Skip to content

v0.10.38 — Fast task recovery after SIGKILL/power outage

Choose a tag to compare

@rafiki270 rafiki270 released this 27 Feb 14:20
· 387 commits to main since this release

Changes

  • fix: reset orphaned tasks immediately when stale runner detected — when a runner dies via SIGKILL (power outage, OOM, etc.), its in-progress task is now reset to pending within ~5 minutes instead of ~30 minutes. Sections unblock immediately on the next wakeup cycle instead of requiring manual intervention.

How it works

When wakeup Step 1 detects a dead runner (stale heartbeat > 5 min):

  1. Opens the project DB
  2. Closes the running invocation (marks as failed)
  3. Resets the in_progress task back to pending
  4. Deletes the task lock immediately (process is definitively dead — unlike the SIGTERM path, which leaves the lock to prevent double-execution)
  5. Then removes the runner row as before

Recovery timeline: ~5 min (was ~30 min).

Compare

https://github.com/williamcotton/steroids-cli/compare/v0.10.37...v0.10.38