Switching over to yourspace as the default since we will be manipulating the Node (mynode) app.
$ kubectl config set-context minikube --namespace=yourspace # or kubens yourspace $ kubectl get pods NAME READY STATUS RESTARTS AGE mynode-68b9b9ffcc-4kkg5 1/1 Running 0 15m mynode-68b9b9ffcc-wspxf 1/1 Running 0 18m
Two pods of mynode based on the example in Step 6 but if you need to get your replicas sorted try
$ kubectl scale deployment/mynode --replicas=2 -n yourspace
and -n yourspace should not be necessary based on the set-context earlier
Poll mynode
#!/bin/bash
while true
do
curl $(minikube ip):$(kubectl get service/mynode -o jsonpath="{.spec.ports[*].nodePort}")
sleep .5;
doneHello from Node.js! 4 on mynode-68b9b9ffcc-jv4fd Hello from Node.js! 2 on mynode-68b9b9ffcc-vq9k5 Hello from Node.js! 5 on mynode-68b9b9ffcc-jv4fd Hello from Node.js! 6 on mynode-68b9b9ffcc-jv4fd
Let’s call this version BLUE (the color does not matter) and we wish to deploy GREEN
Modify hello-http.js to say Bonjour
$ cd hello/nodejs
$ vi hello-http.js
res.end('Bonjour from Node.js! ' + cnt++ + ' on ' + process.env.HOSTNAME + '\n');
esc-w-q
and build a new Docker image
$ docker build -t 9stepsawesome/mynode:v2 . Sending build context to Docker daemon 5.12kB Step 1/7 : FROM node:8 ---> ed145ef978c4 Step 2/7 : MAINTAINER Burr Sutter "[email protected]" ---> Using cache ---> 16e077cca62b Step 3/7 : EXPOSE 8000 ---> Using cache ---> 53d9c47ace0d Step 4/7 : WORKDIR /usr/src/ ---> Using cache ---> 5e74464b9671 Step 5/7 : COPY hello-http.js /usr/src ---> 308423270e08 Step 6/7 : COPY package.json /usr/src ---> d13548c1332b Step 7/7 : CMD ["/bin/bash", "-c", "npm start" ] ---> Running in cbf11794596f Removing intermediate container cbf11794596f ---> 73011d539094 Successfully built 73011d539094 Successfully tagged 9stepsawesome/mynode:v2
and check out the image
$ docker images | grep 9steps 9stepsawesome/mynode v2 73011d539094 6 seconds ago 673MB 9stepsawesome/myboot v2 d0c16bffe5f0 38 minutes ago 638MB 9stepsawesome/myboot v1 f66e4bb1f1cf About an hour ago 638MB 9stepsawesome/mynode v1 26d9e4e9f3b1 2 hours ago 673MB
Run the image to see if you built it correctly
$ docker run -it -p 8000:8000 9stepsawesome/mynode:v2 $ curl $(minikube ip):8000 Bonjour from Node.js! 0 on 001b160eaa82
Now, there is a 2nd deployment yaml for mynodeNew
apiVersion: extensions/v1beta1
kind: Deployment
metadata:
labels:
app: mynodenew
name: mynodenew
spec:
replicas: 1
selector:
matchLabels:
app: mynodenew
template:
metadata:
labels:
app: mynodenew
spec:
containers:
- name: mynodenew
image: 9stepsawesome/mynode:v2
ports:
- containerPort: 8000
$ kubectl create -f kubefiles/mynode-deployment-new.yml
You now have the new pod as well as the old ones
$ kubectl get pods NAME READY STATUS RESTARTS AGE mynode-68b9b9ffcc-jv4fd 1/1 Running 0 23m mynode-68b9b9ffcc-vq9k5 1/1 Running 0 23m mynodenew-5fc946f544-q9ch2 1/1 Running 0 25s
Yet your client/user is still seeing the old one only
$ curl $(minikube ip):$(kubectl get service/mynode -o jsonpath="{.spec.ports[*].nodePort}")
Hello from Node.js! 11 on mynode-68b9b9ffcc-jv4fd
You can tell the new pod carries the new code by with an exec
$ kubectl exec -it mynodenew-5fc946f544-q9ch2 /bin/bash root@mynodenew-5fc946f544-q9ch2:/usr/src# curl localhost:8000 Bonjour from Node.js! 0 on mynodenew-5fc946f544-q9ch2 $ exit
Now update the single Service to point to the new pod and go GREEN
$ kubectl patch svc/mynode -p '{"spec":{"selector":{"app":"mynodenew"}}}'
You have just flipped all users to Bonjour (GREEN) and if you wish to flip back
$ kubectl patch svc/mynode -p '{"spec":{"selector":{"app":"mynode"}}}'
Note: Our deployment yaml did not have a live & ready probe, things worked out OK here because we waited until long after mynodenew was up and running before flipping the service selector.
