Question

Postgres 14. A schema migration (`ALTER TABLE accounts ADD COLUMN ...`) was deployed at 16:30 and the deploy 'hung'. Within 2 minutes, p99 on the entire `accounts`-touching API spiked and `pg_stat_activity` filled with sessions in `wait_event = relation` / `Lock`. The ALTER itself is shown as `active` but waiting. There's also a long-running analytics `SELECT` on `accounts` that started at 16:25 and a steady stream of normal app queries. Explain the lock cascade, how you triage with the lock views, the safe mitigation, and how to ship such migrations without an outage.