Introduction

This action recreates a percentage of the node instances in a GKE node pool by calling recreateInstances on the underlying Managed Instance Group(s). GCP terminates the selected nodes and provisions fresh replacements while preserving the node pool's size — unlike deleteInstances, which permanently shrinks the pool.

Pods on affected nodes are drained by kubelet's node-shutdown handling and reschedule onto surviving or freshly-recreated nodes (subject to PodDisruptionBudgets, taints, tolerations, and topology spread constraints).

Use Cases

• Validate Pod Disruption Budgets under partial node loss.
• Verify taints/tolerations and topology-spread constraints keep critical workloads spread across surviving nodes.
• Measure cold-start cost on new nodes: image pull, CSI volume re-attach, init containers, DaemonSet startup.

Rollback

This attack does not support rollback. GKE recreates the terminated nodes automatically. Cancellation is a no-op.

Parameters