v2.10.5 is a focused reliability release for distributed execution. It improves recovery from corrupted file-backed records after abrupt shutdowns and ensures cancellation requests received during startup are not lost.
Fixed
-
fix: recover corrupt distributed records (#2376) @yohamta0
- Hardened file-backed writes against corruption caused by process or host crashes.
- Stale corrupted lease and active-run records are now safely removed, while authoritative updates can replace corrupted records immediately.
- A single corrupted record no longer blocks queue capacity checks and leaves scheduled workflows stuck in the queued state.
-
fix: preserve startup cancellation (#2377) @yohamta0
- Cancellation and termination requests received before execution starts are now preserved, preventing steps from running after a stop has already been requested.
- Also improved artifact-reference detection for typed YAML values and opaque runtime data.
Contributors
Thanks to our contributors for this release:
| Contribution | Contributor |
|---|---|
| bug: scheduled workflows getting queued due to corrupt records in the file backend (#2375) | @pommetjehorlepiep (report) |