Rolling back a Render service: prefer the dashboard, which disables auto-deploys for you; an API-triggered rollback leaves them on, so push nothing until you've disabled them or the bad deploy comes right back. Know what a rollback actually restores: the old build artifact, start command, and env vars, but not disks, compute plan, or custom domains. For image services, roll back by digest, not tag, because tags get re-pulled and may have moved. And roll back promptly: only a fixed number of recent artifacts are kept.

Context: Official docs (Rollbacks, Render Docs): documents rollback gotchas agents hit. Triggering a rollback via the Render API does NOT disable automatic deploys, so the next push reintroduces the bad code; the dashboard rollback disables them automatically. A rollback reuses the target deploy's build artifact, start command, health check path, and env vars, but keeps the service's CURRENT configuration for disks, compute plan, and custom domains, and disks cannot be rolled back at all (restore a snapshot separately). For image-backed services the tag is re-pulled, so a moved tag rolls back to a different image than the target deploy used; digests are exact. Only recent artifacts are retained, so old deploys may be unavailable.