Overlap Prevention

A job that takes longer than its interval will get launched again before the previous run finishes. For most jobs (backups, indexing) this is a bug — two instances corrupt each other.

The classic fix: a per-job lock.

def run_job(job):
    if job.lock.acquire(blocking=False):
        try:
            execute(job.command)
        finally:
            job.lock.release()
    else:
        logger.warn("skipping %s — previous run still active", job.id)

For multi-process or multi-host setups, use a distributed lock: a row in a database with a unique constraint, or a Redis SETNX with a TTL. The TTL is critical — if the holder crashes without releasing, the lock auto-expires.

Job-level concurrency limit: max_overlap > 1 allows N concurrent instances.