I want to show how you can enable Kubernetes in your day to day development workflow. So that you get the feel of production deployment locally from day 1.

I have a flask application which I am working on. The basic directory structure looks like this:

$ ll
total 24
-rw-rw-r--. 1 foo foo  427 Apr 23 16:23 app.py
-rw-rw-r--. 1 foo foo  201 Apr 23 16:55 docker-compose.yml
-rw-rw-r--. 1 foo foo  363 Apr 23 16:21 Dockerfile
-rwxrwxr-x. 1 foo foo   82 Dec  5 19:41 entrypoint.sh
-rw-rw-r--. 1 foo foo 3010 Dec  5 19:41 README.adoc
-rw-rw-r--. 1 foo foo   11 Dec  5 19:41 requirements.txt

You can find all of these files in this github repo.

For having a local cluster I am using minikube. So follow instructions to setup minikube. Once you follow the instructions you will have a vm running a single node kubernetes cluster and a locally available kubectl binary.

Before running this application on the minikube cluster we need configurations that kubernetes understands. Since we already have docker-compose file we will generate configs from this file with the help from tool called kompose. Install kompose as per instructions as given on docs.

Generating configs:

$ mkdir configs
$ kompose convert -o configs/
WARN Kubernetes provider doesnt support build key - ignoring
INFO file "configs/hitcounter-service.yaml" created
INFO file "configs/redis-service.yaml" created
INFO file "configs/hitcounter-deployment.yaml" created
INFO file "configs/redis-deployment.yaml" created

Before we deploy the app we need to make some changes in the deployment files, that were converted from docker-compose service having build construct in them. In our case only python app hitcounter is built is being built from Dockerfile.

We will edit file hitcounter-deployment.yaml in configs directory, to not pull image but read image from the local docker storage. Add a field after image called imagePullPolicy: IfNotPresent. Make changes as shown in following diff:

$ git diff
diff --git a/configs/hitcounter-deployment.yaml b/configs/hitcounter-deployment.yaml
index 7b1116d..0ef35b3 100644
--- a/configs/hitcounter-deployment.yaml
+++ b/configs/hitcounter-deployment.yaml
@@ -17,6 +17,7 @@ spec:
         - name: REDIS_HOST
           value: redis
         image: hitcounter
+        imagePullPolicy: IfNotPresent
         name: hitcounter
         ports:
         - containerPort: 5000

Now we are ready with the configs, but we need to build container image for our app. So here you will need to have docker-compose installed on your machine. For that follow docs here.

Build image in the minikube

eval $(minikube docker-env)
docker-compose build

Once the build is complete, we are good to the deployment in kubernetes.

$ kubectl create -f configs/
deployment "hitcounter" created
service "hitcounter" created
deployment "redis" created
service "redis" created

To verify that the app is running, find out the exposed IP Address as follows:

$ kubectl get svc
NAME         CLUSTER-IP   EXTERNAL-IP   PORT(S)          AGE
hitcounter   10.0.0.244   <pending>     5000:30476/TCP   6s
kubernetes   10.0.0.1     <none>        443/TCP          3d
redis        10.0.0.21    <none>        6379/TCP         6s

Now hit the externally exposed port 30476 of service hitcounter as:

$ curl $(minikube ip):30476

Now everytime you make changes to code do the following:

docker-compose build
kubectl scale deployment hitcounter --replicas=0
kubectl scale deployment hitcounter --replicas=1

Above we are removing all containers with old image and asking it to use the new image. For OpenShift we can do oc deploy hitcounter --latest and it will trigger the deployment but I could not find anything similar with kubernetes.

FAQ

  • - Why do I need to make changes in the kompose generated configs?

    Because by the default the config that kompose generates will not set imagePullPolicy and hence Kubernetes assumes its value to be Always. So if you don’t make changes and try to deploy then Kubernetes will try to find the image from docker hub. Which it won’t find and then that deployment will fail. So we need to tell Kubernetes to look for the image in local docker storage.

  • - Can I use the same configs in the production servers as well?

    Yes you can use it just remove the change we did in the imagePullPolicy: IfNotPresent. The change is done to enable you to use the locally built images without having to push the image to any container registry.

  • - How do I get images when I am deploying in production level cluster?

    Make sure your cluster can pull images from some private container registry. And then set up a build pipeline from your code repo to build container on every change of it’s stable branch.

  • - I have build defined in my docker-compose service why do I need to mention image name?

    With docker-compose this is okay. But kompose cannot make up a name on it’s own and create deployment. The issue is tracked in kompose. But for now with build also provide the image name you would expect.

  • - I get error running docker commands with minikube?

    If you face problem accessing the docker daemon running inside the minikube VM like one of this

    $ eval $(minikube docker-env)
    $ docker ps
    could not read CA certificate "/etc/docker/ca.pem": open /etc/docker/ca.pem: no such file or directory
    

    This could be because there is a mismatch in docker client and docker daemon version, so to solve this issue just copy the docker client from the minikube VM to the local machine.

    Enter in the VM

    minikube ssh
    

    Copy the binary to host machine

    scp $(which docker) foo@192.168.122.1:/home/foo/
    

    Now put the binary in PATH.

If you have any other questions please ask it, I would like to add those here in FAQ section.