Bulkhead interceptor
Bulkhead is opt-in. Give work a bounded number of active slots and a bounded waiting queue:
import { createBulkheadInterceptor } from '@connectum/interceptors';
const bulkhead = createBulkheadInterceptor({ capacity: 20, queueSize: 20 });Use bulkhead: { capacity: 20, queueSize: 20 } in createDefaultInterceptors() for the same policy. In the default chain bulkhead wraps the retry loop, so attempts for one logical request share its slot.
Refusal and occupied slots
A request that finds every slot and the whole queue taken is rejected with ResourceExhausted and Bulkhead capacity exceeded (capacity: N, queueSize: M). The text contains configured limits, not live occupancy. Branch on the code; the human-readable text can change between releases.
Retry waits for already running work to settle after cancellation. This keeps the bulkhead slot occupied until that work actually finishes; an outer timeout or transport can stop the caller's wait earlier. A slot limit cannot forcibly stop signal-unaware work.
A queued call leaves the queue as soon as its signal aborts, freeing the place without running its handler. It fails with Canceled, preserving the abort reason itself when that is a ConnectError. An already aborted call fails the same way without taking a slot or queue place. A slot already held remains occupied until downstream work settles.
Option limits
capacity must be an integer of at least 1; queueSize an integer of at least 0. Both default to 10 when enabled. Wrong types throw TypeError; invalid numeric ranges throw RangeError at factory construction, before requests.
skipStreaming defaults to true and accepts only booleans; other types throw TypeError at construction.