Kubernetes/GKE

From Christoph's Personal Wiki
Revision as of 21:11, 23 August 2019 by Christoph (Talk | contribs) (External links)

Jump to: navigation, search

Google Kubernetes Engine (GKE) is a managed, production-ready environment for deploying containerized applications in Kubernetes.

Deployments

A Deployment's rollout is triggered if and only if the Deployment's Pod template (that is, .spec.template) is changed, for example, if the labels or container images of the template are updated. Other updates, such as scaling the Deployment, do not trigger a rollout.


Trigger a deployment rollout
  • To update the version of nginx in the deployment, execute the following command:
$ kubectl set image deployment.v1.apps/nginx-deployment nginx=nginx:1.9.1 --record
$ kubectl rollout status deployment.v1.apps/nginx-deployment
$ kubectl rollout history deployment nginx-deployment
Trigger a deployment rollback

To roll back an object's rollout, you can use the kubectl rollout undo command.

To roll back to the previous version of the nginx deployment, execute the following command:

$ kubectl rollout undo deployments nginx-deployment
  • View the updated rollout history of the deployment.
$ kubectl rollout history deployment nginx-deployment

deployments "nginx-deployment"
REVISION  CHANGE-CAUSE
2         kubectl set image deployment.v1.apps/nginx-deployment nginx=nginx:1.9.1 --record=true
3         <none>
  • View the details of the latest deployment revision:
$ kubectl rollout history deployment/nginx-deployment --revision=3

The output should look like the example. Your output might not be an exact match but it will show that the current revision has rolled back to nginx:1.7.9.

deployments "nginx-deployment" with revision #3
Pod Template:
  Labels:       app=nginx
        pod-template-hash=3123191453
  Containers:
   nginx:
    Image:      nginx:1.7.9
    Port:       80/TCP
    Host Port:  0/TCP
    Environment:        <none>
    Mounts:     <none>
  Volumes:      <none>

Perform a canary deployment

A canary deployment is a separate deployment used to test a new version of your application. A single service targets both the canary and the normal deployments. And it can direct a subset of users to the canary version to mitigate the risk of new releases. The manifest file nginx-canary.yaml that is provided for you deploys a single pod running a newer version of nginx than your main deployment. In this task, you create a canary deployment using this new deployment file.

apiVersion: apps/v1
kind: Deployment
metadata:
  name: nginx-canary
  labels:
    app: nginx
spec:
  replicas: 1
  selector:
    matchLabels:
      app: nginx
  template:
    metadata:
      labels:
        app: nginx
        track: canary
        Version: 1.9.1
    spec:
      containers:
      - name: nginx
        image: nginx:1.9.1
        ports:
        - containerPort: 80

The manifest for the nginx Service you deployed in the previous task uses a label selector to target the Pods with the app: nginx label. Both the normal deployment and this new canary deployment have the app: nginx label. Inbound connections will be distributed by the service to both the normal and canary deployment Pods. The canary deployment has fewer replicas (Pods) than the normal deployment, and thus it is available to fewer users than the normal deployment.

  • Create the canary deployment based on the configuration file.
$ kubectl apply -f nginx-canary.yaml

When the deployment is complete, verify that both the nginx and the nginx-canary deployments are present.

$ kubectl get deployments

Switch back to the browser tab that is connected to the external LoadBalancer service ip and refresh the page. You should continue to see the standard "Welcome to nginx" page.

Switch back to the Cloud Shell and scale down the primary deployment to 0 replicas.

$ kubectl scale --replicas=0 deployment nginx-deployment

Verify that the only running replica is now the Canary deployment:

$ kubectl get deployments

Switch back to the browser tab that is connected to the external LoadBalancer service ip and refresh the page. You should continue to see the standard "Welcome to nginx" page showing that the Service is automatically balancing traffic to the canary deployment.

Note: Session affinity The Service configuration used in the lab does not ensure that all requests from a single client will always connect to the same Pod. Each request is treated separately and can connect to either the normal nginx deployment or to the nginx-canary deployment. This potential to switch between different versions may cause problems if there are significant changes in functionality in the canary release. To prevent this you can set the sessionAffinity field to ClientIP in the specification of the service if you need a client's first request to determine which Pod will be used for all subsequent connections.

For example:

apiVersion: v1
kind: Service
metadata:
  name: nginx
spec:
  type: LoadBalancer
  sessionAffinity: ClientIP
  selector:
    app: nginx
  ports:
  - protocol: TCP
    port: 60000
    targetPort: 80

Jobs and CronJobs

  • Simple example:
$ kubectl run pi --image perl --restart Never -- perl -Mbignum bpi -wle 'print bpi(2000)'
Parallel Job with fixed completion count
$ cat << EOF > my-app-job.yaml
apiVersion: batch/v1
kind: Job
metadata:
  name: my-app-job
spec:
  completions: 3
  parallelism: 2
  template:
    spec:
[...]
EOF
spec:
  backoffLimit: 4
  activeDeadlineSeconds: 300
Example#1
Create and run a Job

You will create a job using a sample deployment manifest called example-job.yaml that has been provided for you. This Job computes the value of Pi to 2,000 places and then prints the result.

apiVersion: batch/v1
kind: Job
metadata:
  # Unique key of the Job instance
  name: example-job
spec:
  template:
    metadata:
      name: example-job
    spec:
      containers:
      - name: pi
        image: perl
        command: ["perl"]
        args: ["-Mbignum=bpi", "-wle", "print bpi(2000)"]
      # Do not restart containers after they exit
      restartPolicy: Never

To create a Job from this file, execute the following command:

$ kubectl apply -f example-job.yaml
$ kubectl describe job
    Host Port:  <none>
    Command:
      perl
    Args:
      -Mbignum=bpi
      -wle
      print bpi(2000)
    Environment:  <none>
    Mounts:       <none>
  Volumes:        <none>
Events:
  Type    Reason            Age   From            Message
  ----    ------            ----  ----            -------
  Normal  SuccessfulCreate  17s   job-controller  Created pod: example-job-gtf7w

$ kubectl get pods
NAME                READY   STATUS      RESTARTS   AGE
example-job-gtf7w   0/1     Completed   0          43s
Clean up and delete the Job

When a Job completes, the Job stops creating Pods. The Job API object is not removed when it completes, which allows you to view its status. Pods created by the Job are not deleted, but they are terminated. Retention of the Pods allows you to view their logs and to interact with them.

To get a list of the Jobs in the cluster, execute the following command:

$ kubectl get jobs

NAME          DESIRED   SUCCESSFUL   AGE
example-job   1         1            2m

To retrieve the log file from the Pod that ran the Job execute the following command. You must replace [POD-NAME] with the node name you recorded in the last task

$ kubectl logs [POD-NAME]
3.141592653589793238...

The output will show that the job wrote the first two thousand digits of pi to the Pod log.

To delete the Job, execute the following command:

$ kubectl delete job example-job

If you try to query the logs again the command will fail as the Pod can no longer be found.

Define and deploy a CronJob manifest

You can create CronJobs to perform finite, time-related tasks that run once or repeatedly at a time that you specify.

In this section, we will create and run a CronJob, and then clean up and delete the Job.

Create and run a CronJob

The CronJob manifest file example-cronjob.yaml has been provided for you. This CronJob deploys a new container every minute that prints the time, date and "Hello, World!".

apiVersion: batch/v1beta1
kind: CronJob
metadata:
  name: hello
spec:
  schedule: "*/1 * * * *"
  jobTemplate:
    spec:
      template:
        spec:
          containers:
          - name: hello
            image: busybox
            args:
            - /bin/sh
            - -c
            - date; echo "Hello, World!"
          restartPolicy: OnFailure

<block> Note

CronJobs use the required schedule field, which accepts a time in the Unix standard crontab format. All CronJob times are in UTC:

  • The first value indicates the minute (between 0 and 59).
  • The second value indicates the hour (between 0 and 23).
  • The third value indicates the day of the month (between 1 and 31).
  • The fourth value indicates the month (between 1 and 12).
  • The fifth value indicates the day of the week (between 0 and 6).

