lesson-cronjob-one-shot-race-condition-duplicat.json

{
  "task_id": "lesson-cronjob-one-shot-race-condition-duplicat",
  "title": "Cronjob One-Shot Race Condition - Duplicate Execution",
  "domain": "agent-network",
  "tags": [
    "node:ZKA",
    "project:Hermes-Agent",
    "severity:critical"
  ],
  "problem": "Scheduler tick() releases file lock BEFORE mark_job_run() sets last_run_at. For jobs taking >60s, next tick sees job still 'due' and fires again. Also fcntl.flock is per-file-descriptor, NOT per-process — multiple gateway threads bypass locking.",
  "solution": "Added threading.Lock (_tick_thread_lock) in scheduler.py. Removed cron-tick.sh from crontab — gateway internal 60s ticker is the only tick source now.",
  "source": "lessons/cronjob-one-shot-race-condition-duplicate-execution.md",
  "test_cmd": "python3 scripts/verify_task.py lesson-cronjob-one-shot-race-condition-duplicat"
}