Introduction

This action deletes a percentage of nodes from a managed AKS node pool via the AKS BeginDeleteMachines API.

With the cluster-autoscaler enabled on the node pool, AKS replaces the deleted nodes within minutes. Without autoscaling, the pool shrinks until manually scaled back. AKS forbids deleting all nodes from a System pool — the experiment fails Prepare with an actionable error in that case.

Use Cases

• Validate that pods reschedule onto surviving nodes within the configured PodDisruptionBudgets.
• Measure cluster-autoscaler scale-up time and the impact on application availability during node churn.
• Verify that StatefulSet, DaemonSet, and PDB-protected workloads recover correctly when nodes disappear.
• Exercise alerting around node count drops or pool health degradation.

Rollback

This action does not support rollback — the AKS delete-machines call is fire-and-forget. Either the cluster-autoscaler restores the pool size, or you scale the pool manually after the experiment.

Parameters

Parameter	Description	Default
Percentage of nodes to terminate	Percentage (1-100) of the node pool's nodes to terminate.	33