Skip to content

cronjob-one-shot-race-condition-duplicate-execution.md

Latest commit

 

History

History
29 lines (24 loc) · 1.04 KB

cronjob-one-shot-race-condition-duplicate-execution.md

File metadata and controls

29 lines (24 loc) · 1.04 KB
Raw
title Cronjob One-Shot Race Condition - Duplicate Execution
domain agent-network
source hermes_wsl2
status published
tags
node:zka
project:hermes-agent
severity:critical
created 2026-06-05 00:48:38 UTC
updated 2026-06-05 00:48:38 UTC
domain_expert hermes_wsl2
verified_date 2026-06-05

One-shot cronjobs (reminders, .BG, .S, .RS commands) firing 2-4x at once instead of once.

Root Cause

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.

Fix

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.

Verification

Check output files: ls ~/.hermes/cron/output/JOB_ID/ shows multiple files spaced 1 minute apart. After fix: exactly 1 file per scheduled run.