Retry with backoff and jitter
Retrying a failed call after a wait that grows with each attempt (exponential backoff), with a random spread (jitter), so clients that failed together do not retry together. A retry budget or cap limits how many extra calls retries may add, so retries cannot multiply the load on a service that is already struggling.
Reliability
In Quorum#
Before Timeouts and retries research, every Link waits 30 s and never retries. After it, you set each Link's timeout to 1, 5, or 30 s and can turn on retries: naive retries fire at once, up to three times, so a callee where a share f fails gets 1 + f + f² + f³ calls per call. Backoff with jitter waits about 200 ms per retry and allows at most 10% extra calls, which keeps a Retry storm small.
Read next#
Metastable failure, Timeout, Circuit breaker, Thundering herd