The schedule field also accepts * and ? as wildcard values. Combining / with ranges specifies that the task should repeat at a regular interval. In the example, */1 * * * * indicates that the task should repeat every minute of every day of every month. </block>

To create a Job from this file, execute the following command:

$ kubectl apply -f example-cronjob.yaml
<pre>

To check the status of this Job, execute the following command, where [job_name] is the name of your job:
<pre>
$ kubectl describe job [job_name]

    Image:      busybox
    Port:       <none>
    Host Port:  <none>
    Args:
      /bin/sh
      -c
      date; echo "Hello, World!"
    Environment:  <none>
    Mounts:       <none>
  Volumes:        <none>
Events:
  Type    Reason            Age   From            Message
  ----    ------            ----  ----            -------
  Normal  SuccessfulCreate  35s   job-controller  Created pod: hello-1565824980-sgdnn

View the output of the Job by querying the logs for the Pod. Replace [POD-NAME] with the name of the Pod you recorded in the last step.

$ kubectl logs <pod-name>

Wed Aug 14 23:23:03 UTC 2019
Hello, World!

To view all job resources in your cluster, including all of the Pods created by the CronJob which have completed, execute the following command:

$ kubectl get jobs

NAME               COMPLETIONS   DURATION   AGE
hello-1565824980   1/1           2s         2m29s
hello-1565825040   1/1           2s         89s
hello-1565825100   1/1           2s         29s

Your job names might be different from the example output. By default, Kubernetes sets the Job history limits so that only the last three successful and last failed job are retained so this list will only contain the most recent three of four jobs.

Clean up and delete the Job

In order to stop the CronJob and clean up the Jobs associated with it you must delete the CronJob.

To delete all these jobs, execute the following command:

$ kubectl delete cronjob hello

To verify that the jobs were deleted, execute the following command:

$ kubectl get jobs
No resources found.

All the Jobs were removed.


Cluster scaling

Think of cluster scaling as a coarse-grain operation that should happen infrequently in pods scaling with deployments as a fine-grain operation that should happen frequently.

Pod conditions that prevent node deletion
  • Not run by a controller
    • e.g., Pods that are not set in a Deployment, ReplicaSet, Job, etc.
  • Has local storage
  • Restricted by constraint rules
  • Pods that have cluster-autoscaler.kubernetes.io/safe-to-evict annotation set to False
  • Pods that have the RestrictivePodDisruptionBudget
  • At the node-level, if the kubernetes.io/scale-down-disabled annotation is set to True
gcloud
  • Create a cluster with autoscaling enabled:
$ gcloud container clusters create <cluster-name> \
  --num-nodes 30 \
  --enable-autoscaling \
  --min-nodes 15 \
  --max-nodes 50 \
  [--zone <compute-zone>]
  • Add a node pool with autoscaling enabled:
$ gcloud container node-pools create <pool-name> \
  --cluster <cluster-name> \
  --enable-autoscaling \
  --min-nodes 15 \
  --max-nodes 50 \
  [--zone <compute-zone>]
  • Enable autoscaling for an existing node pool:
$ gcloud container clusters update \
  <cluster-name> \
  --enable-autoscaling \
  --min-nodes 1 \
  --max-nodes 10 \
  --zone <compute-zone> \
  --node-pool <pool-name>
  • Disable autoscaling for an existing node pool:
$ gcloud container clusters update \
  <cluster-name> \
  --no-enable-autoscaling \
  --node-pool <pool-name> \
  [--zone <compute-zone> --project <project-id>]

Configuring Pod Autoscaling and NodePools

Create a GKE cluster

In Cloud Shell, type the following command to create environment variables for the GCP zone and cluster name that will be used to create the cluster for this lab.

export my_zone=us-central1-a
export my_cluster=standard-cluster-1
  • Configure tab completion for the kubectl command-line tool.
source <(kubectl completion bash)
  • Create a VPC-native Kubernetes cluster:
$ gcloud container clusters create $my_cluster \
   --num-nodes 2 --enable-ip-alias --zone $my_zone
  • Configure access to your cluster for kubectl:
$ gcloud container clusters get-credentials $my_cluster --zone $my_zone
Deploy a sample web application to your GKE cluster

Deploy a sample application to your cluster using the web.yaml deployment file that has been created for you:

apiVersion: extensions/v1beta1
kind: Deployment
metadata:
  name: web
spec:
  replicas: 1
  selector:
    matchLabels:
      run: web
  template:
    metadata:
      labels:
        run: web
    spec:
      containers:
      - image: gcr.io/google-samples/hello-app:1.0
        name: web
        ports:
        - containerPort: 8080
          protocol: TCP

This manifest creates a deployment using a sample web application container image that listens on an HTTP server on port 8080.

  • To create a deployment from this file, execute the following command:
$ kubectl create -f web.yaml --save-config

  • Create a service resource of type NodePort on port 8080 for the web deployment:
$ kubectl expose deployment web --target-port=8080 --type=NodePort

  • Verify that the service was created and that a node port was allocated:
$ kubectl get service web
NAME   TYPE       CLUSTER-IP    EXTERNAL-IP   PORT(S)          AGE
web    NodePort   10.12.6.154   <none>        8080:30972/TCP   5m4s

Your IP address and port number might be different from the example output.

Configure autoscaling on the cluster

In this section, we will configure the cluster to automatically scale the sample application that we deployed earlier.

Configure autoscaling
  • Get the list of deployments to determine whether your sample web application is still running:
$ kubectl get deployment
NAME   DESIRED   CURRENT   UP-TO-DATE   AVAILABLE   AGE
web    1         1         1            1           94s
  • To configure your sample application for autoscaling (and to set the maximum number of replicas to four and the minimum to one, with a CPU utilization target of 1%), execute the following command:
$ kubectl autoscale deployment web --max 4 --min 1 --cpu-percent 1

When you use kubectl autoscale, you specify a maximum and minimum number of replicas for your application, as well as a CPU utilization target.

  • Get the list of deployments to verify that there is still only one deployment of the web application:
$ kubectl get deployment
Inspect the HorizontalPodAutoscaler object

The kubectl autoscale command you used in the previous task creates a HorizontalPodAutoscaler object that targets a specified resource, called the scale target, and scales it as needed. The autoscaler periodically adjusts the number of replicas of the scale target to match the average CPU utilization that you specify when creating the autoscaler.

  • To get the list of HorizontalPodAutoscaler resources, execute the following command:
$ kubectl get hpa
NAME   REFERENCE        TARGETS   MINPODS   MAXPODS   REPLICAS   AGE
web    Deployment/web   1%/1%     1         4         1          50s
  • To inspect the configuration of HorizontalPodAutoscaler in YAML form, execute the following command:
$ kubectl describe horizontalpodautoscaler web
<pre>
Name:                                                  web
Namespace:                                             default
Labels:                                                <none>
Annotations:                                           <none>
CreationTimestamp:                                     Thu, 15 Aug 2019 12:32:37 -0700
Reference:                                             Deployment/web
Metrics:                                               ( current / target )
  resource cpu on pods  (as a percentage of request):  1% (1m) / 1%
Min replicas:                                          1
Max replicas:                                          4
Deployment pods:                                       1 current / 1 desired
Conditions:
  Type            Status  Reason              Message
  ----            ------  ------              -------
  AbleToScale     True    ReadyForNewScale    recommended size matches current size
  ScalingActive   True    ValidMetricFound    the HPA was able to successfully calculate a replica count from cpu resource utilization (percentage of request)
  ScalingLimited  False   DesiredWithinRange  the desired count is within the acceptable range
Events:           <none>
Test the autoscale configuration

You need to create a heavy load on the web application to force it to scale out. You create a configuration file that defines a deployment of four containers that run an infinite loop of HTTP queries against the sample application web server.

