EKS cluster upgrades now reversible within 7 days

EKS cluster upgrades are now reversible within a seven-day window.

What changed

• Amazon EKS now supports automated rollbacks for Kubernetes version upgrades.
• Rollbacks can be initiated within seven days of an upgrade.
• The feature applies to managed node groups and Fargate profiles.

Why it matters

This feature addresses a significant operational pain point for EKS users: the risk associated with cluster version upgrades. Previously, a failed upgrade could necessitate manual intervention, potentially leading to cluster rebuilds and extended downtime. The ability to automatically reverse an upgrade within a defined window substantially reduces the blast radius of a bad deployment. This is a pragmatic improvement for any team running production workloads on EKS, particularly those with complex add-on configurations that might interfere with new Kubernetes versions.

The catch

The honest version: While a seven-day rollback window is a substantial improvement, it's not infinite. Workloads that require immediate rollback beyond this period will still face manual remediation. Watch out: The rollback functionality is limited to managed node groups and Fargate profiles. Self-managed nodes will not benefit from this automated feature, requiring traditional manual rollback procedures. Furthermore, the announcement does not specify the exact rollback mechanism or any potential data consistency issues that might arise for stateful applications during the rollback process.

Ship it

If you are running production EKS clusters on managed node groups or Fargate, enable version rollbacks proactively before your next planned upgrade. Review the EKS documentation for specific configuration steps and ensure your upgrade strategy accounts for the seven-day window. Consider testing the rollback mechanism in a non-production environment to understand its behavior with your specific application stack.

Bottom line: EKS upgrades can now be undone within 7 days, reducing risk for managed nodes and Fargate.

— Filed to /blog