PATTERN Cited by 1 source
Cluster sharding for blast radius¶
Definition¶
Split a large streaming cluster into multiple smaller shards (independent clusters), each handling a subset of topics/tenants. During an incident, traffic from an affected shard is quickly routed to a healthy one — limiting the blast radius to a fraction of total throughput rather than the entire platform.
Trade-offs¶
- Pro: Reduces blast radius; enables partial failover within minutes (~15 min at Atlassian).
- Pro: Limits the "hot broker" problem to one shard rather than cross-contaminating all traffic.
- Con: More clusters to manage; cross-shard coordination requires routing intelligence.
- Con: Receiving shard must have spare capacity or accept rate-limited traffic during failover.
Seen in¶
- sources/2026-07-28-atlassian-scaling-streamhub-transitioning-from-kinesis-to-kafka — Atlassian StreamHub shards clusters to reduce per-incident blast radius