You create the load on your web application by deploying the loadgen application using the loadgen.yaml file that has been provided for you.

apiVersion: apps/v1
kind: Deployment
metadata:
  name: loadgen
spec:
  replicas: 4
  selector:
    matchLabels:
      app: loadgen
  template:
    metadata:
      labels:
        app: loadgen
    spec:
      containers:
      - name: loadgen
        image: k8s.gcr.io/busybox
        args:
        - /bin/sh
        - -c
        - while true; do wget -q -O- http://web:8080; done
  • Get the list of deployments to verify that the load generator is running:
$ kubectl get deployment
NAME      DESIRED   CURRENT   UP-TO-DATE   AVAILABLE   AGE
loadgen   4         4         4            4           11s
web       1         1         1            1           9m9s
  • Inspect HorizontalPodAutoscaler:
$ kubectl get hpa
NAME   REFERENCE        TARGETS   MINPODS   MAXPODS   REPLICAS   AGE
web    Deployment/web   20%/1%    1         4         1          7m58s

Once the loadgen Pod starts to generate traffic, the web deployment CPU utilization begins to increase. In the example output, the targets are now at 35% CPU utilization compared to the 1% CPU threshold.

  • After a few minutes, inspect the HorizontalPodAutoscaler again:
$ kubectl get hpa
NAME   REFERENCE        TARGETS   MINPODS   MAXPODS   REPLICAS   AGE
web    Deployment/web   68%/1%    1         4         4          9m39s

$ kubectl get deployment
NAME      DESIRED   CURRENT   UP-TO-DATE   AVAILABLE   AGE
loadgen   4         4         4            4           2m44s
web       4         4         4            3           11m
  • To stop the load on the web application, scale the loadgen deployment to zero replicas.
$ kubectl scale deployment loadgen --replicas 0
  • Get the list of deployments to verify that loadgen has scaled down.
$ kubectl get deployment
NAME      DESIRED   CURRENT   UP-TO-DATE   AVAILABLE   AGE
loadgen   0         0         0            0           3m25s
web       4         4         4            3           12m

The loadgen deployment should have zero replicas.

Wait 2 to 3 minutes, and then get the list of deployments again to verify that the web application has scaled down to the minimum value of 1 replica that you configured when you deployed the autoscaler.

$ kubectl get deployment
NAME      DESIRED   CURRENT   UP-TO-DATE   AVAILABLE   AGE
loadgen   0         0         0            0           4m
web       1         1         1            1           15m

You should now have one deployment of the web application.

Managing node pools

In this section, we will create a new pool of nodes using preemptible instances, and then will constrain the web deployment to run only on the preemptible nodes.

Add a node pool
  • To deploy a new node pool with three preemptible VM instances, execute the following command:
$ gcloud container node-pools create "temp-pool-1" \
  --cluster=$my_cluster --zone=$my_zone \
  --num-nodes "2" --node-labels=temp=true --preemptible

If you receive an error that no preemptible instances are available you can remove the --preemptible option to proceed with the lab.

  • Get the list of nodes to verify that the new nodes are ready:
$ kubectl get nodes
NAME                                                STATUS   ROLES    AGE   VERSION
gke-standard-cluster-1-default-pool-61fba731-01mc   Ready    <none>   21m   v1.12.8-gke.10
gke-standard-cluster-1-default-pool-61fba731-bvfx   Ready    <none>   21m   v1.12.8-gke.10
gke-standard-cluster-1-temp-pool-1-e8966c96-nccc    Ready    <none>   46s   v1.12.8-gke.10
gke-standard-cluster-1-temp-pool-1-e8966c96-pk21    Ready    <none>   43s   v1.12.8-gke.10

You should now have 4 nodes. (Your names will be different from the example output.)

All the nodes that you added have the temp=true label because you set that label when you created the node-pool. This label makes it easier to locate and configure these nodes.

  • To list only the nodes with the temp=true label, execute the following command:
$ kubectl get nodes -l temp=true
NAME                                               STATUS   ROLES    AGE    VERSION
gke-standard-cluster-1-temp-pool-1-e8966c96-nccc   Ready    <none>   2m1s   v1.12.8-gke.10
gke-standard-cluster-1-temp-pool-1-e8966c96-pk21   Ready    <none>   118s   v1.12.8-gke.10
Control scheduling with taints and tolerations

To prevent the scheduler from running a Pod on the temporary nodes, you add a taint to each of the nodes in the temp pool. Taints are implemented as a key-value pair with an effect (such as NoExecute) that determines whether Pods can run on a certain node. Only nodes that are configured to tolerate the key-value of the taint are scheduled to run on these nodes.

To add a taint to each of the newly created nodes, execute the following command. You can use the temp=true label to apply this change across all the new nodes simultaneously.

$ kubectl taint node -l temp=true nodetype=preemptible:NoExecute
node/gke-standard-cluster-1-temp-pool-1-e8966c96-nccc tainted
node/gke-standard-cluster-1-temp-pool-1-e8966c96-pk21 tainted

$ kubectl describe nodes | grep ^Taints
Taints:             <none>
Taints:             <none>
Taints:             nodetype=preemptible:NoExecute
Taints:             nodetype=preemptible:NoExecute

To allow application Pods to execute on these tainted nodes, you must add a tolerations key to the deployment configuration.

Edit the web.yaml file to add the following key in the template's spec section:

tolerations:
- key: "nodetype"
  operator: Equal
  value: "preemptible"

The spec section of the file should look like the following:

...
    spec:
      tolerations:
      - key: "nodetype"
        operator: Equal
        value: "preemptible"
      containers:
      - image: gcr.io/google-samples/hello-app:1.0
        name: web
        ports:
        - containerPort: 8080
          protocol: TCP

To force the web deployment to use the new node-pool add a nodeSelector key in the template's spec section. This is parallel to the tolerations key you just added.

     nodeSelector:
        temp: "true"

Note: GKE adds a custom label to each node called cloud.google.com/gke-nodepool that contains the name of the node-pool that the node belongs to. This key can also be used as part of a nodeSelector to ensure Pods are only deployed to suitable nodes.

The full web.yaml deployment should now look as follows.

apiVersion: extensions/v1beta1
kind: Deployment
metadata:
  name: web
spec:
  replicas: 1
  selector:
    matchLabels:
      run: web
  template:
    metadata:
      labels:
        run: web
    spec:
      tolerations:
      - key: "nodetype"
        operator: Equal
        value: "preemptible"
      nodeSelector:
        temp: "true"
      containers:
      - image: gcr.io/google-samples/hello-app:1.0
        name: web
        ports:
        - containerPort: 8080
          protocol: TCP

To apply this change, execute the following command:

kubectl apply -f web.yaml

If you have problems editing this file successfully you can use the pre-prepared sample file called web-tolerations.yaml instead.

  • Get the list of Pods:
$ kubectl get pods
NAME                   READY     STATUS    RESTARTS   AGE
web-7cb566bccd-pkfst   1/1       Running   0          1m

To confirm the change, inspect the running web Pod(s) using the following command

$ kubectl describe pods -l run=web

A Tolerations section with nodetype=preemptible in the list should appear near the bottom of the (truncated) output.

