The migration that locked prod at 2am

The deploy went out at 01:58. By 02:03 the on-call phone was lit up, the API was returning timeouts, and a migration that had passed every check in staging was holding every write in production behind