Total time budget

The problem

Bounding retries by count gives you an unpredictable total time: with exponential backoff, six retries might take half a second or half a minute. When you have a request budget or an SLA, what you actually care about is the wall-clock time.

The solution

maxElapsedTime stops retrying once that much time has elapsed in total. The final wait is trimmed so it never overshoots the budget.

await retry(callApi, {
  forever: true,
  initialDelayTime: 200,
  maxElapsedTime: 30_000, // keep trying for at most 30s, then reject
});

It composes with everything: combine it with forever for "try as hard as you can within 30s", or with times as a belt-and-braces upper bound.

Try it

The request never recovers here, so the budget is what stops it — note how it gives up partway through, not after a fixed number of attempts.

attempt timelineidle

• failed
• timed out
• bailed
• pending
• succeeded

Retries stop the moment the elapsed time crosses the budget.

maxElapsedTimetotal budget, ms (0 = off)foreverignore times; until success/abortinitialDelayTimems before the first retryfailUntilattempts that fail before success

1. Press Run to start.