PipeCD Multi-Cluster Workshop

reading settings
font:
size:
spacing:

A hands-on workshop for Pre-DevFest Ibadan 2026. Attendees run PipeCD and two Kubernetes clusters on their own laptop, deploy one app to both, and practise previewing, canary rollouts and rolling back.

Kubernetes clusters on one laptop
2
hands-on exercises
4
setup script
1

The problem

Trying the multi-cluster plugin normally means setting up PipeCD and several Kubernetes clusters before anything happens. That is too much for one session, so most people never see a canary rollout across clusters for themselves.

What changed

  • One setup script brings up the PipeCD control plane and two kind clusters on a laptop, and a no-install path lets attendees without Docker still do every exercise.
  • Four short exercises cover plan preview, a canary rollout, updating one cluster first, and undoing a bad update, each using a branch, a preview and a merge.

What I worked on

  • I designed the workshop and its exercises for the Pre-DevFest Ibadan 2026 Cloud Session, building on my LFX work on PipeCD's multi-cluster plugin.
  • I wrote the setup, piped, preview, status and cleanup scripts, the attendee guide, worked solutions and a facilitator runbook, and tested it end to end before the session.

I ran this workshop at the Pre-DevFest Ibadan 2026 Cloud Session on 31 October 2026. It turns the Kubernetes multi-cluster plugin I built for PipeCD into something people can try in one sitting, on their own machine.

Each attendee forks the repo and runs one setup script. It starts the PipeCD console and two kind clusters, and a small agent called piped deploys the same app, hello-ibadan, to both clusters whenever they push to their fork.

The exercises

Every exercise follows the same loop: make a branch, edit the app, preview what will change, then merge to deploy and check both clusters.

  1. Change the version. A plain update that reaches both clusters at once, which raises the question of what happens if the new version is broken.
  2. Update slowly. A canary rollout that starts one copy of the new version, waits, then updates everything.
  3. One cluster first. Roll out to a single cluster before the other.
  4. Undo a bad update. Roll back when a release goes wrong.

The setup downloads about 1.2 GB, so the guide tells attendees to start it straight away and follow along on the projector until it finishes. Anyone without Docker, or with a struggling laptop, can use the no-install path and still work through every exercise. The workshop was tested end to end on macOS with OrbStack, and on Windows it runs inside WSL2.

Built with PipeCD, Kubernetes, kind and shell scripts.