Question

Mid-morning, the orders API latency p99 jumps from 80ms to 8s and error rate climbs as request timeouts pile up. The DB (MySQL/InnoDB) CPU is only at 30% — it's not saturated. Your APM shows hundreds of `UPDATE orders SET status=...` statements stuck in "waiting for lock." `SHOW ENGINE INNODB STATUS` and `sys.innodb_lock_waits` reveal they're all blocked behind a single transaction that's been open for 22 minutes and holds row locks on a swath of the orders table. That transaction came from an analytics/reporting connection. Recent context: a data analyst started running an ad-hoc "reconcile yesterday's orders" script before standup. Triage and mitigate.