that times out (large repo, busy node, many untracked files), the exception ends the attempt and the completed agent run is discarded and replayed. We hit this on SWE-bench Verified django__django-15916: a finished 21-minute run was lost to the 120 s timeout and re-run from scratch. This raises both to 600 s (module constant _GIT_STAGING_TIMEOUT) and asserts it in test_get_patch_resets_index_after_capture.

get_patch() and the pre-verifier `git add -A` run over the whole worktree
after the agent has finished, but use the 120 s default command timeout.
On a large repository or a busy node this can time out, which raises out
of the attempt and discards the completed agent run; the task is then
replayed from scratch. We hit this running SWE-bench Verified locally
(django__django-15916: a finished 21-minute run lost to the 120 s
timeout and re-run). Use a 600 s timeout for both.
@EdwardGlockner

Copy link
Copy Markdown
Author

@microsoft-github-policy-service agree

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