There are at least two types of deployments that some folks consider "canary deployments" in Kubernetes. The first is simply the rolling update strategy with the health check (liveness probe), if the liveness check fails, it knows to undo the deployment.
Switching back to focusing on myboot and myspace
$ kubectl config set-context minikube --namespace=mypace $ kubectl get pods kubectl get pods NAME READY STATUS RESTARTS AGE myboot-859cbbfb98-4rvl8 1/1 Running 0 55m myboot-859cbbfb98-rwgp5 1/1 Running 0 55m
Make sure myboot has 2 replicas
$ kubectl scale deployment/myboot --replicas=2
and let’s attempt to put some bad code into production
Go into hello/springboot/MyRESTController.java and add a System.exit(1) into the /health logic
@RequestMapping(method = RequestMethod.GET, value = "/health")
public ResponseEntity<String> health() {
System.exit(1);
return ResponseEntity.status(HttpStatus.OK)
.body("I am fine, thank you\n");
}
Obviously this sort of thing would never pass through your robust code reviews and automated QA but let’s assume it does.
Build the code
$ mvn clean package
Build the docker image for v3
$ docker build -t 9stepsawesome/myboot:v3 .
Terminal 1: Start a poller
$ ./poll_myboot.sh
Terminal 2: Watch pods
$ kubectl get pods -w
Terminal 3: Watch events
$ kubectl get events -w
Terminal 4: rollout the v3 update
$ kubectl set image deployment/myboot myboot=9stepsawesome/myboot:v3
and watch the fireworks
$ kubectl get pods -w myboot-5d7fb559dd-qh6fl 0/1 Error 1 11m myboot-859cbbfb98-rwgp5 0/1 Terminating 0 6h myboot-859cbbfb98-rwgp5 0/1 Terminating 0 6h myboot-5d7fb559dd-qh6fl 0/1 CrashLoopBackOff 1 11m myboot-859cbbfb98-rwgp5 0/1 Terminating 0 6h
$ kubectl get events -w
2018-08-02 19:42:19 -0400 EDT 2018-08-02 19:42:16 -0400 EDT 2 myboot-5d7fb559dd-qh6fl.154735c94d1446ce Pod spec.containers{myboot} Warning BackOff kubelet, minikube Back-off restarting failed container
And yet your polling client, stays with the old code & old pod
Aloha from Spring Boot! 133 on myboot-859cbbfb98-4rvl8 Aloha from Spring Boot! 134 on myboot-859cbbfb98-4rvl8
If you watch a while, the CrashLoopBackOff will continue and the restart count will increment.
Now, go fix the MyRESTController and also change from Hello to Aloha
No more System.exit()
@RequestMapping(method = RequestMethod.GET, value = "/health")
public ResponseEntity<String> health() {
return ResponseEntity.status(HttpStatus.OK)
.body("I am fine, thank you\n");
}
And change the greeting response to something you recognize.
Save
$ mvn clean package $ docker build -t 9stepsawesome/myboot:v3 .
and now rollout the change to v3
$ kubectl set image deployment/myboot myboot=9stepsawesome/myboot:v3
Go back to v1
$ kubectl set image deployment/myboot myboot=9stepsawesome/myboot:v1
Next, we will use a 2nd Deployment like we did with Blue/Green.
$ kubectl create -f kubefiles/myboot-deployment-canary.yml
And you can see a new pod being born
$ kubectl get pods
And this is the v3 one
$ kubectl get pods -l app=mybootcanary $ kubectl exec -it mybootcanary-6ddc5d8d48-ptdjv curl localhost:8080/
Now we add a label to both v1 and v3 Deployments PodTemplate, causing new pods to be born
$ kubectl patch deployment/myboot -p '{"spec":{"template":{"metadata":{"labels":{"newstuff":"withCanary"}}}}}'
$ kubectl patch deployment/mybootcanary -p '{"spec":{"template":{"metadata":{"labels":{"newstuff":"withCanary"}}}}}'
Tweak the Service selector for this new label
$ kubectl patch service/myboot -p '{"spec":{"selector":{"newstuff":"withCanary","app": null}}}'
You should see approximately 30% canary mixed in with previous deployment
Hello from Spring Boot! 23 on myboot-d6c8464-ncpn8 Hello from Spring Boot! 22 on myboot-d6c8464-qnxd8 Aloha from Spring Boot! 83 on mybootcanary-74d99754f4-tx6pj Hello from Spring Boot! 24 on myboot-d6c8464-ncpn8
You can then manipulate the percentages via the replicas associated with each deployment 20% Aloha (Canary)
$ kubectl scale deployment/myboot --replicas=4 $ kubectl scale deployment/mybootcanary --replicas=1
The challenge with this model is that you have to have the right pod count to get the right mix. If you want a 1% canary, you need 99 of the non-canary pods.
The concept of the Canary rollout gets a lot smarter and more interesting with Istio. You also get the concept of dark launches which allows you to push a change into the production environment, send traffic to the new pod(s) yet no responses are actual sent back to the end-user/client.