Migrate Calico data from an etcdv3 datastore to a Kubernetes datastore

Big picture

Switch your Calico datastore from etcdv3 to Kubernetes on a live cluster.

Value

Using Kubernetes as your datastore provides a number of benefits over using etcdv3 directly, including fewer components and better support for role based access control. For most users, using the Kubernetes data store will provide a better experience. We provide a seamless way to migrate your data from an existing cluster with an etcdv3 datastore to a Kubernetes datastore. For the complete set of advantages of using a Kubernetes datastore over an etcd datastore, see Calico Datastore documentation.

Before you begin

• Ensure that your Calico installation is configured to use an etcdv3 datastore. This guide does not apply to clusters installed using the Kubernetes API datastore.

• The latest version of calicoctl must be installed and configured to access etcd.

  note

  Since the steps below require modifying calicoctl configuration, we do not recommend using calicoctl installed as a Kubernetes pod for this procedure. Instead, install the binary directly on a host with access to etcd and the Kubernetes API.

How To

Migrate the datastore

To migrate contents of the datastore, we will be using the calicoctl datastore migrate command and subcommands. For more information, see the calicoctl datastore migrate documentation.

1. Lock the etcd datastore for migration. This prevents any changes to the data from affecting the cluster.

   calicoctl datastore migrate lock

   note

   After running the above command, you cannot make changes to the configuration of your cluster until the migration is complete. New pods will not be started until after the migration.

2. Export the datastore contents to a file.

   calicoctl datastore migrate export > etcd-data

3. Configure calicoctl to access the Kubernetes datastore.

4. Import the datastore contents from your exported file.

   calicoctl datastore migrate import -f etcd-data

5. Verify that the datastore was properly imported. This can be accomplished by using calicoctl to query for any Calico resources that exist in the etcd datastore (e.g. networkpolicy).

   calicoctl get networkpolicy

6. Configure Calico to read from the Kubernetes datastore. Follow the directions to install Calico with the Kubernetes datastore. The installation instructions contain the relevant version of the calico.yaml file to apply.

   kubectl apply -f calico.yaml

7. Wait for Calico to perform a rolling update before continuing by monitoring the following command

   kubectl rollout status daemonset calico-node -n kube-system

8. Unlock the datastore. This allows the Calico resources to affect the cluster again.

   calicoctl datastore migrate unlock

   note

   Once the Kubernetes datastore is unlocked, the datastore migration cannot be rolled back. Make sure that the Kubernetes datastore is populated with all of the expected Calico resources prior to unlocking the datastore.

Roll back the datastore migration

Rolling back the datastore migration can only be done if the original etcd datastore still exists and the Kubernetes datastore was not unlocked after the datastore resources were imported. The following steps delete the Calico resources imported into the Kubernetes datastore and configure the cluster to once again read from the original etcd datastore.

1. Lock the Kubernetes datastore.

   calicoctl datastore migrate lock

2. Delete all of the Calico CRDs. This will remove all of the data imported into the Kubernetes datastore.

   kubectl delete $(kubectl get crds -o name | grep projectcalico.org)

3. Configure Calico to read from the etcd datastore. Follow the directions to install Calico with the etcd datastore. The installation instructions contain the relevant version of the calico.yaml file to apply.

   kubectl apply -f calico.yaml

4. Configure calicoctl to access the etcd datastore.

5. Unlock the etcd datastore. This allows the Calico resources to affect the cluster again.

   calicoctl datastore migrate unlock