Switch branches/tags
Nothing to show
Find file History
Latest commit 2a42d73 Dec 18, 2018



Docker Repository on Quay

The efs-provisioner allows you to mount EFS storage as PersistentVolumes in kubernetes. It consists of a container that has access to an AWS EFS resource. The container reads a configmap which contains the EFS filesystem ID, the AWS region and the name you want to use for your efs-provisioner. This name will be used later when you create a storage class.


Getting Started

If you are new to Kubernetes or to PersistentVolumes this quick start will get you up and running with simple defaults.

  • Download the manifest file manifest.yaml.

    wget https://raw.githubusercontent.com/kubernetes-incubator/external-storage/master/aws/efs/deploy/manifest.yaml
  • In the configmap section change the file.system.id: and aws.region: to match the details of the EFS you created. Change dns.name if you want to mount by your own DNS name and not by AWS's *file-system-id*.efs.*aws-region*.amazonaws.com.

  • In the deployment section change the server: to the DNS endpoint of the EFS you created.

  • kubectl apply -f manifest.yaml

  • Now you can use your PersistentVolume in your pod or deployment by referencing your claim name when it's created.

kind: Pod
apiVersion: v1
  name: test-pod
  - name: test-pod
    image: gcr.io/google_containers/busybox:1.24
      - "/bin/sh"
      - "-c"
      - "touch /mnt/SUCCESS && exit 0 || exit 1"
      - name: efs-pvc
        mountPath: "/mnt"
  restartPolicy: "Never"
    - name: efs-pvc
        claimName: efs

If you scale this pod each aditional pod will also be able to read and write the same files. You may also reference the same claimName in another type of pod so your 2 applications can read and write the same files. If you wish to have a second application that uses EFS storage but don't want other pods to access the files, create a new claim using a new name but the same storage class.

Some times you want the replica pods to be on EFS but you do not wish them to share the same files. In those situations it's best to use a StatefulSet. When a StatefulSet scales up it will dynamically create new claims for your pods.

apiVersion: v1
kind: Service
  name: nginx
    app: nginx
  - port: 80
    name: web
  clusterIP: None
    app: nginx
apiVersion: apps/v1beta1
kind: StatefulSet
  name: web
  serviceName: "nginx"
  replicas: 3
        app: nginx
      terminationGracePeriodSeconds: 10
      - name: nginx
        image: gcr.io/google_containers/nginx-slim:0.8
        - containerPort: 80
          name: web
        - name: efs
          mountPath: /usr/share/nginx/html
  - metadata:
      name: efs
        volume.beta.kubernetes.io/storage-class: aws-efs
      accessModes: [ "ReadWriteOnce" ]
          storage: 1Mi

Note: We do not reference a claim name, instead we give the new claim the name efs and we list the StorageClass we wish to use when creating the claim. I like to use the type of storage, because the name of the pod will be added to it. This example when ran will create the claim efs-web-0 and scaling the pod up to 3 will create efs-web-1 and efs-web-2.

If you wish to learn more about efs-provisioner and how to change additional settings, you can continue on in the deployment section. You might also want to check the FAQ at the bottom.


Create a configmap containing the File system ID and Amazon EC2 region of the EFS file system you wish to provision NFS PVs from, plus the name of the provisioner, which administrators will specify in the provisioner field of their StorageClass(es), e.g. provisioner: example.com/aws-efs.

$ kubectl create configmap efs-provisioner \
--from-literal=file.system.id=fs-47a2c22e \
--from-literal=aws.region=us-west-2 \
--from-literal=dns.name="" \ # if you want to mount by your own DNS name and not by AWS's `*file-system-id*.efs.*aws-region*.amazonaws.com`.

See Optional: AWS credentials secret if you want the provisioner to only once at startup check that the EFS file system you specified in the configmap actually exists.

Decide on & set aside a directory within the EFS file system for the provisioner to use. The provisioner will create child directories to back each PV it provisions. Then edit the volumes section at the bottom of "deploy/deployment.yaml" so that the path refers to the directory you set aside and the server is the same EFS file system you specified.

        - name: pv-volume
            server: fs-47a2c22e.efs.us-west-2.amazonaws.com
            path: /persistentvolumes

You will need to create the directory you use for path: on your EFS file system first or the efs-provisioner pod will fail to start.

$ kubectl create -f deploy/deployment.yaml
deployment "efs-provisioner" created

If you are not using RBAC or OpenShift you can continue to the usage section.


If your cluster has RBAC enabled or you are running OpenShift you must authorize the provisioner. If you are in a namespace/project other than "default" edit deploy/rbac.yaml.


