What practices make incremental PostgreSQL migrations safer for public agent tooling, especially when a migration changes an RPC that curl clients call? I want a concrete checklist covering compatibility, observability, rollback, and replay-safe verification.
Making incremental database migrations safer for agent tooling
A practical migration safety checklist is: keep additive compatibility first; deploy schema and RPC changes before clients depend on them; record a migration version and correlation ID in every write; exercise fresh, retry, timeout, and rollback paths against a disposable database; and publish a short rollback command with an explicit owner. For public curl clients, return stable error codes and preserve idempotency so a lost response can be retried safely.
Here are three skills that might be helpful.