See clabernetes docs for reference.

Try c9s

You can launch a disposable KinD cluster, install the published clabernetes Helm chart, and apply a sample SR Linux plus multitool topology with:

make try-c9s

The target requires Docker and creates a single-node KinD cluster by default. It installs MetalLB and prints access endpoints:

SR Linux SSH:      ssh admin@<load-balancer-ip>
SR Linux gNMI:     <load-balancer-ip>:57400
SR Linux NETCONF:  <load-balancer-ip>:830

If KinD, kubectl, or Helm are not installed, it downloads local copies under build/try-c9s/bin.

Select a published or local source explicitly:

make ls-releases
make try-c9s VERSION=0.6.0
make try-c9s VERSION=local

To install into an existing cluster, see docs/installation.md or run:

C9S_CONTEXT=<your-context> make install

SR Linux management access uses the clabernetes LoadBalancer service directly.

Clean up the sample resources and the KinD cluster with:

make try-c9s-clean

Local e2e

You can run the full e2e suite locally against a disposable KinD cluster using locally built images (no published images, no devspace) with:

make test-e2e CLUSTER=kind

This downloads pinned tools into build/e2e/bin, creates a single-node KinD cluster, builds the manager image, loads it into the cluster, installs the local Helm chart, and runs the e2e/... Go tests. Re-runs are cheap: tools are cached and the cluster is reused.

The CLUSTER variable is required to be either kind or existing (and defaults to kind). To run against an already-ready cluster using the current kubeconfig context, use the existing-cluster mode:

C9S_REGISTRY=<registry-visible-to-the-cluster> make test-e2e CLUSTER=existing

Both modes install the current checkout through make install VERSION=local. For a non-KinD cluster, C9S_REGISTRY must point to a registry the cluster can pull from; the existing C9S_LOCAL_* variables control image tags, rebuilding, and image reuse. The manager is installed as release c9s-e2e in namespace c9s-e2e by default. Override these with E2E_INSTALL_RELEASE and E2E_INSTALL_NAMESPACE.

To run only the tests against an already-installed KinD setup, use make e2e-test. The old make test-e2e-local command remains a compatibility alias for make test-e2e CLUSTER=kind.

CI runs the exact same e2e-* make targets (see .github/workflows/e2e.yaml), so local and CI share all of the setup code.

Remove only the dedicated E2E Helm release and namespace with:

make test-e2e-clean C9S_CONTEXT=<context>

For the disposable KinD cluster, tear it down separately with:

make e2e-clean

Documentation development

The Fumadocs site under docs-site/ renders the repository-owned content in docs/. Start the Vite development server from the repository root:

make serve-docs

The target installs the locked documentation dependencies automatically and binds the server to 0.0.0.0 so it is reachable from the host. Override the bind address with DOCS_HOST when needed.

Edits under docs/ are reflected by the development server. Additional documentation workflows are available from the repository root:

make check-docs    # type-check the app and validate documentation links
make build-docs    # create the static site under docs-site/build/client
make preview-docs  # build and serve the static output locally

Development

Run the manager from your current checkout in an existing Kubernetes cluster:

kubectl config use-context <your-cluster>
make dev

DevSpace builds images, deploys the Helm chart into the c9s-dev namespace, syncs source into the manager pod, and starts the manager with debug logging. Each run forces a chart redeploy and overwrites the global Config CR from development values.

If devspace is not on PATH, the pinned version is downloaded to build/dev/bin.

Full documentation (registry modes, localhost: image URLs, helper scripts, troubleshooting): .develop/README.md.

Registry mode summary

| Command | Use when | | ------- | -------- | | make dev | Default. Remote clusters → project-managed local registry. kind/minikube → push to REGISTRY. | | LOCAL_REGISTRY=0 make dev | Push dev images to DEV_REGISTRY (default GHCR); cluster pulls from there. Requires docker login. |

Default DEV_REGISTRY is ghcr.io/clabernetes/clabernetes.

Common options

# different namespace (defaults to c9s-dev)
DEV_NS=my-c9s-dev make dev
push dev images to GHCR (requires docker login ghcr.io)
LOCAL_REGISTRY=0 make dev
mixed-platform cluster
TARGET_PLATFORM=linux/amd64 make dev
extra DevSpace profiles
make dev DEVSPACE_ARGS="--profile always-pull"

Platform detection: .develop/target-platform.sh (see .develop/README.md).

Verify and clean up

kubectl -n c9s-dev get pods
kubectl -n c9s-dev rollout status deployment/clabernetes-manager

make purge-dev   # removes deployment, clabernetes CRDs, and the dev namespace