...
Node-Selectors:  <none>
Tolerations:     node.kubernetes.io/not-ready:NoExecute for 300s
                 node.kubernetes.io/unreachable:NoExecute for 300s
                 nodetype=preemptible
Events:
...

The output confirms that the Pods will tolerate the taint value on the new preemptible nodes, and thus that they can be scheduled to execute on those nodes.

To force the web application to scale out again scale the loadgen deployment back to four replicas.

$ kubectl scale deployment loadgen --replicas 4

You could scale just the web application directly but using the loadgen app will allow you to see how the different taint, toleration and nodeSelector settings that apply to the web and loadgen applications affect which nodes they are scheduled on.

Get the list of Pods using thewide output format to show the nodes running the Pods

$ kubectl get pods -o wide

This shows that the loadgen app is running only on default-pool nodes while the web app is running only the preemptible nodes in temp-pool-1.

The taint setting prevents Pods from running on the preemptible nodes so the loadgen application only runs on the default pool. The toleration setting allows the web application to run on the preemptible nodes and the nodeSelector forces the web application Pods to run on those nodes.

NAME        READY STATUS    [...]         NODE
Loadgen-x0  1/1   Running   [...]         gke-xx-default-pool-y0
loadgen-x1  1/1   Running   [...]         gke-xx-default-pool-y2
loadgen-x3  1/1   Running   [...]         gke-xx-default-pool-y3
loadgen-x4  1/1   Running   [...]         gke-xx-default-pool-y4
web-x1      1/1   Running   [...]         gke-xx-temp-pool-1-z1
web-x2      1/1   Running   [...]         gke-xx-temp-pool-1-z2
web-x3      1/1   Running   [...]         gke-xx-temp-pool-1-z3
web-x4      1/1   Running   [...]         gke-xx-temp-pool-1-z4

Deploying Kubernetes Engine via Helm Charts

Ensure your user account has the cluster-admin role in your cluster.

$ kubectl create clusterrolebinding user-admin-binding \
   --clusterrole=cluster-admin \
   --user=$(gcloud config get-value account)
  • Create a Kubernetes service account that is Tiller - the server side of Helm, can be used for deploying charts.
$ kubectl create serviceaccount tiller --namespace kube-system
  • Grant the Tiller service account the cluster-admin role in your cluster:
$ kubectl create clusterrolebinding tiller-admin-binding \
   --clusterrole=cluster-admin \
   --serviceaccount=kube-system:tiller
  • Execute the following commands to initialize Helm using the service account:
$ helm init --service-account=tiller

$ kubectl -n kube-system get pods | grep ^tiller
tiller-deploy-8548d8bd7c-l548r                                 1/1     Running   0          18s

$ helm repo update

$ helm version
Client: &version.Version{SemVer:"v2.6.2", GitCommit:"be3ae4ea91b2960be98c07e8f73754e67e87963c", GitTreeState:"clean"}
Server: &version.Version{SemVer:"v2.6.2", GitCommit:"be3ae4ea91b2960be98c07e8f73754e67e87963c", GitTreeState:"clean"}

Execute the following command to deploy a set of resources to create a Redis service on the active context cluster:

$ helm install stable/redis

A Helm chart is a package of resource configuration files, along with configurable parameters. This single command deployed a collection of resources.

A Kubernetes Service defines a set of Pods and a stable endpoint by which network traffic can access them. In Cloud Shell, execute the following command to view Services that were deployed through the Helm chart:

$ kubectl get services
NAME                               TYPE        CLUSTER-IP     EXTERNAL-IP   PORT(S)    AGE
kubernetes                         ClusterIP   10.12.0.1      <none>        443/TCP    3m24s
opining-wolverine-redis-headless   ClusterIP   None           <none>        6379/TCP   11s
opining-wolverine-redis-master     ClusterIP   10.12.5.246    <none>        6379/TCP   11s
opining-wolverine-redis-slave      ClusterIP   10.12.14.196   <none>        6379/TCP   11s

A Kubernetes StatefulSet manages the deployment and scaling of a set of Pods, and provides guarantees about the ordering and uniqueness of these Pods. In Cloud Shell, execute the following commands to view a StatefulSet that was deployed through the Helm chart:

$ kubectl get statefulsets
NAME                             DESIRED   CURRENT   AGE
opining-wolverine-redis-master   1         1         59s
opining-wolverine-redis-slave    2         2         59s

A Kubernetes ConfigMap lets you storage and manage configuration artifacts, so that they are decoupled from container-image content. In Cloud Shell, execute the following commands to view ConfigMaps that were deployed through the Helm chart:

$ kubectl get configmaps
NAME                             DATA   AGE
opining-wolverine-redis          3      95s
opining-wolverine-redis-health   6      95s

A Kubernetes Secret, like a ConfigMap, lets you store and manage configuration artifacts, but it's specially intended for sensitive information such as passwords and authorization keys. In Cloud Shell, execute the following commands to view some of the Secret that was deployed through the Helm chart:

$ kubectl get secrets
NAME                      TYPE     DATA   AGE
opining-wolverine-redis   Opaque   1      2m5s

You can inspect the Helm chart directly using the following command:

$ helm inspect stable/redis

If you want to see the templates that the Helm chart deploys you can use the following command:

$ helm install stable/redis --dry-run --debug
Test Redis functionality

You store and retrieve values in the new Redis deployment running in your Kubernetes Engine cluster.

Execute the following command to store the service ip-address for the Redis cluster in an environment variable:

$ export REDIS_IP=$(kubectl get services -l app=redis -o json | jq -r '.items[].spec | select(.selector.role=="master")' | jq -r '.clusterIP')

Retrieve the Redis password and store it in an environment variable:

$ export REDIS_PW=$(kubectl get secret -l app=redis -o jsonpath="{.items[0].data.redis-password}"  | base64 --decode)
  • Display the Redis cluster address and password:
$ echo Redis Cluster Address : $REDIS_IP
$ echo Redis auth password   : $REDIS_PW
  • Open an interactive shell to a temporary Pod, passing in the cluster address and password as environment variables:
$ kubectl run redis-test --rm --tty -i --restart='Never' \
    --env REDIS_PW=$REDIS_PW \
    --env REDIS_IP=$REDIS_IP \
    --image docker.io/bitnami/redis:4.0.12 -- bash
  • Connect to the Redis cluster:
# redis-cli -h $REDIS_IP -a $REDIS_PW
  • Set a key value:
set mykey this_amazing_value

This will display OK if successful.

  • Retrieve the key value:
get mykey

This will return the value you stored indicating that the Redis cluster can successfully store and retrieve data.

Network security

Network policy

A Pod-level firewall restricting access to other Pods and Services. (Disabled by default in GKE.)

