By default, ten revisions. The Helm reference says history-max defaults to 10 for both upgrade and rollback, and 0 means no limit. A rollback with no revision goes to the previous release. Customers can name a revision from the history list, as long as it is still saved.
Definitions
A Helm revision is one recorded state of a release. The history command lists them with numbers. The rollback command takes a release name and optionally a revision number, and if you omit it or pass zero, it goes back to the previous release.
Useful rollback flags
| Flag | Effect per the reference |
|---|---|
| wait | Wait for resources to become ready, up to the timeout |
| cleanup-on-fail | Allow Helm to delete resources the rollback created if the rollback fails |
| force-replace | Force resource updates by replacement |
| force-conflicts | Force changes through conflicts when using server side apply |
| history-max | Cap saved revisions, default 10, 0 for none |
| no-hooks | Skip hooks during the rollback |
The page I read does not list a recreate pods flag, so do not put one in your guide without testing it on your supported Helm version.
What vendors should write down
- State how many revisions customers should keep. Ten is the default. If your upgrades are frequent, say whether to raise it.
- State what a rollback does not undo. It restores the Kubernetes objects. It does not restore data, and it does not reverse a database migration. Name the first release where a migration is not reversible.
- Warn about hooks. A rollback runs hooks unless told not to, so test yours in both directions.
- Say whether rolling back is acceptable after a security release. Going back means running code with a known flaw. See is rolling back safe if the old version has known vulnerabilities.
A practical test
Before each release, upgrade a test cluster from the previous version, roll back, and confirm the product starts and serves traffic. Then upgrade again. If the second upgrade fails, you found a hook or migration problem before a customer did.
The security tradeoff
A rollback is a reasonable first move when an upgrade fails. It should not become a way of staying on a vulnerable version. Tell customers to treat a rollback as temporary and to contact you with the failure details, so you can fix the upgrade path.
If you want a second look at what you ship, Safeguard can generate and scan an SBOM and run dependency scanning with reachability analysis, and you can try the free scan.
Quick check
What is the default number of saved Helm revisions?
Ten, per the Helm reference. Zero means no limit.
Does helm rollback restore my data?
No. It restores Kubernetes resources to an earlier revision. Data and schema changes need their own plan.
Sources
Last reviewed: 19 October 2026.
