Skip to main content

Preemptible node groups in a Managed Kubernetes cluster

In a preemptible node group, each node runs for a maximum of 24 hours. During this time, nodes can be stopped by Servercore at any time — for example, if the virtual host does not have enough resources for other clusters.

Recovery starts immediately after the node is stopped. If there are enough free resources for recovery, the duration of node recovery will depend on the boot volume type:

  • a node with a network boot volume is recovered in less than one minute;
  • a node with a local boot volume is recovered in up to 10 minutes.

Data recovery on the node also depends on the boot volume type. If a network boot volume is used, all data is preserved; if a local one is used, it is deleted.

After each recovery, the 24-hour countdown starts over.

Preemptible node groups support all features available for regular node groups, and the cost of resources in such groups is 70% lower on average.

You can make a node group preemptible in the control panel when you create a cluster or add a node group. You can make a node group of any configuration preemptible.

Limitations

Preemptible node groups are available only in clusters on cloud servers in the regions of St. Petersburg, Moscow, and Novosibirsk.

We do not guarantee the same availability level as for regular node groups — preemptible node groups are not subject to the Kubernetes cluster SLA.

Cost

The cost of resources in a preemptible node group is 70% lower on average than the cost of resources in a regular node group with the same configuration.

During operation, resources in a preemptible node group are charged according to the cloud platform resource billing model.

After interruption:

  • vCPU, RAM, GPU, and local disks are no longer charged starting from the next hour after the stop;
  • network volumes continue to be charged.