عرض عملي: مهمة ليلية تعمل مرة واحدة عبر ثلاث نسخ

خطوة بخطوة

1. Scale orders to three replicas; the @Scheduled reconciliation job runs three times.
2. Add ShedLock with a JDBC lock provider (verify library and version before recording) and annotate with @SchedulerLock(name = "recon", lockAtMostFor = "10m").
3. Run again: one replica executes, others skip.
4. Planned mistake: set lockAtMostFor shorter than the job. A second replica starts while the first still runs. Fix: size it above the worst-case runtime and fence the writes.
5. Add a lock-generation number as a fencing token on the state table.
6. Pause the lock holder with docker pause, let the lease expire, let another run, unpause, and show the stale writes are rejected.