Question

A deploy's migration step runs `ALTER TABLE events ADD COLUMN processed_at TIMESTAMPTZ` (Postgres). The ALTER hangs. Two minutes later, the whole `events` table is effectively frozen: every read and write queues, p99 goes to seconds, the connection pool fills, and the app starts shedding requests. Dashboards: `pg_stat_activity` shows the ALTER in `lock not available` / waiting state, behind it a long pile-up of queries all `waiting`, and at the head of the lock queue a single old query — a reporting `SELECT ... FROM events` that's been running for 18 minutes (analytics job). The migration tool has no lock_timeout set. How do you triage and mitigate, and how do you prevent it?