---
title: "Rollbacks"
description: "Undo a release in seconds: a pointer flip, not a redeploy - and what rollback will not do."
canonical: "https://orbit-handbook.sitesmojo.com/docs/deployments/rollbacks/"
---

# Rollbacks

Bad releases happen. Orbit's contract: **rolling back is faster than
explaining what went wrong.**

## Undo the last release

```bash
orbit rollback --env production
```

That is the whole procedure. A rollback is a pointer flip to the previous
release - no rebuild, no redeploy, no waiting on CI. It completes in about a
second.

## Roll back to a specific release

```bash
orbit releases --env production   # list what is on disk
orbit rollback --env production --to 41
```

`release.keep` in [configuration](/docs/configuration/) controls how many
releases stay addressable (default 10).

## What a rollback does not do

- It does not run migrations in reverse. Data is yours; Orbit will not guess.
- It does not touch other environments.
- It does not delete the bad release - it stays on disk for the post-mortem.

## Practice it

A rollback you have never run is a rollback you do not have. Once a quarter,
roll staging back on purpose:

```bash
orbit rollback --env staging && orbit status --env staging
```

If that makes anyone nervous, that nervousness is the finding.