Backend Development › Queues & Async Processing · also in Reliability & Resilience

Jitter

Randomizing retry delays so clients don't all retry in sync.

Backend Engineer · Senior Frontend Engineer · Senior Data Engineer · Senior

Also known as: jitter, exponential backoff with jitter, retry jitter

Jitter is a small random variation added to timing — especially retry and backoff delays — so that independent clients don’t all act at the same instant. Without it, clients that fail together tend to retry together, hitting the recovering service with a synchronized spike that knocks it over again.

no jitter:   all clients retry at t=1s, 2s, 4s, ...   (synchronized burst)
with jitter: retry at 1s ± random, 2s ± random, ...    (spread out)

It’s a standard part of retry strategy: exponential backoff (wait longer after each failure) plus jitter (randomise within the backoff window). The backoff reduces load; the jitter desynchronises.

The classic mistakes:

Fixed-interval retries. Every client waits exactly the same time and retries simultaneously — a self-inflicted thundering herd on the service that just recovered.
Backoff without jitter. Even growing delays synchronize if every client follows the same schedule. Jitter is what breaks the lockstep.
Retry storms after an outage. When a dependency recovers, thousands of queued retries fire at once. Jitter plus a retry budget/circuit breaker prevents the recovery itself from being the next outage.
Applying jitter only client-side. Servers and queues also benefit: delayed jobs, scheduled task start times, and cache expiry can all be jittered to spread load (see cache warming).
Over-randomising. Too much jitter makes timing unpredictable and hard to reason about; a bounded random factor around the backoff is the usual balance.
Unbounded retries. Jitter doesn’t fix a hopeless request; combine with a retry limit and dead-lettering (see dead-letter queue).
Forgetting the server side. Client jitter spreads requests, but the server also needs backpressure and circuit breaking to protect itself.

How to use it: on any retry path, use exponential backoff with jitter (randomise the wait), cap attempts, and pair it with a circuit breaker. Jitter is a tiny change with an outsized effect on stability — it turns synchronized failure-and-recovery cycles into smooth, spread-out load. It’s a standard tool for backpressure and resilience.