Skip to content

fix(streaming): prevent duplicate output when resuming after crashes - #558

Open
rikkurea-boop wants to merge 1 commit into
OpenDCAI:mainfrom
rikkurea-boop:publish/stream-crash-safe-resume-20261014
Open

rikkurea-boop wants to merge 1 commit into
OpenDCAI:mainfrom
rikkurea-boop:publish/stream-crash-safe-resume-20261014

Conversation

@rikkurea-boop

Copy link
Copy Markdown
Contributor

Summary

Streaming pipelines may duplicate records when interrupted after writing output but before saving progress.

This PR adds an optional safer resume mode for compatible local CSV/JSONL pipelines.

Changes

  • Restore output to the last saved progress before retrying.
  • Validate saved data and reject unsafe recovery.
  • Keep existing streaming behavior unchanged by default.
  • Add regression tests for crash recovery and resource cleanup.

Testing

  • 56 focused regression tests passed.
  • All 8 Ubuntu/Windows validation jobs passed, including integration with current storage fixes.

This mode requires compatible operators and a single local writer. External side effects and power-loss durability are not covered.

Add opt-in crash-consistent streaming resume for compatible local CSV/JSONL
operators, with durable progress tracking, recovery, and regression tests.
Default streaming behavior remains unchanged.

This branch has not been deployed

No deployments
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