Stopniowe wdrażanie za pomocą Argo Rollouts

Progressive delivery means releasing a new model in controlled steps, with checks at each stage, instead of one risky flip to everyone. 🚦

Argo Rollouts is a Kubernetes controller that replaces the plain Deployment so you can run canary and blue-green releases natively in the cluster.

You swap your Deployment for a Rollout object. It looks familiar but adds a strategy block that describes how new pods take over traffic.

Inside the canary strategy you list steps: set a traffic weight, then pause. Argo walks them in order, shifting users to the new version slowly.

An empty pause halts the rollout until you approve it by hand. A timed pause waits a set duration, then continues on its own.

An AnalysisTemplate queries metrics, like error rate from Prometheus, and tells Argo whether the canary at this step is healthy.

If an analysis run fails its success condition, Argo aborts the rollout and shifts all traffic back to the stable version with no human needed.

For precise percentages, Argo drives a traffic provider like Istio or NGINX. Without one, weight is approximated by pod counts instead.

Argo also supports blue-green: it spins up the new version fully, lets you test it on a preview service, then flips all traffic at once.

You drive a paused rollout with the CLI: promote moves to the next step, while abort cancels and returns everything to stable.

The get rollout command shows each step, the current weight, and analysis results, so you can follow the promotion in real time.

How does Argo Rollouts react to a failing canary?

Argo Rollouts turns a Deployment into staged canary or blue-green releases, runs metric analysis at each step, and auto-aborts to stable on failure. ✅