Must be enabled:

  • Requires at least 2 nodes of n1-standard-1 or higher (recommended minimum of 3 nodes)
  • Requires nodes to be recreated
  • Enable network policy for a new cluster:
$ gcloud container clusters create <name> \
  --enable-network-policy
  • Enable a network policy for an existing cluster:
$ gcloud container clusters update <name> \
  --update-addons-NetworkPolicy=ENABLED
$ gcloud container cluster update <name> \
  --enable-network-policy
  • Disabling a network policy:
$ gcloud container clusters create <name> \
  --no-enable-network-policy
Writing a network policy
apiVersion: networking.k8s.io/v1
kind: NetworkPolicy
metadata:
  name: demo-network-policy
  namespace: default
spec:
  podSelector:
    matchLabels:
      role: demo-app
  policyTypes:
  - Ingress
  - Egress
  ingress:
  - from:
    - ipBlock:
      cidr: 172.17.0.0/16
      except:
      - 172.17.1.0/16
    - namespaceSelector:
        matchLabels:
          project: myproject
    - podSelector:
        matchLabels:
          role: frontend
    ports:
    - protocol: TCP
      port: 6379

  egress:
  - to:
    - ipBlock:
        cidr: 10.0.0.0/24
    ports:
    - protocol: TCP
      port: 5978
Network policy defaults
  • Pros:
    • Limits "attack surface" of Pods in your cluster.
  • Cons:
    • A lot of work to manage (use Istio instead)
metadata:
  name: default-deny
spec:
  podSelector: {}
  policyTypes:
  - Ingress
metadata:
  name: default-deny
spec:
  podSelector: {}
  policyTypes:
  - Egress
metadata:
  name: default-deny
spec:
  podSelector: {}
  policyTypes:
  - Ingress
  - Egress
metadata:
  name: allow-all
spec:
  podSelector: {}
  policyTypes:
  - Ingress
  ingress:
  - {}
metadata:
  name: allow-all
spec:
  podSelector: {}
  policyTypes:
  - Egress
  egress:
  - {}

Setup a private GKE cluster

In the Cloud Shell, enter the following command to review the details of your new cluster:

$ gcloud container clusters describe private-cluster --region us-central1-a
  • The following values appear only under the private cluster:
privateEndpoint 
an internal IP address. Nodes use this internal IP address to communicate with the cluster master.
publicEndpoint 
an external IP address. External services and administrators can use the external IP address to communicate with the cluster master.
  • You have several options to lock down your cluster to varying degrees:
    • The whole cluster can have external access.
    • The whole cluster can be private.
    • The nodes can be private while the cluster master is public, and you can limit which external networks are authorized to access the cluster master.

Without public IP addresses, code running on the nodes cannot access the public Internet unless you configure a NAT gateway such as Cloud NAT.

You might use private clusters to provide services such as internal APIs that are meant only to be accessed by resources inside your network. For example, the resources might be private tools that only your company uses. Or they might be backend services accessed by your frontend services, and perhaps only those frontend services are accessed directly by external customers or users. In such cases, private clusters are a good way to reduce the surface area of attack for your application.

Restrict incoming traffic to Pods

First, we will create a GKE cluster to use for the demos below.

Create a GKE cluster
  • In Cloud Shell, type the following command to set the environment variable for the zone and cluster name:
export my_zone=us-central1-a
export my_cluster=standard-cluster-1
  • Configure kubectl tab completion in Cloud Shell:
source <(kubectl completion bash)
  • Create a Kubernetes cluster (note that this command adds the additional flag --enable-network-policy. This flag allows this cluster to use cluster network policies):
$ gcloud container clusters create $my_cluster \
  --num-nodes 2 \
  --enable-ip-alias \
  --zone $my_zone \
  --enable-network-policy
  • Configure access to your cluster for the kubectl command-line tool:
$ gcloud container clusters get-credentials $my_cluster --zone $my_zone

Run a simple web server application with the label app=hello, and expose the web application internally in the cluster:

$ kubectl run hello-web --labels app=hello \
  --image=gcr.io/google-samples/hello-app:1.0 --port 8080 --expose
Restrict incoming traffic to Pods
  • The following NetworkPolicy manifest file defines an ingress policy that allows access to Pods labeled app: hello from Pods labeled app: foo:
$ cat << EOF > hello-allow-from-foo.yaml
kind: NetworkPolicy
apiVersion: networking.k8s.io/v1
metadata:
  name: hello-allow-from-foo
spec:
  policyTypes:
  - Ingress
  podSelector:
    matchLabels:
      app: hello
  ingress:
  - from:
    - podSelector:
        matchLabels:
          app: foo
EOF

$ kubectl apply -f hello-allow-from-foo.yaml

$ kubectl get networkpolicy
NAME                   POD-SELECTOR   AGE
hello-allow-from-foo   app=hello      7s
Validate the ingress policy
  • Run a temporary Pod called test-1 with the label app=foo and get a shell in the Pod:
$ kubectl run test-1 --labels app=foo --image=alpine --restart=Never --rm --stdin --tty

The kubectl switches used here in conjunction with the run command are important to note:

--stdin (alternatively -i
creates an interactive session attached to STDIN on the container.
--tty (alternatively -t
allocates a TTY for each container in the pod.
--rm 
instructs Kubernetes to treat this as a temporary Pod that will be removed as soon as it completes its startup task. As this is an interactive session it will be removed as soon as the user exits the session.
--label (alternatively -l
adds a set of labels to the pod.
--restart 
defines the restart policy for the Pod
  • Make a request to the hello-web:8080 endpoint to verify that the incoming traffic is allowed:
/ # wget -qO- --timeout=2 http://hello-web:8080
Hello, world!
Version: 1.0.0
Hostname: hello-web-75f66f69d-qgzjb
/ #
  • Now, run a different Pod using the same Pod name but using a label, app=other, that does not match the podSelector in the active network policy. This Pod should not have the ability to access the hello-web application:
$ kubectl run test-1 --labels app=other --image=alpine --restart=Never --rm --stdin --tty
  • Make a request to the hello-web:8080 endpoint to verify that the incoming traffic is not allowed:
/ # wget -qO- --timeout=2 http://hello-web:8080
wget: download timed out
/ #

The request times out.

Restrict outgoing traffic from the Pods

You can restrict outgoing (egress) traffic as you do incoming traffic. However, in order to query internal hostnames (such as hello-web) or external hostnames (such as www.example.com), you must allow DNS resolution in your egress network policies. DNS traffic occurs on port 53, using TCP and UDP protocols.

The following NetworkPolicy manifest file defines a policy that permits Pods with the label app: foo to communicate with Pods labeled app: hello on any port number, and allows the Pods labeled app: foo to communicate to any computer on UDP port 53, which is used for DNS resolution. Without the DNS port open, you will not be able to resolve the hostnames:

$ cat << EOF > foo-allow-to-hello.yaml
kind: NetworkPolicy
apiVersion: networking.k8s.io/v1
metadata:
  name: foo-allow-to-hello
spec:
  policyTypes:
  - Egress
  podSelector:
    matchLabels:
      app: foo
  egress:
  - to:
    - podSelector:
        matchLabels:
          app: hello
  - to:
    ports:
    - protocol: UDP
      port: 53
EOF

$ kubectl apply -f foo-allow-to-hello.yaml

$ kubectl get networkpolicy
NAME                   POD-SELECTOR   AGE
foo-allow-to-hello     app=foo        7s
hello-allow-from-foo   app=hello      5m
Validate the egress policy
  • Deploy a new web application called hello-web-2 and expose it internally in the cluster:
$ kubectl run hello-web-2 --labels app=hello-2 \
  --image=gcr.io/google-samples/hello-app:1.0 --port 8080 --expose
  • Run a temporary Pod with the app=foo label and get a shell prompt inside the container:
$ kubectl run test-3 --labels app=foo --image=alpine --restart=Never --rm --stdin --tty
  • Verify that the Pod can establish connections to hello-web:8080:
/ # wget -qO- --timeout=2 http://hello-web:8080
Hello, world!
Version: 1.0.0
Hostname: hello-web-75f66f69d-qgzjb
/ #
  • Verify that the Pod cannot establish connections to hello-web-2:8080
wget -qO- --timeout=2 http://hello-web-2:8080

This fails because none of the Network policies you have defined allow traffic to Pods labelled app: hello-2.

