So far, we have discussed storage solutions that stored the data outside the Kubernetes cluster (except for emptyDir and HostPath, which are not persistent). Flocker is a little different. It is Docker-aware. It was designed to let Docker data volumes transfer with their container when the container is moved between nodes. You may want to use the Flocker volume plugin if you're migrating a Docker-based system that use a different orchestration platform, such as Docker compose or Mesos, to Kubernetes and you use Flocker for orchestrating storage. Personally, I feel that there is a lot of duplication between what Flocker does and what Kubernetes does to abstract storage.

Flocker has a control service and agents on each node. Its architecture is very similar to Kubernetes with its API server and the Kubelet running on each node. The Flocker control service exposes a REST API and manages the configuration of the state across the cluster. The agents are responsible for ensuring that the state of their node matches the current configuration. For example, if a dataset needs to be on node X, then the Flocker agent on node X will create it.

The following diagram showcases the Flocker architecture:

In order to use Flocker as persistent volumes in Kubernetes, you first must have a properly configured Flocker cluster. Flocker can work with many backing stores (again, very similar to Kubernetes persistent volumes).

Then you need to create Flocker datasets and at that point you're ready to hook it up as a persistent volume. After all your hard work, this part is easy and you just need to specify the Flocker dataset name:

apiVersion: v1
kind: Pod
metadata:
  name: some-pod
spec:
  containers:
    - name: some-container
      image: kubernetes/pause
      volumeMounts:
          # name must match the volume name below
          - name: flocker-volume
            mountPath: "/flocker"
  volumes:
    - name: flocker-volume
      flocker:
        datasetName: some-flocker-dataset