Release analysis
Test Migrations Against Production-Shaped Data, Not Fixtures
Fixtures are clean by construction, while production data carries the accumulated exceptions of years, and migrations fail precisely on what fixtures omit.
Migration failures rarely involve the data everyone remembers creating. They involve the null from an old bug, the format from two schemas ago, the record that predates the rule.
Making rehearsal realistic
- Sample real distributions, including the ancient and malformed.
- Measure migration timing at production scale, not fixture scale.
- Rehearse the abort: interrupt the migration and verify recovery.
Use the changelog, blackboard schema, and benchmarks to let the rehearsal meet the same mess the rollout will.