A release pattern that routes a small fraction of traffic to a new version first, so you find problems before they hit everyone.
A canary deployment is a release strategy that routes a small percentage of production traffic (the canary, often 1-5%) to a new version of a service while the rest of traffic stays on the old version. If the canary's error rate, latency, or business metrics regress, the rollout halts and the canary is rolled back; if it stays healthy, traffic shifts in increments until 100% is on the new version. The approach reduces blast radius for bad releases by orders of magnitude.
All-at-once deploys make every release a coin flip, the bug either ships to nobody or to everybody. Canary deploys cap the worst case at the canary slice, which means an SRE team can deploy faster (no week-long manual QA gates) without raising production risk. Pair canaries with automated rollback on burn-rate breach to remove the human reaction loop entirely.
See the part of the platform that handles canary deployment in production.