Restore etcd backup to a different namespace

[jungopro]

Hello

I'm trying to perform a restore to an etcd cluster in a different namespace than the operator and fails. I read through the docs and couldn't find the exact spec to make it work.

Scenario

• operator in the default namespace
• cluster in mp-temp namespace

Backup worked great using the following method:

• create the aws secret and the backup_cr.yaml in the same namespace as the operator (default)
• modified backup_cr.yaml to include the full cluster path:

  apiVersion: "etcd.database.coreos.com/v1beta2"
  kind: "EtcdBackup"
  metadata:
  name: example-etcd-cluster-backup
  spec:
  etcdEndpoints: [http://mp-etcd-client.mp-temp.svc.cluster.local:2379]
  storageType: S3
  s3:
  # The format of "path" must be: "<s3-bucket-name>/<path-to-backup-file>"
  # e.g: "mybucket/etcd.backup"
  path: "mybucket/etcd.backup"
  awsSecret: aws

The restore process, not so great:

• re-created the cluster in the mp-temp namespace:

  $ kg pod -n mp-temp -l app=etcd
  NAME                 READY   STATUS    RESTARTS   AGE
  mp-etcd-8pzxtltnj4   1/1     Running   0          48m
  mp-etcd-cn6khprk55   1/1     Running   0          49m
  mp-etcd-lvdfwlsjj2   1/1     Running   0          49m

• tried running the restore in both namespaces with different specs but failed:

  piVersion: "etcd.database.coreos.com/v1beta2"
  kind: "EtcdRestore"
  metadata:
  # The restore CR name must be the same as spec.etcdCluster.name
  name: mp-etcd
  spec:
  etcdCluster:
  # The namespace is the same as this EtcdRestore CR
  name: mp-etcd
  namespace: mp-temp
  backupStorageType: S3
  s3:
  # The format of "path" must be: "<s3-bucket-name>/<path-to-backup-file>"
  # e.g: "mybucket/etcd.backup"
  path: "mybucket/etcd.backup"
  awsSecret: aws

• operator logs:

  time="2019-02-22T13:38:14Z" level=error msg="error syncing restore request (default/mp-etcd): prepare seed failed: failed to get reference EtcdCluster(default/mp-etcd): etcdclusters.etcd.database.coreos.com \"mp-etcd\" not found" pkg=controller
  time="2019-02-22T13:39:38Z" level=error msg="error syncing restore request (default/mp-etcd): prepare seed failed: failed to get reference EtcdCluster(default/mp-etcd): etcdclusters.etcd.database.coreos.com \"mp-etcd\" not found" pkg=controller
  time="2019-02-22T13:41:10Z" level=error msg="error syncing restore request (default/mp-etcd.mp-temp): prepare seed failed: failed to get reference EtcdCluster(default/mp-etcd.mp-temp): etcdclusters.etcd.database.coreos.com \"mp-etcd.mp-temp\" not found" pkg=controller
  time="2019-02-22T13:47:34Z" level=error msg="error syncing restore request (default/mp-etcd): prepare seed failed: failed to get reference EtcdCluster(default/mp-etcd): etcdclusters.etcd.database.coreos.com \"mp-etcd\" not found" pkg=controller

I tried the following modifications to the restore yaml:

• add the full path to the name of the cluster
• add a namespace spec in the metadata and / or the etcdCluster spec

Nothing worked. Any suggestions?

Thanks Omer

[alaypatel07]

@jungopro The restore-operator only works if the restore-operator and the etcd cluster you are trying to restore are in the same namespace.

In the operator logs you provided, the restore operator is run in default namespace and the etcdcluster is in mp-temp namespace, hence it is failing.

Can I ask what exactly are you trying to achieve in having the restore to work across namespaces?

[jungopro]

hi @alaypatel07 thanks for the reply. I thought the restore can work across namespaces, same as the backup. I basically wanted to deploy the operator once, in a single namespace, and have it manage multiple clusters in different namespaces Since this feature is already possible for creating and backing-up clusters, I thought it only makes sense it will be possible for restoring cluster data as well