Automatic Volume Snapshots on Kubernetes.
Switch branches/tags
Clone or download
miracle2k Merge pull request #69 from moepot/master
Update README how to set AWS credentials.
Latest commit bdc7f94 Oct 22, 2018

Automatic Volume Snapshots on Kubernetes

How is it useful? Simply add an annotation to your PersistentVolume or PersistentVolumeClaim resources, and let this tool create and expire snapshots according to your specifications.

Supported Environments:

  • Google Compute Engine disks.
  • AWS EBS disks.

Want to help adding support for other backends? It's pretty straightforward. Have a look at the API that backends need to implement.


Let's run k8s-snapshots in your cluster:

cat <<EOF | kubectl create -f -
apiVersion: extensions/v1beta1
kind: Deployment
  name: k8s-snapshots
  namespace: kube-system
  replicas: 1
        app: k8s-snapshots
      - name: k8s-snapshots
        image: elsdoerfer/k8s-snapshots:v2.0

Add a backup annotation to one of your persistent volumes:

kubectl patch pv pvc-01f74065-8fe9-11e6-abdd-42010af00148 -p \
  '{"metadata": {"annotations": {"": "P1D P30D P360D"}}}'

k8s-snapshots will now run in your cluster, and per the deltas given, it will create a daily snapshot of the volume. It will keep 30 daily snapshots, and then for one year it will keep a monthly snapshot. If the daemon is not running for a while, it will still try to approximate your desired snapshot scheme as closely as possible.

A tip for kops users

k8s-snapshots need EBS and S3 permissions to take and save snapshots. Under the kops IAM Role scheme, only Masters have these permissions. The easiest solution is to run k8s-snapshots on Masters.

To run on a Master, we need to:

To do this, add the following to the above manifest for the k8s-snapshots Deployment:

      - key: ""
        operator: "Equal"
        value: ""
        effect: "NoSchedule"
      nodeSelector: master

How do the deltas work

The expiry logic of tarsnapper is used.

The generations are defined by a list of deltas formatted as ISO 8601 durations (this differs from tarsnapper). PT60S or PT1M means a minute, PT12H or P0.5D is half a day, P1W or P7D is a week. The number of backups in each generation is implied by it's and the parent generation's delta.

For example, given the deltas PT1H P1D P7D, the first generation will consist of 24 backups each one hour older than the previous (or the closest approximation possible given the available backups), the second generation of 7 backups each one day older than the previous, and backups older than 7 days will be discarded for good.

The most recent backup is always kept.

The first delta is the backup interval.

Running the daemon

Use the example deployment file given above to start off. If you run the daemon on a Google Container Engine cluster, it should already have access to all the resources it needs.

However, depending on your configuration, you may need to assign the correct RBAC rules, or give it access to a Google Cloud identity that has permissions to create snapshots. See below for more on the available configuration options.

How to enable backups

To backup a volume, you should add an annotation with the name to either your PersistentVolume or PersistentVolumeClaim resources.

Since PersistentVolumes are often created automatically for you by Kubernetes, you may want to annotate the volume claim in your resource definition file. Alteratively, you can kubectl edit pv a PersistentVolume created by Kubernetes and add the annotation.

The value of the annotation are a set of deltas that define how often a snapshot is created, and how many snapshots should be kept. See the section above for more information on how deltas work.

In the end, your annotation may look like this: PT1H P2D P30D P180D

There is also the option of manually specifying the volume names to be backed up as options to the k8s-snapshots daemon. See below for more information.


Configure access permissions to Google Cloud

If there are no default credentials to Kubernetes and the Cloud snapshot API, or the default credentials do not have the required access scope, you may need to configure these.

GCLOUD_PROJECT Name of the Google Cloud project. This is required to use the Google Cloud API, but if it's not given, we try to read the value from the [instance metadata service]( which will usually work.
GCLOUD_JSON_KEYFILE_NAME Filename to the JSON keyfile that is used to authenticate. You'll want to mount it into the container.
GCLOUD_JSON_KEYFILE_STRING The contents of the JSON keyfile that is used to authenticate.
KUBE_CONFIG_FILE Authentification with the Kubernetes API. By default, the pod service account is used.

