CDK on GCP

The Charmed Distribution of Kubernetes® will run seamlessly on Google Cloud Platform(GCP). With the addition of the gcp-integrator, your cluster will also be able to use GCP native features directly.

GCP Credentials

If you have set up a service account with IAM roles as your credential for Juju, there may be some additional authorisations you will need to make to access all features of GCP with CDK.

If you have a GCP project set up specifically for CDK, the quickest route is to simply add the service account as an Owner of that project in the GCP console.

If you chose a more fine-grained approach to role administration, the service account should have at least:

  • roles/compute.loadBalancerAdmin
  • roles/compute.instanceAdmin.v1
  • roles/compute.securityAdmin
  • roles/iam.serviceAccountUser

A full description of the various pre-defined roles is available in the GCP Documentation.

GCP integrator

The gcp-integrator charm simplifies working with CDK on GCP. Using the credentials provided to Juju, it acts as a proxy between CDK and the underlying cloud, granting permissions to dynamically create, for example, storage volumes.

Installing

If you use the recommended install method with conjure-up, the integrator charm will be installed by default, and trust granted automatically.

If you install CDK using the Juju bundle, you can add the gcp-integrator at the same time by using the following overlay file (download it here):

applications:
  gcp-integrator:
    charm: cs:~containers/gcp-integrator
    num_units: 1
relations:
  - ['gcp-integrator', 'kubernetes-master']
  - ['gcp-integrator', 'kubernetes-worker']

To use this overlay with the CDK bundle, it is specified during deploy like this:

juju deploy canonical-kubernetes --overlay ~/path/gcp-overlay.yaml

Then run the command to share credentials with this charm:

juju trust gcp-integrator

… and remember to fetch the configuration file!

juju scp kubernetes-master/0:config ~/.kube/config

For more configuration options and details of the permissions which the integrator uses, please see the charm readme.

Using persistent storage

Many pods you may wish to deploy will require storage. Although you can use any type of storage supported by Kubernetes (see the storage documentation), you also have the option to use the native GCP storage types.

GCP storage currently comes in two types - SSD (pd-ssd) or ‘standard’(pd-standard). To use these, we need to create a storage classes in Kubernetes.

For the standard disks:

kubectl create -f - <<EOY
apiVersion: storage.k8s.io/v1
kind: StorageClass
metadata:
  name: gcp-standard
provisioner: kubernetes.io/gce-pd
parameters:
  type: pd-standard
EOY

Or for SSD:

kubectl create -f - <<EOY
apiVersion: storage.k8s.io/v1
kind: StorageClass
metadata:
  name: gcp-ssd
provisioner: kubernetes.io/gce-pd
parameters:
  type: pd-ssd
EOY

You can confirm this has been added by running:

kubectl get sc

which should return:

NAME           PROVISIONER            AGE
gcp-ssd        kubernetes.io/gce-pd   9s
gcp-standard   kubernetes.io/gce-pd   45s

To actually create storage using this new class, you can make a Persistent Volume Claim:

kubectl create -f - <<EOY
kind: PersistentVolumeClaim
apiVersion: v1
metadata:
  name: testclaim
spec:
  accessModes:
    - ReadWriteOnce
  resources:
    requests:
      storage: 100Mi
  storageClassName: gcp-standard
EOY

This should finish with a confirmation. You can check the current PVCs with:

kubectl get pvc

…which should return something similar to:

NAME        STATUS   VOLUME                                     CAPACITY   ACCESS MODES   STORAGECLASS   AGE
testclaim   Bound    pvc-e1d42bae-44e6-11e9-8dff-42010a840007   1Gi        RWO            gcp-standard   15s

This PVC can then be used by pods operating in the cluster. As an example, the following deploys a busybox pod:

kubectl create -f - <<EOY
apiVersion: v1
kind: Pod
metadata:
  name: busybox
  namespace: default
spec:
  containers:
    - image: busybox
      command:
        - sleep
        - "3600"
      imagePullPolicy: IfNotPresent
      name: busybox
      volumeMounts:
        - mountPath: "/pv"
          name: testvolume
  restartPolicy: Always
  volumes:
    - name: testvolume
      persistentVolumeClaim:
        claimName: testclaim
EOY

To set this type of storage as the default, you can use the command:

kubectl patch storageclass gcp-standard -p '{"metadata": {"annotations":{"storageclass.kubernetes.io/is-default-class":"true"}}}'

Note: If you create persistent disks and subsequently tear down the cluster, check with the GCP console to make sure all the associated resources have also been released.

Using GCP Loadbalancers

With the gcp-integrator charm in place, actions which invoke a loadbalancer in Kubernetes will automatically generate a GCP Target Pool and the relevant forwarding rules. This can be demonstrated with a simple application. Here we will create an application running in five pods:

kubectl run hello-world --replicas=5 --labels="run=load-balancer-example" --image=gcr.io/google-samples/node-hello:1.0  --port=8080

You can verify that the application and replicas have been created with:

 kubectl get deployments hello-world

Which should return output similar to:

 NAME              READY   UP-TO-DATE   AVAILABLE   AGE
 hello-world      5/5               5                            5             2m38s

To create a target pool load balancer, the application should now be exposed as a service:

 kubectl expose deployment hello-world --type=LoadBalancer --name=hello

To check that the service is running correctly:

kubectl describe service hello

…which should return output similar to:

Name:                     hello
Namespace:                default
Labels:                   run=load-balancer-example
Annotations:              <none>
Selector:                 run=load-balancer-example
Type:                     LoadBalancer
IP:                       10.152.183.63
LoadBalancer Ingress:     34.76.144.215
Port:                     <unset>  8080/TCP
TargetPort:               8080/TCP
NodePort:                 <unset>  31864/TCP
Endpoints:                10.1.54.11:8080,10.1.54.12:8080,10.1.54.13:8080 + 2 more...
Session Affinity:         None
External Traffic Policy:  Cluster
Events:
  Type    Reason                Age    From                Message
  ----    ------                ----   ----                -------
  Normal  EnsuringLoadBalancer  9m21s  service-controller  Ensuring load balancer
  Normal  EnsuredLoadBalancer   7m37s  service-controller  Ensured load balancer

You can see that the LoadBalancer Ingress is now associated with a new ingress address in front of the five endpoints of the example deployment. You can test this address:

curl 34.76.144.215:8080
Hello Kubernetes!

Upgrading the integrator-charm

The gcp-integrator is not specifically tied to the version of CDK installed and may generally be upgraded at any time with the following command:

juju upgrade-charm gcp-integrator

Troubleshooting

If you have any specific problems with the gcp-integrator, you can report bugs on Launchpad.

Any activity in GCP can be monitored from the Operations console. If you are using a service account with IAM roles, it is relatively easy to see the actions that particular account is responsible for.

For logs of what the charm itself believes the world to look like, you can use Juju to replay the log history for that specific unit:

juju debug-log --replay --include gcp-integrator/0