Any example using glacier?

[bbigras]

Do I just need to use a glacier s3 bucket for cold and a regular s3 bucket for hot?

There's more than 1 Glacier StorageClass, and it seems a normal s3 bucket can transition its file to glacier https://registry.terraform.io/providers/hashicorp/aws/latest/docs/resources/s3_bucket#transition

[aawsome]

rustic doesn't support single repositories/buckets which have objects that can be in different "states" (something like this is not contained in the restic REST protocol; rclone therefore isn't able to provide this kind of information).

So, to use S3 glacier, you should use 2 buckets: A hot one which is always accessible and a cold one where all object should be transitioned/transferred directly to Glacier when they are stored. The "hot" objects are then in fact stored in both buckets, but this shouldn't be much overhead as it is only metadata which usually is less than 1% of the repo size. (The side effect is that the cold repo contains a full restic-compatible repository; if you warm up it completely, you can use it a standard repository within rustic and even restic).

TL;DR: I think you have to do the following:

1. To store data (e.g. init or backup command):

• Create a regular bucket and use it as repo-hot
• Create a bucket for cold and configure it to store data directly in Glacier (don't know if you can configure this in rclone or in AWS)

2. To retrieve data (e.g. restore):

• configure warm-up-command such that objects from the cold buckets are restored, see e.g https://docs.aws.amazon.com/AmazonS3/latest/userguide/restoring-objects.html. That is, the config file should look have a line like warm-up-command = 'aws s3api restore-object --bucket BUCKET --key path/data/%id --restore-request '{"Days":25,"GlacierJobParameters":{"Tier":"Standard"}}' ' (may need some adaption)
• configure a suitable warm-up-wait to ensure that the data is warmed up and can be accessed after that
• Note: All commands which need to read cold data now will warm this data up and wait for the given time before processing the data

3. removing cold data (prune)
   This actually should work out of the box, but note that by default prune doesn't repack any cold pack file; instead it keeps them until the last blob within is no longer used.

• to do more repacking, have a look at the repack-cacheable option which does allow to repack cold pack files.
• Use the keep-pack option if you have some minimum holding duration, i.e. if you would get charged if you remove objects too early. This option ensures that only pack files older than the given time will be removed.

Once you get a working configuration, can please share it so other users can also use it?

[mprasil]

I'm linking it here because this came up when I initially investigated how to use Rustic with Glacier.

I have written two articles on the topic - economic side of hosting rustic repo in AWS Glacier and the technical side of migrating to Glacier, restoring and such. It's really quite complex topic from both points of view and there are some sharp edges especially around restoring the backups.

Hopefully it helps someone.

[mprasil]

Yes, feel free to link it.

[aawsome]

Thanks, is now linked.
And, BTW, the first article of rustic (that I am aware of) 🎉

[mprasil]

I think I need to post couple updates, but before I do I figured I'll check it with you @aawsome.

• The copied repo has similar size, but about 10x fewer pack files. (so going from ~300k -> ~30k) Is that down to bigger pack sizes? I have copied over all snapshots and the data seems to be there. (the overall repo size is similar)
  • This seems to save on upfront cost quite a bit. So nice side effect, just making sure it's expected.
• All of the snapshots that were copied over show ? instead of number in the Files, Dirs and Size column. Newly created snapshots show actual numbers there.
  • Why is that?

[aawsome]

@mprasil Your source repo is a restic repository, right?

• rustic uses a much higher pack size for data packs, depending on the actual repository size
• if you copy from a snapshot, which doesn't have these metadata saved, it won't be saved in the target repository. What does rustic snapshots (using the profile you are using with copy show? The snapshots in the target repository should be identical. If your source snapshots have been generated by restic, they lack this information, this is WIP in restic.

[mprasil]

Yes exactly, the source is restic. Thanks for confirming.