PATTERN Cited by 1 source
Failover cluster escape hatch¶
Definition¶
When a managed service's control plane becomes unavailable during an incident (blocking scaling, reboots, or reassignment), recovery can stall indefinitely. The failover-cluster escape hatch pre-provisions a clean parallel cluster using infrastructure-as-code templates so operators can route producers and consumers to an entirely fresh environment without depending on the degraded cluster's management APIs.
How it works (Atlassian StreamHub)¶
A four-stage runbook:
- Shard failover (~15 min): Route traffic from the affected shard to a healthy shard; apply rate limits and quarantine to stay within target capacity.
- Prepare (~2 hours): Provision a new MSK cluster via IaC templates (topics, tiered storage config, ACLs, client config).
- Evaluate: Continue mitigating the primary cluster while the clean environment initializes.
- Cluster failover: If the primary remains stuck past a decision threshold, update routing so all producers/consumers move to the clean cluster.
Why this pattern exists¶
Managed streaming services abstract away operations but introduce a hidden dependency: the control plane. When the control plane is unhealthy, you cannot scale brokers, expand storage, rebalance partitions, or complete automated recovery — precisely the levers needed during incidents. The escape hatch ensures recovery does not depend entirely on the failing control plane becoming healthy.
Seen in¶
- sources/2026-07-28-atlassian-scaling-streamhub-transitioning-from-kinesis-to-kafka — Atlassian StreamHub at 150B events/day