  • Verify that the Pod cannot establish connections to external websites, such as www.example.com:
wget -qO- --timeout=2 http://www.example.com

This fails because the network policies do not allow external http traffic (tcp port 80).

/ # ping 8.8.8.8 -c 3
PING 8.8.8.8 (8.8.8.8): 56 data bytes

--- 8.8.8.8 ping statistics ---
3 packets transmitted, 0 packets received, 100% packet loss

Creating Services and Ingress Resources

Create Pods and services to test DNS resolution
  • Create a service called dns-demo with two sample application Pods called dns-demo-1 and dns-demo-2:
$ cat << EOF > dns-demo.yaml
apiVersion: v1
kind: Service
metadata:
  name: dns-demo
spec:
  selector:
    name: dns-demo
  clusterIP: None
  ports:
  - name: dns-demo
    port: 1234
    targetPort: 1234
---
apiVersion: v1
kind: Pod
metadata:
  name: dns-demo-1
  labels:
    name: dns-demo
spec:
  hostname: dns-demo-1
  subdomain: dns-demo
  containers:
  - name: nginx
    image: nginx
---
apiVersion: v1
kind: Pod
metadata:
  name: dns-demo-2
  labels:
    name: dns-demo
spec:
  hostname: dns-demo-2
  subdomain: dns-demo
  containers:
  - name: nginx
    image: nginx
EOF

$ kubectl apply -f dns-demo.yaml

$ kubectl get pods
NAME         READY   STATUS    RESTARTS   AGE
dns-demo-1   1/1     Running   0          19s
dns-demo-2   1/1     Running   0          19s
Access Pods and services by FQDN
  • Test name resolution for pods and services from the Cloud Shell and from Pods running inside your cluster (note: you can find the IP address for dns-demo-2 by displaying the details of the Pod):
$ kubectl describe pods dns-demo-2

You will see the IP address in the first section of the below the status, before the details of the individual containers:

kubectl describe pods dns-demo-2
Name:               dns-demo-2
Namespace:          default
Priority:           0
PriorityClassName:  <none>
Node:               gke-standard-cluster-1-default-pool-a6c9108e-05m2/10.128.0.5
Start Time:         Mon, 19 Aug 2019 16:58:11 -0700
Labels:             name=dns-demo
Annotations:        [...]
Status:             Running
IP:                 10.8.2.5
Containers:
  nginx:

In the example above, the Pod IP address was 10.8.2.8. You can query just the Pod IP address on its own using the following syntax for the kubectl describe pods command:

$ echo $(kubectl get pod dns-demo-2 --template={{.status.podIP}})
10.8.2.5

The format of the FQDN of a Pod is hostname.subdomain.namespace.svc.cluster.local. The last three pieces (svc.cluster.local) stay constant in any cluster, however, the first three pieces are specific to the Pod that you are trying to access. In this case, the hostname is dns-demo-2, the subdomain is dns-demo, and the namespace is default, because we did not specify a non-default namespace. The FQDN of the dns-demo-2 Pod is therefore dns-demo-2.dns-demo.default.svc.cluster.local.

  • Ping dns-demo-2 from your local machine (or from the Cloud Shell):
$ ping dns-demo-2.dns-demo.default.svc.cluster.local
ping: dns-demo-2.dns-demo.default.svc.cluster.local: Name or service not known

The ping fails because we are not inside the cluster itself.

To get inside the cluster, open an interactive session to Bash running from dns-demo-1.

$ kubectl exec -it dns-demo-1 /bin/bash

Now that we are inside a container in the cluster, our commands run from that context. However, we do not have a tool to ping in this container, so the ping command will not work.

  • Update apt-get and install a ping tool (from within the container):
root@dns-demo-1:/# apt-get update && apt-get install -y iputils-ping
  • Ping dns-demo-2:
root@dns-demo-1:/# ping dns-demo-2.dns-demo.default.svc.cluster.local -c 3
PING dns-demo-2.dns-demo.default.svc.cluster.local (10.8.2.5) 56(84) bytes of data.
64 bytes from dns-demo-2.dns-demo.default.svc.cluster.local (10.8.2.5): icmp_seq=1 ttl=62 time=1.46 ms
64 bytes from dns-demo-2.dns-demo.default.svc.cluster.local (10.8.2.5): icmp_seq=2 ttl=62 time=0.397 ms
64 bytes from dns-demo-2.dns-demo.default.svc.cluster.local (10.8.2.5): icmp_seq=3 ttl=62 time=0.387 ms

--- dns-demo-2.dns-demo.default.svc.cluster.local ping statistics ---
3 packets transmitted, 3 received, 0% packet loss, time 16ms
rtt min/avg/max/mdev = 0.387/0.748/1.461/0.504 ms

This ping should succeed and report that the target has the IP address you found earlier for the dns-demo-2 Pod.

  • Ping the dns-demo service's FQDN, instead of a specific Pod inside the service:
ping dns-demo.default.svc.cluster.local

This ping should also succeed but it will return a response from the FQDN of one of the two demo-dns Pods. This Pod might be either demo-dns-1 or demo-dns-2.

When you deploy applications, your application code runs inside a container in the cluster, and thus your code can access other services by using the FQDNs of those services. This approach is better than using IP addresses or even Pod names because those are more likely to change.

Deploy a sample workload and a ClusterIP service

In this section, we will create a deployment for a set of Pods within the cluster and then expose them using a ClusterIP service.

Deploy a sample web application to your GKE cluster
  • Deploy a sample web application container image that listens on an HTTP server on port 8080:
$ cat << EOF > hello-v1.yaml
apiVersion: extensions/v1beta1
kind: Deployment
metadata:
  name: hello-v1
spec:
  replicas: 3
  selector:
    matchLabels:
      run: hello-v1
  template:
    metadata:
      labels:
        run: hello-v1
        name: hello-v1
    spec:
      containers:
      - image: gcr.io/google-samples/hello-app:1.0
        name: hello-v1
        ports:
        - containerPort: 8080
          protocol: TCP
EOF

$ kubectl create -f hello-v1.yaml

$ kubectl get deployments
NAME       DESIRED   CURRENT   UP-TO-DATE   AVAILABLE   AGE
hello-v1   3         3         3            3           10s
Define service types in the manifest
  • Deploy a Service using a ClusterIP:
$ cat << EOF > hello-svc.yaml
apiVersion: v1
kind: Service
metadata:
  name: hello-svc
spec:
  type: ClusterIP
  selector:
    name: hello-v1
  ports:
  - protocol: TCP
    port: 80
    targetPort: 8080
EOF

$ kubectl apply -f ./hello-svc.yaml

This manifest defines a ClusterIP service and applies it to Pods that correspond to the selector. In this case, the manifest is applied to the hello-v1 Pods that we deployed. This service will automatically be applied to any other deployments with the name: hello-v1 label.

  • Verify that the Service was created and that a Cluster-IP was allocated:
$ kubectl get service hello-svc
NAME        TYPE        CLUSTER-IP    EXTERNAL-IP   PORT(S)   AGE
hello-svc   ClusterIP   10.12.1.159   <none>        80/TCP    29s

No external IP is allocated for this service. Because the Kubernetes Cluster IP addresses are not externally accessible by default, creating this Service does not make your application accessible outside of the cluster.

Test your application
  • Attempt to open an HTTP session to the new Service using the following command:
$ curl hello-svc.default.svc.cluster.local
curl: (6) Could not resolve host: hello-svc.default.svc.cluster.local

The connection should fail because that service is not exposed outside of the cluster.

Now, test the Service from inside the cluster using the interactive shell you have running on the dns-demo-1 Pod. Return to your first Cloud Shell window, which is currently redirecting the STDIN and STDOUT of the dns-demo-1 Pod.

