Skip to content

kind E2E tests

Run a target from the repository root (Docker, kind, kubectl, Python 3 and curl required):

bash scripts/e2e/run.sh argocd
bash scripts/e2e/run.sh prometheus-operator
bash scripts/e2e/run.sh prometheus
bash scripts/e2e/run.sh grafana
bash scripts/e2e/run.sh mysql
bash scripts/e2e/run.sh helm       # also requires Helm
bash scripts/e2e/run.sh kustomize  # also requires standalone Kustomize

Each invocation creates a disposable kind cluster, waits for CoreDNS, verifies workload readiness and an application endpoint, then deletes the cluster even on failure. Prometheus Operator also checks that Prometheus discovers and successfully scrapes itself. Run local targets sequentially to avoid exhausting Docker Desktop resources. The versions come from the sample manifests, including the pinned Operator bundle in prometheus-operator/operator/kustomization.yaml.

Every kubectl call uses an explicit --context and --kubeconfig. Both kind create and delete use a fresh, private kubeconfig. Neither $HOME/.kube/config nor files in an existing $KUBECONFIG are written; your current-context stays unchanged. Tests do not reuse or delete existing clusters. Do not replace k with bare kubectl inside a case.

The default node image is kindest/node:v1.36.1. To test another Kubernetes version:

KIND_NODE_IMAGE=kindest/node:v1.37.0 bash scripts/e2e/run.sh argocd

Failure diagnostics (events, pod descriptions/logs and kind logs) are saved in .e2e-artifacts/<target>/ and uploaded by CI. Override this with E2E_ARTIFACTS. A forced process kill can prevent cleanup; use the cluster name in the run log and an explicit disposable --kubeconfig when deleting a leftover cluster.

Prometheus tests readiness and a successful self-scrape query. Grafana tests startup with the repository's dashboard/datasource provisioning and database health; it does not provision the optional MySQL or RabbitMQ dashboard backends.

Grafana Operator tests the HA sample with PostgreSQL 18.6, two ready Grafana replicas, database health through the Service, and Grafana-created SQL tables. Optional Prometheus datasources and external dashboards are outside this test. Run it with bash scripts/e2e/run.sh grafana-operator.

Helm tests the hello-world chart's install, connection hook, upgrade with optional Ingress/HPA resources, HTTP response and uninstall. Kustomize tests the portable contents/kustomize/example dev/prod overlays, content and replicas. These tests do not cover helm-vs-kustomize's older Flask/MySQL example or the packaged Helm archive. CI installs Helm and Kustomize versions from their README version links, so version-only README updates exercise the new binaries. Local runs use the binaries on PATH and print their versions; use the documented versions for parity.

Changed targets in CI

PRs select targets from the merge-base diff, including deleted and renamed files. contents/argocd/** runs only Argo CD; contents/prometheus-operator/** runs only Prometheus Operator. contents/prometheus/** and contents/grafana/** select the corresponding standalone sample. The hello-world chart and Helm README select helm; contents/kustomize/** selects kustomize. Changes to an individual case run that case; shared runner, selector or workflow changes run all registered targets. Registry edits compare the PR merge-base and head: only added targets or targets with changed path patterns run. Deleted targets are not scheduled, and registry formatting/order changes do not create clusters. Unit-test-only changes run the selection/isolation checks without clusters; the tool-version helper selects only Helm and Kustomize. New commits cancel superseded E2E runs for the same PR. Unrelated changes skip cluster jobs. README changes under a target are included because installation versions and instructions can live there. Manual workflow_dispatch runs all. status-check-e2e is the stable aggregate check, including when no targets match. Go-only examples remain covered by the separate Go workflow.

ECK tests the pinned operator and Elasticsearch/Kibana fresh installation, an authenticated document write/read over the Elasticsearch Service, and Kibana API health. It needs roughly 4 GiB of additional free cluster memory. The historical Elastic Helm/Filebeat examples and upgrades of existing data are not covered. Run it with bash scripts/e2e/run.sh eck.

Add coverage

  1. Add scripts/e2e/cases/<target>.sh, using k for all kubectl operations.
  2. Register source paths and any shared dependencies in scripts/e2e/targets.json.
  3. Check rollout readiness and actual application behavior with bounded timeouts.
  4. Run the case locally, and python3 -m unittest discover -s scripts/e2e.

Only registered targets have E2E coverage; this does not claim coverage of every sample in the repository. Helm cases must pass --kube-context "$context" and --kubeconfig "$KUBECONFIG" explicitly.

MySQL tests the Helm/Kustomize shared database dependency with authenticated SQL through its Service: table creation, insert, select and delete. It does not cover the older Flask application or migration of an existing MySQL 5.6 database.

Istio downloads the release pinned in its README and tests the demo profile, Bookinfo sidecar injection, the ingress Service, v1 routing and header-based v2 routing. It also requires tar. Optional Gateway API, dashboards and experimental examples are outside this test. Run bash scripts/e2e/run.sh istio.