Warp Migrate

Problem

Adding a NOT NULL column to a 1.2 million row table locked writes for four minutes on the primary. The team’s answer was a Sunday 2am maintenance window, roughly monthly.

The existing zero-downtime tools were written for MySQL, or assumed you could install an extension the managed database would not let you install.

Approach

Warp creates a shadow table with the target schema, backfills it in bounded batches that yield between chunks, and keeps it current with triggers on the source.

Cutover is a single transaction that renames both tables, and it refuses to run while replication lag is above a threshold you set. If anything is wrong it does nothing, loudly.

Every run starts with a dry run that prints the exact statements. No migration tool earns trust by being clever about what it is about to do.

Result

The Sunday windows stopped. Migrations now run on a Tuesday afternoon while the site takes traffic, with peak write latency up six milliseconds during the backfill.

Around 900 stars and a steady trickle of issues from people running it against managed Postgres, which is exactly who it was for.