Do I need gh-ost/pt-online-schema-change if I’m on Postgres?

Not for most cases. Postgres has `CREATE INDEX CONCURRENTLY`, `ALTER TABLE ... VALIDATE CONSTRAINT`, and can add columns without rewrite if you avoid defaults. Use `pg_repack` for heavy rewrites (table bloat, clustering) without long locks.

How do I keep replicas from falling over during backfill?

Throttle batch size and sleep between chunks; keep transactions small. Monitor replication lag and pause the job if you exceed your gate (e.g., 2s). Consider running backfill on the primary only and let replicas catch up before cutover.

Can I collapse steps if I have a maintenance window?

Yes, but treat that as a separate runbook. If your business approves a window, you can do blocking DDL. Most teams can’t, so the expand/contract approach is safer and repeatable.

Use them for scaffolding, not for prod DDL. Generate migrations, then hand-edit for online semantics. For example, convert generated Postgres index creation to `CREATE INDEX CONCURRENTLY` and remove `NOT NULL DEFAULT` on large tables.

How long should the soak period be before contracting?

Depends on risk. We usually keep old paths for 1–2 sprints. If data volumes are huge or the change is core to revenue, leave it longer and schedule the contract step with the same gates.

Guides · Nov 5, 2025 · 10 minute read

The Migration That Didn’t Wake PagerDuty: A Real Zero‑Downtime Schema Strategy

I’ve watched teams brick prod with a “simple” ALTER TABLE. Here’s the expand/contract playbook—with metrics, commands, and guardrails—that actually ships without waking the on-call.

Back to all posts

The Migration That Didn’t Wake PagerDuty: A Real Zero‑Downtime Schema Strategy

Key takeaways

Implementation checklist