Retries, timeouts, and cancellation

Resource methods and request() retry temporary connection failures and limit how long each request attempt can run. The low-level fetch() method does not use these behaviors.

Retry behavior

By default, the client retries a NETWORK_ERROR or TIMEOUT up to two times. The first call plus two retries means a request can make three attempts.

HTTP errors are not retried. This includes rate limits and server error status codes.

The client waits for a randomized delay before each retry. The first delay is at most 500 milliseconds. The upper limit doubles for later retries and stops growing at 5 seconds.

You can change the retry count when you create a client:

import { RankedClient } from "mcsrranked-sdk";

const noRetries = new RankedClient({ retries: 0 });
const moreRetries = new RankedClient({ retries: 4 });

retries must be a finite, non-negative integer.

Query libraries

TanStack Query and SWR can also retry. Disable retries in one or the other, so failures don't stack retries.

Timeouts

Each attempt has a 10 second timeout by default:

const ranked = new RankedClient({ timeout: 5_000 });

The timeout is measured in milliseconds and must be a finite positive number. When an attempt times out, the client can retry it. If no retries remain, the request throws RankedError with code TIMEOUT.

Cancellation

The low-level request() method accepts an AbortSignal:

const controller = new AbortController();

const pending = ranked.request("users/NotAva1able", {
  signal: controller.signal,
});

controller.abort();

await pending;

An aborted request throws RankedError with code ABORTED. Cancellation also stops a pending retry delay.

Resource-method limitation

Resource methods such as users.get() do not accept an AbortSignal in the current beta. Use the low-level request() method when cancellation is required.

See the errors reference for the error properties.