Communication

Give a recovering service room to recover

Editorial edition · 16 November 2026

The idea

Retries can amplify an outage when every client sends another request at the same instant. Backoff spreads attempts over time. It belongs with a maximum retry budget, otherwise a failed operation can occupy resources long after it remains useful.

Design choice

Increase delays between transient failures and add bounded randomness to avoid synchronised clients. Honour a meaningful server retry hint. Do not automatically retry requests whose side effects cannot be repeated safely.

Worked example

An uploader retries a temporary outage three times within a total deadline. After the deadline, it preserves the local work and offers an explicit retry action.

Before you ship

  • Set a total deadline
  • Add jitter to retry intervals
  • Distinguish transient from permanent failures

Examples are illustrative. Adapt the checks to your own environment.

Further reading

HTTP Semantics, RFC 9110

Search the library