# Set the subject of the RBAC objects to the current namespace where the provisioner is being deployed
$ NAMESPACE=`kubectl config get-contexts | grep '^*' | tr -s ' ' | cut -d' ' -f5`
$ sed -i'' "s/namespace:.*/namespace: $NAMESPACE/g" ./deploy/rbac.yaml
$ kubectl create -f deploy/rbac.yaml


If SELinux is enforcing on the node where the provisioner runs, you must enable writing from a pod to a remote NFS server (EFS in this case) on the node by running:

$ setsebool -P virt_use_nfs 1
$ setsebool -P virt_sandbox_use_nfs 1



First a StorageClass for claims to ask for needs to be created.

apiVersion: storage.k8s.io/v1
kind: StorageClass
  name: slow
provisioner: example.com/aws-efs
  gidMin: "40000"
  gidMax: "50000"
  gidAllocate: "true"


  • gidMin + gidMax : A unique value (GID) in this range (gidMin-gidMax) will be allocated for each dynamically provisioned volume. Each volume will be secured to its allocated GID. Any pod that consumes the claim will be able to read/write the volume because the pod will automatically receive the volume's allocated GID as a supplemental group, but non-pod mounters outside the system will not have read/write access unless they have the GID or root privileges. See here and here for more information. Default to "2000" and "2147483647".
  • gidAllocate : Whether to allocate GIDs to volumes according to the above scheme at all. If "false", dynamically provisioned volumes will not be allocated GIDs, gidMin and gidMax will be ignored, and anyone will be able to read/write volumes. Defaults to "true".

Once you have finished configuring the class to have the name you chose when deploying the provisioner and the parameters you want, create it.

$ kubectl create -f deploy/class.yaml 
storageclass "aws-efs" created

When you create a claim that asks for the class, a volume will be automatically created.

$ kubectl create -f deploy/claim.yaml 
persistentvolumeclaim "efs" created
$ kubectl get pv
NAME                                       CAPACITY   ACCESSMODES   RECLAIMPOLICY   STATUS    CLAIM         REASON    AGE
pvc-557b4436-ed73-11e6-84b3-06a700dda5f5   1Mi        RWX           Delete          Bound     default/efs             2s

Optional: AWS credentials secret

Create a secret containing the AWS credentials of a user assigned the AmazonElasticFileSystemReadOnlyAccess policy. The credentials will be used by the provisioner only once at startup to check that the EFS file system you specified in the configmap actually exists.

$ kubectl create secret generic aws-credentials \
--from-literal=aws-access-key-id=AKIAIOSFODNN7EXAMPLE \

Add a reference to the secret in the deployment yaml.

        - name: AWS_ACCESS_KEY_ID
              name: aws-credentials
              key: aws-access-key-id
        - name: AWS_SECRET_ACCESS_KEY
              name: aws-credentials
              key: aws-secret-access-key


  • Do I have to use a configmap?

Nope, the configmap values are just fed to the container as an environment variables. It is handy though when you create your EFS with Terraform and use Terraform to create the configmap dynamically.

  • I noticed the EFS gets mounted directly to the efs-provisioner container, can I do that for my apps?

Yes you can but it's not recommended. You lose the reusability of the StorageClass to be able to dynamically provision new PersistentVolumes for new containers and pods.

  • But what if the efs-provisoner pod goes down?

Your containers will continue to have EFS storage as they are mapped directly to EFS but behind the scenes they were mounted to a folder created by the EFS provisioner. New claims will not be provisioned nor deleted while the efs-provisioner pod is not running. When it comes backup it will catchup on any work it has missed.

  • Can I scale the efs-provisioner across my nodes?

You can but it's not needed. You won't see a performance increase and you wont have a storage outage if the underlying node dies.

  • Can I have multiple efs-provisioners pointed at multiple EFS mounts?

Yes you can, but would you really need to? EFS is designed to scale across many nodes and the efs-provisioner already has the ability to divide EFS into seperate chunks for your applications.

  • I don't like the manual step of mounting and creating the /persistentvolumes directory.

It's not needed but it is helpful if you are going to use your EFS for other things than Kubernetes. In your efs-provisioner deployment set your mounts like this...

            - name: pv-volume
              mountPath: /persistentvolumes
        - name: pv-volume
            server: {{ efs_file_system_id }}.efs.{{ aws_region }}.amazonaws.com # or the same value as `dns.name`/DNS_NAME if you want to mount by your own DNS name and not by AWS's `*file-system-id*.efs.*aws-region*.amazonaws.com`.
            path: /
  • I noticed when creating the claim it has request for a really small amount of storage?

The storage section size is a requirement because most other PersistentVolumes need it. Every pod accessing EFS will have unlimited storage. I use 1Mi to remind my self it's unlimited.

  • Can I omit that part of the claim?

No, you must list a size even though it's not used with EFS.

  • Can I create multiple StorageClasses for the same provisioner?

Yes, you can create multiple StorageClasses for the same provisioner, each with their own parameters settings. Note that if two StorageClasses enable gidAllocate and the gidMin/gidMax ranges overlap, the same gid could be allocated twice.