Summary

#50414 (closed completed 2026-04-25) fixed the original bug where cron jobs interrupted by gateway restart were silently marked completed. The current behavior in 2026.5.2 (8b2a6e5) flipped to the opposite problem: interrupted jobs are correctly marked failed, but state.consecutiveErrors increments on every restart-induced interruption — making the metric useless for monitoring real agent health when the gateway gets restarted normally.

Evidence

Job state observed today after a sequence of routine gateway restarts:

"state": {
  "lastRunStatus": "error",
  "lastError": "cron: job interrupted by gateway restart",
  "consecutiveErrors": 4,
  "lastDeliveryError": "cron: job interrupted by gateway restart"
}

All four "errors" came from openclaw config set triggering hot-reloads during config tuning, not from agent failures. The job ran successfully in every case where the gateway didn't restart mid-run.

Expected behavior

gateway-restart interruptions should:

• Be distinguishable from agent-turn failures (e.g. lastRunStatus: "interrupted", separate lastInterruptionReason: "gateway-restart").
• Not increment consecutiveErrors, or increment a separate consecutiveInterruptions counter.

Suggested fix

In the cron runner, detect SIGUSR1/SIGTERM-driven shutdown and emit a distinct interruption status. Or expose a --reset-errors flag on openclaw cron edit for operators to clear the counter manually.

Environment

• OpenClaw 2026.5.2 (8b2a6e5)
• Linux x86_64
• Cron job: kind: every, everyMs: 600000, sessionTarget: isolated