Data migrations without the weekend war room

The risk in migrations is rarely the SQL—it is the unknown unknowns in production data. Our playbook assumes duplicates, nulls, and one-off rows that never appeared in staging.

Incremental beats big bang

We prefer expanding schema, backfilling in batches, validating counts and checksums, then cutting reads and writes over in steps. Big-bang cutovers are a last resort with a very narrow window.

Reversible steps

Every phase should have a rollback or forward-fix path. If you cannot answer “what do we do at 2 a.m. if this step fails,” the plan is not ready.

Communicate with stakeholders

Data moves affect reporting, integrations, and finance. We build a comms checklist so nobody discovers schema changes through broken dashboards on Monday morning.