Skip to main content

What's new in CMK v2

CMK v2 is the next version of Crusoe Managed Kubernetes (CMK). Starting November 7, 2026, every new cluster is a CMK v2 cluster, except in us-southcentral1-a, where new clusters are still CMK v1 clusters. Every node pool in a CMK v2 cluster is a CMK v2 node pool. Existing clusters stay on CMK v1, and so do node pools you add to them later.

You manage both versions with the same CLI commands, API, Terraform resources, and console pages. This page lists what changes.

Check whether a cluster is CMK v2​

A CMK v2 cluster reports its Kubernetes version without a -cmk suffix, such as 1.35.8. A CMK v1 cluster reports a version such as 1.34.5-cmk.23. Check the cluster's version, not a node pool's, because node pool versions keep the -cmk suffix in both.

The API also returns control_plane_generation on the cluster: 2 for CMK v2 and 1 for CMK v1.

Changes​

  • Cluster version: When you create a cluster, specify the Kubernetes minor version, such as 1.35, and CMK uses the latest supported patch. If you specify a patch, CMK uses it and ignores a -cmk suffix. CMK v2 supports fewer minor versions, so check existing commands and Terraform configurations against crusoe kubernetes versions list.
  • Node pool settings: Node pools have new settings for scale-down and remediation, and report their running node count and health issues. All other settings stay the same. See Node pools in CMK v2.
  • Changes apply to running nodes: Node pool changes, including labels and taints, apply to existing nodes. Kubernetes version, ephemeral storage for containerd, and SSH key changes still apply only to new nodes.
  • Scale-down: If you turn on a pool's scale-down setting, lowering its node count removes nodes. Don't turn it on for a pool that the Cluster Autoscaler manages.
  • Node pool states: A pool that is short of its node count because nodes couldn't be created reports STATE_RUNNING_WITH_ERROR instead of STATE_UNHEALTHY.
  • AutoClusters: AutoClusters is installed on all clusters and is configured on each node pool with a remediation policy. By default, Crusoe repairs issues automatically as they arise. You can also set a remediation policy where Crusoe proposes each repair with a node label and you approve it with a different node label, or turn off remediation entirely. Crusoe reports faults as node conditions and repair progress as Kubernetes events. See AutoClusters in CMK v2.