Jitter is the cheapest reliability fix you are not using

2026-09-10

Distributed systems synchronise themselves without being told to: shared events such as deploys, cache flushes, or connection loss turn independent workloads into recurring load spikes. Add jitter to initial retries, TTL refreshes, heartbeats, token renewals, and scheduled work. This preserves average load while preventing retry storms from repeatedly hitting a recovering dependency in lockstep.