Backoff Lab

An API hiccups and a crowd of clients all retry at once. Tune exponential backoff and jitter, watch whether they stampede the rate limit or spread out politely, then copy the retry policy.

Scenario what goes wrong

Clients retrying

Initial arrival spread0 means everyone fires the same instant (a cron job, a deploy, a reconnect).

Outage lengthEvery request fails while the API is down.

Rate limit after recoveryRequests over the limit in any 100 ms slice get a 429.

Retry policy

Base delay

Multiplier

Max delay (cap)

Max retries

…

–succeeded

–p95 time to success

–peak load vs limit

–requests per client

Load on the API

servedrejected (down or 429)rate limitoutage

Every attempt, per client

failed attemptsuccessgave up

Same storm, every jitter strategy same seed

Strategy	Success	p95	Peak	Req/client

Worst-case schedule (no jitter luck)

Retry	Wait up to	Total elapsed

Copy the policy

Simulated locally in your browser with a seeded random generator. Nothing is sent anywhere. Settings save to this browser.