Skip to content

Retries, backoff and jitter

When a retry helps and when it starts a retry storm: amplification across layers, capped exponential backoff, full jitter and retry budgets.

30 min7 steps

By the end you can
  • Decide which failures are worth retrying and which are not
  • Compute how retries at several layers multiply load on a failing dependency
  • Compute capped exponential backoff schedules and explain why jitter matters
  • Bound total retry load with a retry budget and choose where in the stack to retry

Retries fix transient failures, and only those

Retry

  • Dropped packet
  • Election in progress
  • 503 / 429 after waiting

Never

  • 400 Bad Request
  • 403 Forbidden
  • Fails identically

Careful

  • Non-idempotent op after timeout
  • Might run twice

Which failures are reasonable to retry, with backoff? Select all that apply.

Select all that apply.

Answer this step to continue.