k8s-volume-discovery is a Kubernetes operator that discovers EBS volumes in AWS tagged with the cluster identifiers and attaches them to the cluster in the form of PersistentVolume and PersistentVolumeClaim.

When rebuilding a Kubernetes cluster (e.g. in case of disaster), we might want to reattach the volumes that were attached to the previous cluster. This operator allows for an easy and automatic volume recovery. The cluster will automatically rediscover the volumes and attach them.

The operator deployment files are expressed in dhall for type safety and customizability. To apply it to your cluster you need to install the dhall-to-yaml tool (inside dhall-json). See https://github.com/dhall-lang/dhall-haskell#pre-built-binaries for more information on installing it.

Once you have dhall-to-yaml installed, you can apply the files to your cluster by doing

export AWS_CLUSTER_NAME=example #set it to your cluster name
make deploy

By default, the operator tries to assume a IAM role called App-k8s-volume-discovery using kube2iam or kiam. You can rename the role by changing k8s/config.dhall. The role needs to have the ec2:DescribeVolumes permission.

k8s-volume-discovery operates by retrieving the AWS EBS volumes and using the information that Kubernetes puts in the tags. k8s-volume-discovery will only consider volumes with the kubernetes.io/cluster/{clusterName} tag that are in the available state.

The following tags are required for k8s-volume-discovery to add the EBS volume to Kubernetes:

• kubernetes.io/created-for/pv/name
• kubernetes.io/created-for/pvc/name
• kubernetes.io/created-for/pvc/namespace These tags are usually automatically created by Kubernetes when creating PersistentVolumes/PersistentVolumeClaims.

If any of these tags is missing, the volume is not added to the cluster.

Optionally, users can add the tag kubernetes.io/created-for/sc/name to signal which storage class should be assigned to the volume. To add the storage class tag, we recommend using k8s-metadata-injector.

The application is developed in Haskell. To build the project you need stack. Run make build to build the project locally. Run make test to test the project locally.

You can deploy a dry run version of the project by adding the --dry-run flag to the deployment.

To build a new docker image, run make build-docker-image.

The project is actively maintained but it's in alpha stage, expect breaking changes.