  • Install curl so you can make calls to web services from the command line:
$ apt-get install -y curl
  • Use the following command to test the HTTP connection between the Pods:
$ curl hello-svc.default.svc.cluster.local
Hello, world!
Version: 1.0.0
Hostname: hello-v1-5574c4bff6-72wzc

This connection should succeed and provide a response similar to the output below. Your hostname might be different from the example output.

Convert the service to use NodePort

In this section, we will convert our existing ClusterIP service to a NodePort service and then retest access to the service from inside and outside the cluster.

  • Apply a modified version of our previous hello-svc Service manifest:
$ cat << EOF > hello-nodeport-svc.yaml
apiVersion: v1
kind: Service
metadata:
  name: hello-svc
spec:
  type: NodePort
  selector:
    name: hello-v1
  ports:
  - protocol: TCP
    port: 80
    targetPort: 8080
    nodePort: 30100
EOF

$ kubectl apply -f ./hello-nodeport-svc.yaml

This manifest redefines hello-svc as a NodePort service and assigns the service port 30100 on each node of the cluster for that service.

  • Verify that the service type has changed to NodePort:
$ kubectl get service hello-svc
NAME        TYPE       CLUSTER-IP    EXTERNAL-IP   PORT(S)        AGE
hello-svc   NodePort   10.12.1.159   <none>        80:30100/TCP   5m30s

Note that there is still no external IP allocated for this service.

Test the application
  • Attempt to open an HTTP session to the new service:
$ curl hello-svc.default.svc.cluster.local
curl: (6) Could not resolve host: hello-svc.default.svc.cluster.local

The connection should fail because that service is not exposed outside of the cluster.

Return to your first Cloud Shell window, which is currently redirecting the STDIN and STDOUT of the dns-test Pod.

  • Test the HTTP connection between the Pods:
$ curl hello-svc.default.svc.cluster.local

Hello, world!
Version: 1.0.0
Hostname: hello-v1-5574c4bff6-72wzc
Deploy a new set of Pods and a LoadBalancer service

We will now deploy a new set of Pods running a different version of the application so that we can easily differentiate the two services. We will then expose the new Pods as a LoadBalancer Service and access the service from outside the cluster.

  • Create a new deployment that runs version 2 of the sample "hello" application on port 8080:
$ cat << EOF > hello-v2.yaml
apiVersion: extensions/v1beta1
kind: Deployment
metadata:
  name: hello-v2
spec:
  replicas: 3
  selector:
    matchLabels:
      run: hello-v2
  template:
    metadata:
      labels:
        run: hello-v2
        name: hello-v2
    spec:
      containers:
      - image: gcr.io/google-samples/hello-app:2.0
        name: hello-v2
        ports:
        - containerPort: 8080
          protocol: TCP
EOF

$ kubectl create -f hello-v2.yaml

$ kubectl get deployments
NAME       DESIRED   CURRENT   UP-TO-DATE   AVAILABLE   AGE
hello-v1   3         3         3            3           8m22s
hello-v2   3         3         3            3           6s
Define service types in the manifest
  • Deploy a LoadBalancer Service:
apiVersion: v1
kind: Service
metadata:
  name: hello-lb-svc
spec:
  type: LoadBalancer
  selector:
    name: hello-v2
  ports:
  - protocol: TCP
    port: 80
    targetPort: 8080

This manifest defines a LoadBalancer Service, which deploys a GCP Network Load Balancer to provide external access to the service. This service is only applied to the Pods with the name: hello-v2 selector.

$ kubectl apply -f ./hello-lb-svc.yaml
$ kubectl get services
NAME           TYPE           CLUSTER-IP    EXTERNAL-IP      PORT(S)        AGE
dns-demo       ClusterIP      None          <none>           1234/TCP       18m
hello-lb-svc   LoadBalancer   10.12.3.30    35.193.235.140   80:30980/TCP   95s
hello-svc      NodePort       10.12.1.159   <none>           80:30100/TCP   10m
kubernetes     ClusterIP      10.12.0.1     <none>           443/TCP        21m

$ export LB_EXTERNAL_IP=35.193.235.140

Notice that the new LoadBalancer Service has an external IP. This is implemented using a GCP load balancer and will take a few minutes to create. This external IP address makes the service accessible from outside the cluster. Take note of this External IP address for use below.

Test your application
  • Attempt to open an HTTP session to the new service:
$ curl hello-lb-svc.default.svc.cluster.local
curl: (6) Could not resolve host: hello-lb-svc.default.svc.cluster.local

The connection should fail because that service name is not exposed outside of the cluster. This occurs because the external IP address is not registered with this hostname.

  • Try the connection again using the External IP address associated with the service:
$ curl ${LB_EXTERNAL_IP}
Hello, world!
Version: 2.0.0
Hostname: hello-v2-7db7758bf4-998gf

This time the connection does not fail because the LoadBalancer's external IP address can be reached from outside GCP.

Return to your first Cloud Shell window, which is currently redirecting the STDIN and STDOUT of the dns-demo-1 Pod.

  • Use the following command to test the HTTP connection between the Pods.
root@dns-demo-1:/# curl hello-lb-svc.default.svc.cluster.local
Hello, world!
Version: 2.0.0
Hostname: hello-v2-7db7758bf4-qkb42

The internal DNS name works within the Pod, and you can see that you are accessing the same v2 version of the application as you were from outside of the cluster using the external IP address.

Try the connection again within the Pod using the External IP address associated with the service (replace the IP with the external IP of the service created above):

root@dns-demo-1:/# curl 35.193.235.140
Hello, world!
Version: 2.0.0
Hostname: hello-v2-7db7758bf4-crxzf

The external IP also works from inside Pods running in the cluster and returns a result from the same v2 version of the applications.

Deploy an Ingress resource

We have two services in our cluster for the "hello" application. One service is hosting version 1.0 via a NodePort service, while the other service is hosting version 2.0 via a LoadBalancer service. We will now deploy an Ingress resource that will direct traffic to both services based on the URL entered by the user.

Create an Ingress resource

Ingress is a Kubernetes resource that encapsulates a collection of rules and configuration for routing external HTTP(S) traffic to internal services.

On GKE, Ingress is implemented using Cloud Load Balancing. When you create an Ingress resource in your cluster, GKE creates an HTTP(S) load balancer and configures it to route traffic to your application.

  • Define and deploy an Ingress resource that directs traffic to our web services based on the path entered:
$ cat << EOF > hello-ingress.yaml
apiVersion: extensions/v1beta1
kind: Ingress
metadata:
  name: hello-ingress
  annotations:
    nginx.ingress.kubernetes.io/rewrite-target: /
spec:
  rules:
  - http:
      Paths:
     - path: /v1
        backend:
          serviceName: hello-svc
          servicePort: 80
      - path: /v2
        backend:
          serviceName: hello-lb-svc
          servicePort: 80
EOF

$ kubectl apply -f hello-ingress.yaml

When we deploy this manifest, Kubernetes creates an ingress resource on your cluster. The ingress controller running in your cluster is responsible for creating an HTTP(S) load balancer to route all external HTTP traffic (on port 80) to the web NodePort service and the LoadBalancer service that we exposed.

