Cancellation & timeouts ​

Cancellation in ReqMind works at two levels: a tracker state machine that every request flows through, and an underlying fetch signal that is actually aborted so sockets/resources are released.

cancel() — the cancellable promise ​

Every request returns a CancellablePromise<ApiResponse<T>>:

const request = api.get("/big-report", { timeout: 15_000 });

// …later
request.cancel();          // rejects with CancelledError
await request;             // throws

External AbortSignal ​

const controller = new AbortController();
api.get("/users", { signal: controller.signal });

controller.abort();             // same cancellation path

An already-aborted signal rejects immediately.

Timeouts ​

const api = createClient({ timeout: 5_000 });        // default for all requests
await api.get("/slow", { timeout: 1_000 });          // per-request override

const api = createClient({ timeout: 0 });            // disable globally

When the deadline hits, the abort fires and the request rejects with TimeoutError. The timeout only applies to the time a single network attempt is allowed to take (not the whole retry series).

Error types ​

isAbortError(err) detects raw AbortErrors (e.g. from a lower-level layer).

Tracker states ​

Every request exposes its Tracker, emitting events as it moves:

idle → pending → retrying · retrying · … → success | error | cancelled

The tracker is available on request/success/error/cancel/retry event payloads and via .state:

api.on("cancel", ({ tracker }) => console.log(tracker.state)); // "cancelled"

Notes ​

• cancelAll() cancels every in-flight request tracked by the client and clears the deduper and in-flight map.
• Cancelled/timed-out flights never populate the cache and never fire success.
• A cancelled in-flight read is removed from the in-flight/dedup registry so a later identical call starts fresh instead of joining a dead flight.

Next: cache invalidation.