Introduction

This action triggers a failover of a STANDARD_HA Memorystore for Redis instance. The replica in the alternate zone is promoted to primary — exercising the same code path a real zonal outage on the primary would.

The endpoint host (10.x.x.x) is stable across the swap; only the underlying zone changes.

Use Cases

• Verify that clients using client-side connection pools cleanly reconnect and re-establish tracking / auth after the primary flips.
• Understand the performance impact of a cold cache immediately after failover, and how quickly the read/write hit-rate recovers.
• Validate that observability alerts fire (or don't fire) as expected during a controlled failover.

Rollback

This attack does not support rollback. Memorystore performs the promotion and rebuilds the replica automatically. Cancellation is a no-op.

Parameters

Notes

Failover requires STANDARD_HA tier. BASIC tier instances cannot fail over (no replica exists). The gcp.memorystore.tier attribute lets you scope this attack in your selector.

If the replica is not caught up when running in LIMITED_DATA_LOSS, GCP will complete the operation as done but leave the primary in place — no role swap occurs. Use FORCE_DATA_LOSS when you need to guarantee an actual promotion for chaos purposes.