advanced

Bulkhead

Isolate pools, queues, threads, or capacity so one dependency or tenant cannot consume resources needed by the whole service.

Bulkheads partition resources — thread pools, connection pools, queue capacity, rate limits — so overload in one area does not exhaust the whole service. Named after ship compartments that contain flooding.

Trade-off: reserved capacity may sit idle while another partition starves if sizing is wrong.

On interviews: apply bulkheads to a service calling both fast cache and slow payment API; explain tenant isolation.

Common pitfalls: one giant pool for all dependencies; bulkheads without metrics; no rejection policy when partition is full.

Checklist:

Separate pools per critical dependency or tenant.
Define max concurrency and queue limits.
Reject or shed load when partition saturates.
Monitor utilization per bulkhead.