I recently got the chance to speak at DevoxxFR. I loved the event, and even more the chat I had in the hallway track. I went there to talk about the different ways of managing stateful applications via Kubernetes, this talk raised a lot of questions which inspired me to write this blog post.

Most of this questions were about the persistent volume provisioning, so let’s dive into this subject.

Basically there are 2 ways of managing it:

Manually creating the persistent volume, which is leading to you creating all the persistent volume on your own (the last thing you want to do)
Automatically creating the persistent volume by letting Kubernetes talk with your storage backend so that it can provision itself what it needs

If you do not know what is a persistent volume, that’s ok . Let say that you can see it as a network disk or a network drive, whatever rings a bell in your mind.

Some would argue that there is also the statefulset object and I would answer that from a state perspective the statefulset is using the automatic creation of the persistent volume, nothing more. Hoever statefulset are providing a number of feature usually useful when dealing with stateful applications. The most useful one, in my opinion, being the ordered deployment and deletion of pods.

Let set some context

To make it easier to understand, let put some context here. Let’s say that we are in a company called BigCo. BigCo has already rolled out Kubernetes to manage all services it provides. John and Bob are 2 employees of BigCo. John is a sysadmin and Bob is a developer. One day Bob come to John with an ask, he would need to run a MySQL database in the Kubernetes cluster hosting his application.

Manually creating the persistent volumes

After some research John is ready to share a small POC with Bob. John just want to validate the concept, therefore he choose to create the persistent manually himself and don’t get into the automatic persistent volume creation.

The flow

So here is how the provisioning process work in this situation:

1.BigCo has a Kubernetes cluster and a storage backend up and running. They can both communicate together:

PV+PVC-18 2.John create a persistent volume:

PV+PVC-19 3.Bob create a persistent volume claim referencing the persistent volume previously created:

PV+PVC-20 4.Bob create a pod and reference the persistent volume claim in this pod:

PV+PVC-21 5.Kubernetes mount the persistent volume in the pod:

PV+PVC-22

The code

Tested on Minikube, if you are not running Minikube, there will some minor change to do, mainly in the persistent volume description.

1.Deploy a namespace (that’s always the first thing to do):

apiVersion: v1
kind: Namespace
metadata:
  name: mysql

2.Deploy a persistent volume:

apiVersion: v1
kind: PersistentVolume
metadata:
  name: mysql-pv
spec:
  storageClassName: manual
  accessModes:
    - ReadWriteOnce
  capacity:
    storage: 10Gi
  hostPath:
    path: /data/mysql-pv/

3.Deploy a persistent volume claim:

apiVersion: v1
kind: PersistentVolumeClaim
metadata:
  name: mysql-pvc
  namespace: mysql
spec:
  storageClassName: manual
  accessModes:
    - ReadWriteOnce
  resources:
    requests:
      storage: 3Gi

4.Deploy a MySQL pod. Here we will be deploying it using a deployment object:

apiVersion: apps/v1
kind: Deployment
metadata:
  name: mysql
  namespace: mysql
spec:
  selector:
    matchLabels:
      app: mysql
  strategy:
    type: Recreate
  template:
    metadata:
      labels:
        app: mysql
    spec:
      containers:
      - image: mysql:5.7
        name: mysql
        env:
        - name: MYSQL_ROOT_PASSWORD
          value: password
        ports:
        - containerPort: 3306
          name: mysql
        volumeMounts:
        - name: mysql-persistent-storage
          mountPath: /var/lib/mysql
        resources:
          requests:
            cpu: 500m
            memory: 1Gi
        livenessProbe:
          exec:
            command: ["mysqladmin", "ping"]
          initialDelaySeconds: 30
          periodSeconds: 10
          timeoutSeconds: 5
        readinessProbe:
          exec:
            command: ["mysqladmin", "ping"]
          initialDelaySeconds: 5
          periodSeconds: 2
          timeoutSeconds: 1
      volumes:
      - name: mysql-persistent-storage
        persistentVolumeClaim:
          claimName: mysql-pvc

5.Expose internally to the cluster and via DNS the MySQL container via a service:

apiVersion: v1
kind: Service
metadata:
  name: mysql
  namespace: mysql
spec:
  ports:
  - port: 3306
  selector:
    app: mysql
  clusterIP: None

6.Test the MySQL database by connecting to it from inside the Kubernetes cluster using this command:

kubectl --namespace=mysql run -it --rm --image=mysql:5.7 --restart=Never mysql-client -- mysql -h mysql -ppassword

The result

Now if the pod dies and is rescheduled somewhere else, the data will not be lost.

