A resilient client

Retry on transient failures, with a deadline on each attempt.

The client

import { corgi } from '@itsy/corgi';
import { withRetry } from '@itsy/corgi/retry';
import { withTimeout } from '@itsy/corgi/timeout';

const api = corgi.create({
  baseURL: 'https://api.example.com',
  plugins: [withRetry({ retries: 3, backoff: 300, maxDelay: 10_000 }), withTimeout(5000)],
});

await api.get('/users');

Listing order doesn't matter. corgi sorts plugins, so retry (300) always ends up outside timeout (400).

Why retry sits outside timeout

Because retry wraps timeout, each attempt gets its own fresh 5s deadline and a timed-out attempt is itself retried. That's almost always what you want, since one slow attempt shouldn't burn the whole budget.

Read more: per-attempt vs total budget →

Tuning retry

The defaults cover the common case: idempotent methods only, replayable bodies only, transient statuses (408/429/500/502/503/504), Retry-After honoured, and full-jitter backoff so many clients don't retry in lockstep.

Observe or tune it with onRetry.

withRetry({
  retries: 3,
  onRetry: ({ attempt, delay, response, error }) => {
    console.warn(`retry #${attempt} in ${delay}ms`, response?.status ?? error);
  },
});

Read more: every retry option →

Cancel-previous for interactive clients

If this client backs a UI that fires overlapping requests (search, filters), add abortPrevious. It takes the outermost slot, so a new call cancels the entire prior chain, retries included.

const api = corgi.create({
  plugins: [abortPrevious(), withRetry(3), withTimeout(5000)],
});