Stackup

A production-shaped Kubernetes cluster on a laptop, in one command.

  • Kubernetes
  • kind
  • ArgoCD
  • Argo Rollouts
  • Prometheus
  • Grafana
Code

The problem

Practising real deployments usually means renting a cloud cluster you pay for every month. And canary rollouts, metrics and GitOps only make sense together: you cannot learn how a bad rollout shows up in a dashboard if the dashboard was never wired in.

How it works

Stackup bootstraps a kind cluster, ArgoCD syncs the rest, Argo Rollouts canaries advance only while the success rate holds and roll back on their own.

  • One make target starts a cluster on a laptop with Docker and kind.
  • ArgoCD then takes over and deploys whatever is in the Git repository, so the cluster is whatever the repo says it is.
  • A new version goes to a small share of traffic first. Prometheus watches the success rate, and the rollout only moves forward while it holds. If it drops, the rollout reverts without anyone stepping in.
  • Grafana shows the same metrics the rollout is gated on.

What it does not do

  • kind runs on one machine, so node failure and cross-node scheduling are out of scope.
  • There is no real autoscaling to practise against.
  • Volumes are local, with no cloud storage class.

Links