Repository navigation
feat(preheat): retry a failed manifest with backoff - #130
Merged
Merged
Conversation
Preheat jobs run for hours through shared upstream exits, and any one file error failed the whole job. On 2026-10-13 at bj-11 a ~13s burst of dropped connections (unexpected EOF on every stream) failed CASIA-LM/ChineseWebText2.0 at 87% of 1.6 TiB, and a single paths-info timeout failed uonlp/CulturaX; both sat failed for hours until resumed by hand, and both only needed another pass. Rerun the manifest after 30s, 1m, 2m, 5m, then 10m (8 retries, ~48 min). A rerun skips completed files and resumes partial ones from their cached offset. No retry when the job was stopped, canceled or shut down, or when the upstream answered 401/403/404. The progress counter restarts with each pass, and the speed sampler no longer underflows when it does. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
cooocy
self-requested a review
October 13, 2026 15:53
cooocy
approved these changes
Oct 13, 2026
TestModelScopeFailedTaskRetriesPinnedCache waits for the job to fail and then resumes it by hand; with auto-retry the job sat in its first 30s backoff instead. Export PreheatRetryBackoff (empty disables retries) and clear it in that test. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
This was referenced Oct 13, 2026
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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
A preheat can run for hours, and any single file error failed the whole job, requiring a manual resume even for transient upstream errors.
Fix
Re-run a failed manifest with backoff 30s, 1m, 2m, 5m, then 10m (8 attempts, ~48 min). Completed files are skipped and partial files resume from the cached offset. No retry when the job context is done (stop/cancel/shutdown) or on 401/403/404. Progress restarts per attempt and the speed sampler guards against counter rollback.
Tests
TestPreheatRetriesTransientFailures,TestPreheatRetryGivesUp.go test -race ./internal/service/task/passes.🤖 Generated with Claude Code