Allow users to specify a Storage Class when backing up to S3

[cullenmcdermott]

Describe the solution you'd like
I would like the ability to specify a storage class in order to avoid having to setup additional lifecycle rules for objects in my bucket(s).

Anything else you would like to add:
Any metadata associated with the backup that is kept in the S3 bucket would probably need to be kept in the 'standard' class. Otherwise S3 costs could balloon if Ark were to frequently try to read from it when it was in the IA tier for example. Also extra consideration would need to be taken for restores from Glacier since it can take up to a few hours to be available and is only available for a certain amount of time.

[nrb]

Thanks for this request, and apologies about the delayed response.

I haven't done a code dive on this yet, but I think we should be able to add this as an entry to a BackupStorageLocation.ObjectStorage.config map, and maybe some validation on the storage class name.

@skriss @rosskukulinski Any other thoughts?

[skriss]

@cullenmcdermott are you envisioning different storage classes for the different files stored in S3 (there's ark-backup.json, the backup tarball itself, a log file, etc.)? Also - are you envisioning Ark/Velero changing these over time? Or is a single static storage class, configurable by you, for each object that gets uploaded sufficient?

[cullenmcdermott]

I think a single static storage class would be fine in most cases. The only potential issue I'd see with that is if there are any files that get uploaded that store metadata that are retrieved before a backup is performed then it could be expensive to retrieve them regularly in the case of IA or could cause the Ark backup to fail in the case of Glacier.

[sladyn98]

@nrb Would like to tackle this issue. Any tips on how to begin since this is my first time contributing here and looks like a good first issue.

[nrb]

@sladyn98 Very sorry for the delay. We're now doing a large triage, and I wondered if you're still interested.

Some suggestions if you are, please read out contributing guide. We would need a design doc that would describe what changes would be made and what impacts it would have.

If I were to do this, I would probably put the storage class in the BackupStorageLocation's config entry, which is used for provider-specific data. Then, the AWS plugin would use that information, if possible.

[tppalani]

can we use glacier while installing velero instead of s3 bucket?