Progressive delivery: printable course notes
Blue-green and canary releases with Argo Rollouts on the Argo CD platform: install from Git, migrate a Deployment, blue-green with a preview, analysis templates on Prometheus, a self-rolling-back canary, Istio and Gateway API traffic routing, team operations, and a failed-release drill.
These pages are the same content as the interactive course, on one page per module, for reading and printing.
Four pieces
| Piece | What it is |
|---|---|
| See the end result | Not the first step: a 30-minute preview of a canary that rolls itself back when its error rate climbs, with few explanations. |
| Full course | 10 modules on the setup you built in the Argo CD course: Argo Rollouts, blue-green, analysis with Prometheus, a canary that rolls itself back, Istio and Gateway API traffic, team operations and a failed-release drill. |
| Reference repo | The finished implementation: the three repos as they look at the end of this course, plus the docs. You copy from it. |
| Your three repos | argocd-platform, checkout-app and checkout-config from the Argo CD course. This course changes all three. |
Your setup
Commands in these notes use these values (shared with the interactive course, stored only in this browser).
Outline
| Module | Part | Time | You are | You will |
|---|---|---|---|---|
| 00 Orientation: deploy is not release | Foundations | 20 min | Platform and product | Know what progressive delivery adds to the Argo CD setup you already run, check your tools, and confirm (or rebuild) the end state of the Argo CD course. |
| 01 Install Argo Rollouts from Git | Foundations | 25 min | Platform team | Install a pinned Argo Rollouts controller and its read-only dashboard with an Application and values in Git, and find your way around the dashboard and the kubectl plugin. |
| 02 From Deployment to Rollout | Foundations | 40 min | Product team | Release checkout chart v2.0.0, in which the Deployment is a Rollout, ship a new version to dev as a canary that waits for a person, promote it, and handle an aborted release through Git. |
| 03 Blue-green in dev | Releases judged by metrics | 30 min | Product team | Run dev as blue-green: every new version comes up as a full copy behind a preview host, waits for a person, and replaces the old one in one switch. |
| 04 Metrics and analysis templates | Releases judged by metrics | 40 min | Platform team | Install a small Prometheus and a load generator from Git, write the platform's shared success-rate check as a ClusterAnalysisTemplate, and let it promote (or stop) dev's blue-green releases. |
| 05 Canary with automatic rollback in prod | Releases judged by metrics | 35 min | Product team | Run prod as a stepped canary with background analysis, ship a good release that walks through the steps alone, and a bad one that aborts itself at the first step. |
| 06 Istio: exact weights and header routes | Traffic | 50 min | Platform and product | Install Istio from Git, put prod and the load generator in the mesh, and release checkout with exact 5% steps, a header route for testers, and mutual TLS required in prod. |
| 07 Gateway API: the lighter option | Traffic | 30 min | Platform and product | Release dev as a canary that splits traffic at the edge with the Gateway API on Traefik and the Argo Rollouts Gateway API plugin, with no mesh, and know when each kind of routing fits. |
| 08 Operate releases with teams | Operate | 30 min | Platform and product | Let team-checkout resume, abort and retry its own releases from Argo CD and nothing more, send every finished or aborted release to a release log, abort releases stuck past their deadline, and write the release runbook. |
| 09 Capstone: the failed-release drill | Operate | 45 min | Platform and product | Handle a release that every check passes and users still feel: notice it, roll it back fast, add the missing check, prove it catches the next one, write the incident, and put catalog on a canary too, without step-by-step commands. |
The release path
- Committed: A new version or setting in Git
- Deployed: New pods run, users not there yet
- Exposed: A share of users, a header, a preview host
- Measured: Prometheus: errors and latency of the new pods
- Decided: Analysis or a person: go on or abort
- Promoted: All traffic on the new version
- Rolled back: Abort, then a revert in Git
Where the cluster runs
kind on your laptop is the default for every module. On an existing cluster, use your Argo CD hostname for the CLI; the rest is the same.
Between you and production
- Analysis on every production release: Every prod Rollout runs at least a success-rate template; a release without one is a review comment.
- Metrics you trust: Prometheus with retention and high availability (or a managed one), and the request metrics your apps actually expose, not podinfo's.
- Alerts on aborted releases: Rollouts notifications to the team channel, not a log: Slack, Teams or a webhook that pages during working hours.
- A runbook per app: Who may abort, who may retry, how to revert in Git, and where the dashboards are. Linked from the repo README.
- Synthetic traffic or real traffic: Analysis needs requests. Low-traffic apps need a load generator or a longer pause, or the template returns no data.
- Mesh or gateway, chosen once: One traffic router for the platform, documented, upgraded monthly by pull request like every other add-on.
- Database changes that survive two versions: Blue-green and canary run old and new side by side: schema changes must be backward compatible (expand, then contract).
- SSO for the dashboards: The Rollouts dashboard and Argo CD behind SSO, with the same team roles.
- Release windows: Argo CD sync windows or a team agreement on when prod releases start, so a canary is never left running overnight.
- Upgrade cadence: Argo Rollouts, its plugin, Istio and Prometheus upgraded by pull request, release notes read.