When using a service account with a custom role to access the Google Cloud API, the following permissions are required:


Configure access permissions on AWS

If there are no default credentials to the Cloud API, or the default credentials do not have the required access scope, you may need to configure these environment variables.

AWS_ACCESS_KEY_ID AWS IAM Access Key ID that is used to authenticate.
AWS_SECRET_ACCESS_KEY AWS IAM Secret Access Key that is used to authenticate.
AWS_REGION The region is usually detected via the meta data service. You can override the value.

For Role-based Access Control (RBAC) enabled clusters

In kubernetes clusters with RBAC, the required permissions need to be provided to the k8s-snapshots pods to watch and list persistentvolume or persistentvolumeclaims.

  kubectl apply -f rbac.yaml

Pinging a third party service

PING_URL We'll send a GET request to this url whenever a backup completes. This is useful for integrating with monitoring services like Cronitor or Dead Man's Snitch.

Make snapshot names more readable

If your persistent volumes are auto-provisioned by Kubernetes, then you'll end up with snapshot names such as pv-pvc-01f74065-8fe9-11e6-abdd-42010af00148. If you want that prettier, set the enviroment variable USE_CLAIM_NAME=true. Instead of the auto-generated name of the persistent volume, k8s-snapshots will instead use the name that you give to your PersistentVolumeClaim.

Manual snapshot rules

It's possible to ask k8s-snapshots to create snapshots of volumes for which no PersistentVolume object exists within the Kubernetes cluster. For example, you might have a volume at your Cloud provider that you use within Kubernetes by referencing it directly.

To do this, we use a custom Kubernetes resource, SnapshotRule.

First, you need to create this custom resource.

On Kubernetes 1.7 and higher:

cat <<EOF | kubectl create -f -
kind: CustomResourceDefinition
  version: v1
  scope: Namespaced
    plural: snapshotrules
    singular: snapshotrule
    kind: SnapshotRule
    - sr

Or on Kubernetes 1.6 and lower:

cat <<EOF | kubectl create -f -
apiVersion: extensions/v1beta1
kind: ThirdPartyResource
description: "Defines snapshot management rules for a disk."
- name: v1

You can then create SnapshotRule resources:

cat <<EOF | kubectl apply -f -
apiVersion: ""
kind: SnapshotRule
  name: mysql
  deltas: P1D P30D
  backend: aws
     region: eu-west-1
     volumeId: vol-0aa6f44aad0daf9f2

This is an example for backing up an EBS disk on the Amazon cloud. The disk option requires different keys, depending on the backend. See the examples folder.

You may also point SnapshotRule resources to PersistentVolumes (or PersistentVolumeClaims). This is intended as an alternative to adding an annotation; it may be desirable for some to separate the snapshot functionality from the resource.

cat <<EOF | kubectl apply -f -
apiVersion: ""
kind: SnapshotRule
  name: mysql
  deltas: P1D P30D
  persistentVolumeClaim: datadir-mysql

Backing up the etcd volumes of a kops cluster

After setting up the custom resource definitions (see previous section), use snapshot rules as defined in the examples/backup-kops-etcd.yml file. Reference the volume ids of your etcd volumes.

Other environment variables

LOG_LEVEL **Default: INFO**. Possible values: DEBUG, INFO, WARNING, ERROR
JSON_LOG **Default: False**. Output the log messages as JSON objects for easier processing.


What if I manually create snapshots for the same volumes that k8s-snapshots manages?

Starting with v0.3, when k8s-snapshots decides when to create the next snapshot, and which snapshots it deletes, it no longer considers snapshots that are not correctly labeled by it.


For local development, you can still connect to an existing Google Cloud Project and Kubernetes cluster using the config options available. If you are lucky, your local workstation is already setup the way you need it. If we can find credentials for Google Cloud or Kubernetes, they will be used automatically.

However, depending on the backend, you need to provide some options that otherwise would be read from the instance metadata:

For AWS:

$ AWS_REGION=eu-west-1 python -m k8s_snapshots

For Google Cloud:

$ GCLOUD_PROJECT=revolving-randy python -m k8s_snapshots