The Runtime Theory

Canary Deployment Rollout: From 1% to 100%

Watch a canary rollout shift traffic through 1% to 100%: deploy the new version, monitor metrics at each step, and only continue when nothing regresses.

The Runtime Theory Team09 stages

trace / request.md

NEW VERSIONIS BUILTONE INSTANCEJOINS THE FLEETMETRICS ARECOMPAREDSHIFT TOTEN PERCENTHALF THE WORKLOADROLLBACK ISALWAYS READYALL TRAFFIC MOVESOLD FLEET DRAINSPIPELINE ISREADY AGAIN

readyThe new version is built and containerized, and its image is pushed to the registry. From this point both versions exist simultaneously — the old one still serving all traffic.

A canary deployment trades a big risk for a sequence of small ones. This diagram walks the rollout: build the new version, deploy one instance, shift traffic through increasing percentages — 1%, 10%, 50%, 100% — with a metrics gate at every step. The mechanism is pure routing: the load balancer controls how much traffic each version receives, so the rollout is a series of configuration changes, not a series of deploys.

The ordering of the stages is the safety design. The canary starts tiny so that if the new version is catastrophic — a panic on boot, a broken request path — only a sliver of users feel it. Each shift is gated on a clean observation window, so problems surface at the smallest possible blast radius. And because rollback is just rerouting traffic, reverting is seconds rather than a redeploy. The failure model is worth stating plainly: a canary reduces the damage from bad releases; it does not eliminate them.