Limitation

John had to create a persistent volume before Bob could do his job, in BigCo, that is usually meaning the Bob had to submit a ticket to John. Which is usually source of slowdown in the process, which we want to avoid.

Automatically creating the persistent volumes

As Bob is happy with the solution proposed by John, he wants spin up more instances. John quickly need to industrialize the process and after some additional research, he put in place the automatique creation of the persistent volume.

The flow

So here is how the provisioning process work in this situation:

1.Again, BigCo has a Kubernetes cluster and a storage backend up and running. They can both communicate together:

DP-24 2.John create one or many storage class, you can see storage class as type of storage (SSD, HDD, …):

DP-25 3.Bob create a persistent volume claim referring this time to one of the storage class:

DP-26 4.Kubernetes create the persistent volume based on the persistent volume claim and the storage class chosen:

DP-27 5.Bob create a pod and reference the persistent volume claim in the pod:

DP-28 6.Kubernetes mount the persistent volume in the pod:

DP-29

The code

Here again tested on Minikube. Should also work on other Kubernetes cluster with minor changes to the storage class.

1.Deploy 2 storage class:

kind: StorageClass
apiVersion: storage.k8s.io/v1
metadata:
  namespace: mysql
  name: hdd
provisioner: k8s.io/minikube-hostpath
---
kind: StorageClass
apiVersion: storage.k8s.io/v1
metadata:
  namespace: mysql
  name: ssd
provisioner: k8s.io/minikube-hostpath

2.Deploy a namespace:

apiVersion: v1
kind: Namespace
metadata:
  name: mysql

3.Deploy a persistent volume claim:

apiVersion: v1
kind: PersistentVolumeClaim
metadata:
  name: mysql-pvc
  namespace: mysql
spec:
  storageClassName: ssd
  accessModes:
    - ReadWriteOnce
  resources:
    requests:
      storage: 20Gi

4.Deploy a MySQL pod, again using a deployment object:

apiVersion: apps/v1
kind: Deployment
metadata:
  name: mysql
  namespace: mysql
spec:
  selector:
    matchLabels:
      app: mysql
  strategy:
    type: Recreate
  template:
    metadata:
      labels:
        app: mysql
    spec:
      containers:
      - image: mysql:5.7
        name: mysql
        env:
        - name: MYSQL_ROOT_PASSWORD
          value: password
        ports:
        - containerPort: 3306
          name: mysql
        volumeMounts:
        - name: mysql-persistent-storage
          mountPath: /var/lib/mysql
        resources:
          requests:
            cpu: 500m
            memory: 1Gi
        livenessProbe:
          exec:
            command: ["mysqladmin", "ping"]
          initialDelaySeconds: 30
          periodSeconds: 10
          timeoutSeconds: 5
        readinessProbe:
          exec:
            command: ["mysqladmin", "ping"]
          initialDelaySeconds: 5
          periodSeconds: 2
          timeoutSeconds: 1
      volumes:
      - name: mysql-persistent-storage
        persistentVolumeClaim:
          claimName: mysql-pvc

5.Expose the MySQL container via a service:

apiVersion: v1
kind: Service
metadata:
  name: mysql
  namespace: mysql
spec:
  ports:
  - port: 3306
  selector:
    app: mysql
  clusterIP: None

6.Test the MySQL database by connecting to it from inside the Kubernetes cluster using this command:

kubectl --namespace=mysql run -it --rm --image=mysql:5.7 --restart=Never mysql-client -- mysql -h mysql -ppassword

The result

Same thing as previously, if the pod dies and is rescheduled somewhere else, the data will not be lost.

We even reused a lot of the code, basically, the biggest changes are the storage class introduction and the fact that the persistent volume claim is now referencing to the storage class instead of the persistent volume.

Additionally, once the many storage class have been created by John, Bob can create without delay the persistent volume he need.

Limitation

As persistent volume are dynamically created, some people tend to forget to manage the all lifecycle, which can lead to persistent volume not deleted even if they are not used anymore. However, this can be easily worked around by reviewing from time to time the list of persistent volume instantiated and not linked to a pod.

Conclusion

To sum up everything, use the automated way, as always. And if you want to manage stateful app on top of Kubernetes, Statefulset is worth a look.

woernfl

How to provision persistent volume in Kubernetes.

Let set some context

Manually creating the persistent volumes

The flow

The code

The result

Limitation

Automatically creating the persistent volumes

The flow

The code

The result

Limitation

Conclusion