Test your application
  • Get the external IP address of the load balancer serving our application:
$ kubectl describe ingress hello-ingress

Name:             hello-ingress
Namespace:        default
Address:          35.244.213.159
Default backend:  default-http-backend:80 (10.8.1.6:8080)
Rules:
  Host  Path  Backends
  ----  ----  --------
  *
        /v1   hello-svc:80 (<none>)
        /v2   hello-lb-svc:80 (<none>)
Annotations:
[...]
  ingress.kubernetes.io/backends:              {"k8s-be-30013--59854b80169ba7aa":"HEALTHY","k8s-be-30100--59854b80169ba7aa":"HEALTHY","k8s-be-30980--59854b80169ba7aa":"HEALTHY"}
[...]
Events:
  Type    Reason  Age    From                     Message
  ----    ------  ----   ----                     -------
  Normal  ADD     6m34s  loadbalancer-controller  default/hello-ingress
  Normal  CREATE  5m16s  loadbalancer-controller  ip: 35.244.213.159

You may have to wait for a few minutes for the load balancer to become active, and for the health checks to succeed, before the external address will be displayed. Repeat the command every few minutes to check if the Ingress resource has finished initializing.

Use the External IP address associated with the Ingress resource, and type the following command, substituting [external_IP] with the Ingress resource's external IP address. Be sure to include the /v1 in the URL path:

$ curl 35.244.213.159/v1
Hello, world!
Version: 1.0.0
Hostname: hello-v1-5574c4bff6-mbn5

The v1 URL is configured in hello-ingress.yaml to point to the hello-svc NodePort service that directs traffic to the v1 application Pods.

Note: GKE might take a few minutes to set up forwarding rules until the Global load balancer used for the Ingress resource is ready to serve your application. In the meantime, you might get errors such as HTTP 404 or HTTP 500 until the load balancer configuration is propagated across the globe.

  • Now, test the v2 URL path from Cloud Shell. Use the External IP address associated with the Ingress resource, and type the following command, substituting [external_IP] with the Ingress resource's external IP address. Be sure to include the /v2 in the URL path.
$ curl [external_IP]/v2
Hello, world!
Version: 2.0.0
Hostname: hello-v2-7db7758bf4-998gf
Inspect the changes to your networking resources in the GCP Console

There are two load balancers listed:

  1. One was created for the external IP of the hello-lb-svc service. This typically has a UID style name and is configured to load balance TCP port 80 traffic to the cluster nodes.
  2. The second was created for the Ingress object and is a full HTTP(S) load balancer that includes host and path rules that match the Ingress configuration. This will have hello-ingress in its name.

Click the load balancer with hello-ingress in the name. This will display the summary information about the protocols, ports, paths and backend services of the Ingress load balancer.

The v2 URL is configured in hello-ingress.yaml to point to the hello-lb-svc LoadBalancer service that directs traffic to the v2 application Pods.

Load balancing objects in GKE

Kubernetes object How implemented in GKE Typical usage scenario
Service of type ClusterIP GKE networking Cluster-internal applications and microservices
Service of type LoadBalancer GCP Network Load Balancer (regional) Application front ends
Ingress object, backed by a Service of type NodePort GCP HTTP(S) Load Balancer (global) Application front ends; gives access to advanced features like Cloud Armor, Identity-Aware Proxy (beta)


Persistent Data and Storage

  • Volume types:
    • emptyDir: Ephemeral. Shares Pod's lifecycle.
    • ConfigMap: Object can be referenced in a volume.
    • Secret: Stores sensitive info, such as passwords.
    • downwardAPI: Makes data about Pods available to containers.
Creating a Pod with an NFS Volume
apiVersion: v1
kind: Pod
metadata:
  name: web
spec:
  containers:
  - name: web
    image: nginx
    volumeMounts:
    - mountPath: /mnt/vol
      name: nfs
  volumes:
  - name: nfs
    server: 10.1.2.3
    path: "/"
    readOnly: false
Creating and using a compute engine persistent disk

NOTE: This is the old way of mounting persistent volumes. It is no longer a best practice to do the following. Showing here for completeness.

$ gcloud compute disks create \
  --size=100GB \
  --zone=us-west2-a demo-disk
[...]
spec:
  containers:
  - name: demo-container
    image: gcr.io/hello-app:1.0
    volumeMounts:
    - mountPath: /demo-pod
      name: pd-volume
  volumes:
  - name: pd-volume
    gcePersistentDisk:
      pdName: demo-disk # <- must match gcloud
      fsType: ext4

A better way is to abstract the persistent volume (PV) from the Pod by separating the PV from a Persistent Volume Claim (PVC).

kind: PersistentVolume
apiVersion: v1
metadata:
  name: pd-volume
spec:
  storageClassName: "standard"
  capacity:
    storage: 100G
  accessModes:
  - ReadWriteOnce:
    gcePersistentDisk:
      pdName: demo-disk
      fsType: ext4

Note: PVC StorageClassName must match the PV StorageClassName.

kind: StorageClass
apiVersion: storage.k8s.io/v1
metadata:
  name: standard
provisioner: kubernetes.io/gce-pd
parameters:
  type: pd-standard
  replication-type: none

In GKE, a PVC with not defined storage class will use the above (default) storage class.

  • Example using SSD:
kind: PersistentVolume
[...]
spec:
  storageClassName: "ssd"
---
kind: StorageClass
[...]
metadata:
  name: ssd
parameters:
  type: pd-ssd
Volume Access Modes

Access Modes determine how the Volume will read or write. The types of access modes that are available depend on the volume type.

  • ReadWriteOnce: mounts the volume as read/write to a single node;
  • ReadOnlyMany: mounts a volume as read-only to many nodes; and
  • ReadWriteMany: mounts volumes as read/write to many nodes.

For most applications, persistent disks are mounted as ReadWriteOnce.

Note: GCP persistent disks do not support ReadWriteMany. However, NFS does.

  • Example Persistent Volume Claim (PVC):
kind: PersistentVolumeClaim
apiVersion: v1
metadata:
  name: pd-volume-claim
spec:
  storageClassName: "standard"
  accessModes:
  - ReadWriteOnce:
  resources:
    requests:
      storage: 100G
  • Use the above PVC in a Pod (i.e., mount it):
kind: Pod
apiVersion: v1
metadata:
  name: demo-pod
spec:
  containers:
  - name: demo-container
    image: gcr.io/hello-app:1.0
    volumeMounts:
    - mountPath: /demo-pod
      name: pd-volume
  volumes:
  - name: pd-volume
    PersistentVolumeClaim:
      claimName: pd-volume-claim

The above method abstracts...

  • An alternative option is "Dynamic Provisioning".
  • Retain the volume:
[...]
spec:
  persistentVolumeReclaimPolicy: Retain
Regional persistent disks

Increases availability by replicating data between zones:

kind: StorageClass
apiVersion: storage.k8s.io/v1
metadata:
  name: ssd
provisioner: kubernetes.io/gce-pd
parameters:
  type: pd-ssd
  '''replication-type: regional-pd'''
  zones: us-west1-a, us-west1-b

In the above example, if there was an outage in one of the zones, GKE will automatically failover to the other (still up) zone.

You can also use persistent volumes for other controllers, such as deployments and stateful sets. Remember, a deployment is simply a Pod template that runs and maintains a set of identical pods, commonly known as replicas. You can use these deployments for stateless applications. Deployment replicas can share an existing persistent volume using ReadOnlyMany or ReadWriteMany access mode. ReadWriteMany access mode can only be used for storage types that support it, such as NFS systems.

The ReadWriteOnce access mode is not recommended for Deployments because the replicas need to attach and reattach to persistent volumes dynamically. If a first pod needs to detach itself, the second pod needs to be attached first. However, the second pod cannot attach because the first pod is already attached. This creates a deadlock. So neither pod can make progress. Stateful sets resolve this deadlock. Whenever your application needs to maintain state in persistent volumes, managing it with a stateful set rather than a deployment is the way to go.

External links