Kubernetes/GKE
Google Kubernetes Engine (GKE) is a managed, production-ready environment for deploying containerized applications in Kubernetes.
Contents
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-evictannotation set to False - Pods that have the
RestrictivePodDisruptionBudget - At the node-level, if the
kubernetes.io/scale-down-disabledannotation 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
