Switch branches/tags
Nothing to show
Find file Copy path
Fetching contributors…
Cannot retrieve contributors at this time
210 lines (163 sloc) 8.74 KB


Multicluster-service-account makes it easy to use Kubernetes service accounts as multicluster identities. It imports and automounts remote service account tokens inside pods, for them to call the Kubernetes APIs of other clusters. Multicluster-service-account works well with multicluster-controller, but any cross-cluster Kubernetes client can benefit from it.

Why? Check out Admiralty's blog post introducing multicluster-service-account.

How it Works

Multicluster-service-account consists of:

  1. a ServiceAccountImport custom resource definition (CRD) and controller to import remote service accounts (and their secrets);
  2. a dynamic admission webhook to automount service account import secrets inside annotated pods, the same way regular service accounts are automounted inside pods;
  3. helper methods to generate client-go configurations from service account imports (as well as generic methods to fall back to kubeconfig contexts and regular service accounts).

Getting Started

We assume that you are a cluster admin on two clusters, associated with, e.g., the contexts "cluster1" and "cluster2" in your kubeconfig. We're going to install multicluster-service-account and run a multicluster client example in cluster1, listing pods in cluster2.

CLUSTER1=cluster1 # change me
CLUSTER2=cluster2 # change me

Step 1: Installation

Install multicluster-service-account in cluster1:

kubectl apply -f $MANIFEST_URL --context $CLUSTER1

Cluster1 is now able to import service accounts, but it hasn't been given permission to import them from cluster2 yet. This is a chicken-and-egg problem: cluster1 needs a token from cluster2, before it can import service accounts from it. To solve this problem, download the kubemcsa binary and run the bootstrap command:

OS=linux # or darwin (i.e., OS X) or windows
ARCH=amd64 # if you're on a different platform, you must know how to build from source
curl -Lo kubemcsa $BINARY_URL
chmod +x kubemcsa
sudo mv kubemcsa /usr/local/bin

kubemcsa bootstrap $CLUSTER1 $CLUSTER2

Step 2: Example

The multicluster-client example includes:

  • in cluster2:
    • a service account named pod-lister in the default namespace, bound to a role that can only list pods in its namespace;
    • a dummy NGINX deployment (to have pods to list);
  • in cluster1:
    • a service account import named cluster2-default-pod-lister, importing pod-lister from the default namespace of cluster2;
    • a multicluster-client job, whose pod is annotated to automount cluster2-default-pod-lister's secret—it will list the pods in the default namespace of cluster2, and stop without restarting (we'll check the logs).
kubectl config use-context $CLUSTER2
kubectl create serviceaccount pod-lister
kubectl create role pod-lister --verb=list --resource=pods
kubectl create rolebinding pod-lister --role=pod-lister \
kubectl run nginx --image nginx

kubectl config use-context $CLUSTER1
cat <<EOF | kubectl create -f -
kind: ServiceAccountImport
  name: $CLUSTER2-default-pod-lister
  clusterName: $CLUSTER2
  namespace: default
  name: pod-lister
apiVersion: batch/v1
kind: Job
  name: multicluster-client
      annotations: $CLUSTER2-default-pod-lister
      restartPolicy: Never
      - name: multicluster-client

In cluster1, check that:

  1. The service account import controller created a secret for the cluster2-default-pod-lister service account import, containing the token and namespace of the remote service account, and the URL and root certificate of the remote Kubernetes API:
    kubectl get secret -l$CLUSTER2-default-pod-lister -o yaml
    # the data is base64-encoded
  2. The service account import secret was mounted inside the multicluster-client pod by the service account import admission controller:
    kubectl get pod -l job-name=multicluster-client -o yaml
    # look at volumes and volume mounts
  3. The multicluster-client pod was able to list pods in the default namespace of cluster2:
    kubectl logs job/multicluster-client

Service Account Imports

Service account imports tell the service account import controller to maintain a secret in the same namespace, containing the remote service account's namespace and token, as well as the URL and root certificate of the remote Kubernetes API, which are all necessary data to configure a Kubernetes client. If a pod needs to call several clusters, it will use several service account imports, e.g.:

kind: ServiceAccountImport
  name: cluster2-default-pod-lister
  clusterName: cluster2
  namespace: default
  name: pod-lister
kind: ServiceAccountImport
  name: cluster3-default-pod-lister
  clusterName: cluster3
  namespace: default
  name: pod-lister


The annotation on a pod (or pod template) tells the service account import admission controller to automount the corresponding secrets inside it. If a pod needs several service account imports, separate their names with commas, e.g.:

apiVersion: v1
kind: Pod
  name: multicluster-client
  annotations: cluster2-default-pod-lister,cluster3-default-pod-lister
  # ...

Note: just like with local service accounts, there is a race condition if a service account import and a pod requesting it are created at the same time: the service account import admission controller will likely reject the pod because the secret to automount won't be ready. Luckily, if the pod is controlled by another object, such as a deployment, job, etc., pod creation will be retried.

Client Configuration

Multicluster-service-account includes a Go library (cf. pkg/config) to facilitate the creation of client-go rest.Config instances from service account imports. From there, you can create kubernetes.Clientset instances as usual. The namespaces of the remote service accounts are also provided:

cfg, ns, err := NamedServiceAccountImportConfigAndNamespace("cluster2-default-pod-lister")
// ...
clientset, err := kubernetes.NewForConfig(cfg)
// ...
pods, err := clientset.CoreV1().Pods(ns).List(metav1.ListOptions{})

Usually, however, you don't want to hardcode the name of the mounted service account import. If you only expect one, you can get a Config for it and its remote namespace like this:

cfg, ns, err := ServiceAccountImportConfigAndNamespace()

If several service account imports are mounted, you can get Configs and namespaces for all of them by name as a map[string]ConfigAndNamespaceTuple:

all, err := AllServiceAccountImportConfigsAndNamespaces()
// ...
for name, cfgAndNs := range all {
  cfg := cfgAndNs.Config
  ns := cfgAndNs.Namespace
  // ...

Generic Client Configuration

The true power of multicluster-service-account's config package is in its generic functions, that can fall back to kubeconfig contexts or regular service accounts when no service account import is mounted:

cfg, ns, err := ConfigAndNamespace()
all, err := AllNamedConfigsAndNamespaces()

The service account import controller uses AllNamedConfigsAndNamespaces() internally. That's how we were able to bootstrap in the getting started guide: the same binary was used with a kubeconfig out-of-cluster and with service account imports in-cluster. The generic client example uses ConfigAndNamespace().

API Reference

For more details on the config package, or to better understand how the service account import controller and admission control work, please refer to the API documentation:


go get
godoc -http=:6060

then http://localhost:6060/pkg/