This example demonstrates the usage of Kubernetes to perform a rolling update on a running group of pods. See here to understand why you need a rolling update. Also check rolling update design document for more information.
The files for this example are viewable in our docs repo here.
This example assumes that you have forked the repository and turned up a Kubernetes cluster:
$ cd kubernetes
$ ./cluster/kube-up.sh
You can use bash job control to run this in the background (note that you must use the default port – 8001 – for the following demonstration to work properly).
This can sometimes spew to the output so you could also run it in a different terminal. You have to run kubectl proxy
in the root of the
Kubernetes repository. Otherwise you will get “404 page not found” errors as the paths will not match. You can find more information about kubectl proxy
here.
$ kubectl proxy --www=docs/user-guide/update-demo/local/ &
I0218 15:18:31.623279 67480 proxy.go:36] Starting to serve on localhost:8001
Now visit the the demo website. You won’t see anything much quite yet.
Now we will turn up two replicas of an image. They all serve on internal port 80.
$ kubectl create -f docs/user-guide/update-demo/nautilus-rc.yaml
After pulling the image from the Docker Hub to your worker nodes (which may take a minute or so) you’ll see a couple of squares in the UI detailing the pods that are running along with the image that they are serving up. A cute little nautilus.
Now we will increase the number of replicas from two to four:
$ kubectl scale rc update-demo-nautilus --replicas=4
If you go back to the demo website you should eventually see four boxes, one for each pod.
We will now update the docker image to serve a different image by doing a rolling update to a new Docker image.
$ kubectl rolling-update update-demo-nautilus --update-period=10s -f docs/user-guide/update-demo/kitten-rc.yaml
The rolling-update command in kubectl will do 2 things:
gcr.io/google_containers/update-demo:kitten
)Watch the demo website, it will update one pod every 10 seconds until all of the pods have the new image. Note that the new replication controller definition does not include the replica count, so the current replica count of the old replication controller is preserved. But if the replica count had been specified, the final replica count of the new replication controller will be equal this number.
$ kubectl delete rc update-demo-kitten
This first stops the replication controller by turning the target number of replicas to 0 and then deletes the controller.
To turn down a Kubernetes cluster:
$ ./cluster/kube-down.sh
Kill the proxy running in the background: After you are done running this demo make sure to kill it:
$ jobs
[1]+ Running ./kubectl proxy --www=local/ &
$ kill %1
[1]+ Terminated: 15 ./kubectl proxy --www=local/
If you want to build your own docker images, you can set $DOCKER_HUB_USER
to your Docker user id and run the included shell script. It can take a few minutes to download/upload stuff.
$ export DOCKER_HUB_USER=my-docker-id
$ ./docs/user-guide/update-demo/build-images.sh
To use your custom docker image in the above examples, you will need to change the image name in docs/user-guide/update-demo/nautilus-rc.yaml
and docs/user-guide/update-demo/kitten-rc.yaml
.
Note that the images included here are public domain.