Exponential Backoff and Jitter — Idempotency & Retry-Safe Systems

Immediate retries hammer a struggling dependency. Use exponential backoff (base × 2^attempt, capped) and jitter (randomization) so clients don't retry in synchronized waves. AWS's analysis shows full jitter (sleep = random(0, min(cap, base·2^n))) or decorrelated jitter spread load best. Cap attempts and total time, and honor server hints (Retry-After).

Backoff strategies

Strategy	Load on recovering service	Completion time
Immediate retry	Worst (synchronized spikes)	Fast when it works
Fixed delay	Bad (synchronized waves)	Predictable
Exponential, no jitter	Better but clustered	Grows quickly
Exponential + full jitter	Best spread	Good
Decorrelated jitter	Very good	Good

import random

def full_jitter(attempt: int, base: float = 0.1, cap: float = 10.0) -> float:
    return random.uniform(0, min(cap, base * 2 ** attempt))

async def with_retries(op, attempts=4):
    for n in range(attempts):
        try:
            return await op()
        except Retryable as e:
            if n == attempts - 1:
                raise
            await asyncio.sleep(e.retry_after or full_jitter(n))

const fullJitter = (attempt: number, base = 100, cap = 10_000) => Math.random() * Math.min(cap, base * 2 ** attempt);

export async function withRetries<T>(op: () => Promise<T>, attempts = 4): Promise<T> {
  for (let n = 0; ; n++) {
    try { return await op(); }
    catch (err) {
      if (!isRetryable(err) || n === attempts - 1) throw err;
      await new Promise((r) => setTimeout(r, retryAfterMs(err) ?? fullJitter(n)));
    }
  }
}

RetryConfig config = RetryConfig.custom()
    .maxAttempts(4)
    .intervalFunction(IntervalFunction.ofExponentialRandomBackoff(Duration.ofMillis(100), 2.0, 0.5, Duration.ofSeconds(10)))
    .retryOnException(e -> e instanceof RetryableException)
    .build();
Retry retry = Retry.of("inventory", config);
Supplier<Stock> call = Retry.decorateSupplier(retry, () -> inventory.stock(sku));

Common pitfalls

No cap on the backoff or attempts.
Jitter omitted in fleets of clients.
Retrying non-retryable errors.

Key points

Exponential backoff with full jitter
Cap attempts and total time
Honor Retry-After

Further reading: AWS: Exponential backoff and jitter

Practice questions

Why add jitter to exponential backoff?

Answer

Without randomness, clients that failed together retry together, creating synchronized load spikes; jitter spreads retries over time.

What is full jitter?

Answer

Choosing the sleep uniformly at random between 0 and the capped exponential backoff value for that attempt.