Progressive Delivery and Automated Rollback

Delivery and GitOps / Guide 46 of 46 / Intermediate

• Kubernetes1.36.4
• Cluster4 nodes
• Runtimecontainerd 2.2.6
• CNICalico v3.32.1
• TimeAbout 20 min

Argo Rollouts is a CNCF incubating project and its CRDs are still v1alpha1. The step and analysis fields used here have been stable across 1.x.

• Argo Rollouts1.9.1
• Prometheus3.14.0

This guide includes

Use this when a new version has to prove itself against real traffic before it takes all of it. This matters because a Deployment offers one rollout strategy and no opinion about whether the new version is any good - progressive delivery adds the gate that decides.

• installing the Argo Rollouts controller with its Rollout and AnalysisTemplate CRDs
• shifting twenty percent of traffic to a canary and reading the rollout's step state
• watching an analysis step query Prometheus and gate promotion on the answer it gets
• promoting past a pause: {} step that would otherwise wait forever
• setting an image tag that cannot run, and watching the rollout stop instead of proceeding

Before you start

• A working Prometheus in the monitoring namespace, for the analysis step.
• The rolling-update guide, since the whole point here is what a Deployment cannot do.

1. 1

   Install the controller, and meet the first surprise

2. 2

   Twenty percent first

3. 3

   The gate asks Prometheus

4. 4

   Promote, and what one promote actually does

5. 5

   A version that cannot run

Official sources

• Argo Rollouts canary strategy
• Analysis and progressive delivery
• Prometheus metric provider
• Deployment strategies