[Proposal] support periodic backups

[rjtsdl]

I already seen a bunch of ask about how to schedule periodic backup.

The current design will not support that, and suggest to use CronJob to create backup CR periodically (Only creation can trigger the actual backup handling function).

Here are some cons of this methods.

• CronJob is scheduled by k8s scheduler. For clusters has 100s etcd-operators, it will be a big load on the scheduler. (It is not the option for me at Azure) @khenidak on more thoughts.
• For users seeking backup operator, they are looking into a solution. Backup scheduling always comes with Backup itself
• It is not very intuitive to create a CronJob with access to k8s api-server. It involves secret handling.

So, i propose to have backupPolicy option.

• If backupPolicy not provided, it will not do periodic backup. It will be the same of existing behavior, which does one-time backup.
• For backupIntervalInSecond
  • unit in Sec
  • >0 means interval. Of course, we will have the minimal interval limit.
  • =0 means explicit disable interval backup, it will just do one time backup.
  • <0 not allowed, failed validation
  • It doesn't need to strictly refer to the interval. For any reason the backup operator get restarted, we can just ignore the last backup time
• For maxBackups
  • specify how many backups we want to keep
  • maxBackups == 0 => unlimited backup

Related Issues: #1787

---

Based on above proposal,

When user specify in backup CR

spec:
  backupPolicy:
    maxBackups: 4
    backupIntervalInSecond: 30
  s3:
     path: "mybucket/mycluster_backup"
  ...

We will have "mybucket/mycluster_backup_v{revision}_{date}"
And for each revision, we keep up to 4 revisions.

The same logic apply to restore operator as well.

[xiang90]

i do not like the idea to overwrite other backups. maybe we can append suffix after the provide path, , do periodic backups, and configured retentions.

[rjtsdl]

@xiang90 updated the proposal a bit, based on your comment.

[rjtsdl]

Update the proposal, we don't need to split it to slices anymore.

[xiang90]

i suggest to group the backup policy related fields together, so that people do not need the feature can simply ignore them as a whole. other than that, it looks fine in general.

[khenidak]

Shall we consider this proposal approved?

[xiang90]

@khenidak

i am ok with it. @fanminshi any opinions?

[fanminshi]

@rjtsdl what do you mean by revision? is this the etcd kv revision?

[fanminshi]

@rjtsdl also if I want to manually trigger an backup now? how would this proposal support that? do we want to support that scenario?

[khenidak]

@fanminshi what is the use case for manual trigger (Specially if a backup is running at short interval)?

[rjtsdl]

@fanminshi revision will mean etcd kv revision.

If you just want to trigger one time backup. Just don't specify values below.

  maxBackups: 4
  backupIntervalInSecond: 30