Introduction

This action recreates a percentage of the RUNNING instances in a Managed Instance Group by calling GCP's recreateInstances API. GCP terminates the selected instances and provisions fresh replacements from the instance template while keeping the MIG's targetSize intact — this is the key difference from deleteInstances, which permanently shrinks the group.

The attack samples target instances uniformly from the current RUNNING set. For small pools it always picks at least one instance (rounded up from the requested percentage).

Use Cases

• Validate that stateless workloads recover cleanly when a fraction of the fleet disappears and is replaced with fresh instances.
• Measure cold-start cost: image pull, boot, startup scripts, connection-pool warmup on the new instances.
• Verify load-balancer health checks correctly drain terminating instances and route only to healthy replacements.

Rollback

This attack does not support rollback. GCP provisions fresh replacements automatically. Cancellation is a no-op.

Parameters