Etcd's backup/restore tooling is good for recovering from data loss in a single etcd cluster. For example, it is a good idea to take a backup of etcd prior to upgrading etcd itself. For more sophisticated management of your Kubernetes cluster backups and restores, we feel that Ark is generally a better approach. It gives you the ability to throw away an unstable cluster and restore your Kubernetes resources and data into a new cluster, which you can't do easily just by backing up and restoring etcd.
Examples of cases where Ark is useful:
- you don't have access to etcd (e.g. you're running on GKE)
- backing up both Kubernetes resources and persistent volume state
- cluster migrations
- backing up a subset of your Kubernetes resources
- backing up Kubernetes resources that are stored across multiple etcd clusters (for example if you run a custom apiserver)
Yes, with some exceptions. For example, when Ark restores pods it deletes the nodeName from the
pod so that it can be scheduled onto a new node. You can see some more examples of the differences
in pod_action.go
We strongly recommend that you use a separate bucket per cluster to store backups. Sharing a bucket across multiple Ark instances can lead to numerous problems - failed backups, overwritten backups, inadvertently deleted backups, etc., all of which can be avoided by using a separate bucket per Ark instance.
Related to this, if you need to restore a backup from cluster A into cluster B, please use restore-only
mode in cluster B's Ark instance (via the --restore-only flag on the ark server command specified
in your Ark deployment) while it's configured to use cluster A's bucket. This will ensure no
new backups are created, and no existing backups are deleted or overwritten.