This repository contains a Go-based Kubernetes Operator (controller-runtime) that automates progressive delivery (10% → 50% → 100%) using Prometheus SLOs with automatic rollback on breach. It also includes Helm charts, Kustomize overlays, ArgoCD manifests, Gatekeeper/OPA policies, k6 load tests, Grafana dashboards, and runbooks.
Highlights
- Go controller watching a
CanaryCRD. - Progressive traffic shifting with SLO guardrails (p95 latency, error rate, error budget).
- Rollback within ~30s on regression.
- GitOps-friendly: Helm chart + Kustomize overlays + ArgoCD.
- Policies via Gatekeeper for image signing and resource limits.
Quick Start
- See
charts/canary-operatorfor Helm deployment. - See
config/overlays/*for Kustomize overlays per env. - See
argocd/*for ArgoCD Apps. - Apply Gatekeeper policies in
policies/gatekeeper. - Run load tests from
loadtest/k6.
The controller can be tested without a Kubernetes cluster. From operator/, run
go test ./... and go vet ./.... From the repository root, run helm lint charts/canary-operator and render each directory under config/overlays with
kustomize build.
The controller samples Prometheus on every reconciliation so an SLO breach can
roll back early, but it will not advance a rollout step until stepInterval has
elapsed. Existing Services are patched by annotation and are never adopted as
children of the Canary resource.