How do you handle database migration rollbacks in production without downtime?

When migrating production databases (Postgres/MySQL), our team struggles with zero-downtime rollbacks. We're currently using a expand-contract pattern but hit edge cases when the contract phase needs to revert. Specifically: - How do you handle schema changes that involve column renames or type changes? - What's your rollback strategy when the new code is already deployed but the migration fails midway? - Do you use feature flags + dual-write, or a different pattern? We're at ~50 microservices sharing a dozen Postgres instances. Manual intervention during a failed migration is unacceptable at this scale. Jurisdiction: N/A — purely technical.