Output of restic version

restic 0.16.2 compiled with go1.21.3 on linux/arm64

What backend/service did you use to store the repository?

AWS S3

Problem description / Steps to reproduce

If you upload backups to S3 with -o s3.storage-class=GLACIER or -o s3.storage-class=DEEP_ARCHIVE it makes your repository partially broken.
If you try to exec restic snapshots from another host, or using --no-cache flag the output for this would be:

 $ restic snapshots                                                                                               
repository fffffffff opened (version 1)
Load(<snapshot/aaaaaaaaa>, 0, 0) returned error, retrying after 508.626707ms: The operation is not valid for the object's storage class
Load(<snapshot/bbbbbbbb>, 0, 0) returned error, retrying after 308.071881ms: The operation is not valid for the object's storage class
Load(<snapshot/aaaaaaaaa>, 0, 0) returned error, retrying after 1.014433406s: The operation is not valid for the object's storage class
Load(<snapshot/bbbbbbbbb>, 0, 0) returned error, retrying after 400.917258ms: The operation is not valid for the object's storage class
Load(<snapshot/bbbbbbbbb>, 0, 0) returned error, retrying after 812.242443ms: The operation is not valid for the object's storage class
  signal interrupt received, cleaning up
Ignoring "aaaaaaaaaaaa...": failed to load snapshot aaaaaaaaaa: context canceled
Ignoring "bbbbbbbbbbb...": failed to load snapshot bbbbbbbbb: context canceled
ID        Time                 Host        Tags        Paths
------------------------------------------------------------------------------------
cccccccc  2023-12-04 00:15:45  my-host               /home/LICENSE
ddddddd  2023-12-04 13:51:33  my-host               /mnt/media/photo
eeeeeeee  2023-12-04 23:21:57 my-host               /home/README.md
------------------------------------------------------------------------------------
3 snapshots

To get this output I executed restic snapshots, and soon after I got few errors - clicked CTRL+C.
Here you can see, in the repository with 5 (A, B, C, D, E) uploaded snapshots, 2 of them (A, B) are uploaded to the Glacier Flexible and Glacier Deep Archive tiers.

This issue not persists to the GLACIER_IR storage class, because all files are 'instant retrieval', so no issue here (the C-snapshot).

It is expected that if data is archived - it needed to be 'retrieved' beforehand in the AWS. And when I did so (on the machine with cache) - I was able to restic restore it.

Expected behavior

When non-instant retrieval storage class is opted-in - metadata is stored in STANDARD storage class, while packfile data is saved to specified tier.

Actual behavior

Snapshot metadata is unreachable because it is also stored in specified storage class.

Do you have any idea what may have caused this?

yes. It is because all files sent to the same tier without filtering metadata.
I've made a fallback fix for this issue, which I tested and it's working as intended: Ignore storage class option for metadata if sotrage class is either GLACIER or DEEP_ARCHIVE

Did restic help you today? Did it make you happy in any way?

Yup, I became quite enthusiastic about opportunity to safely backup significant media in cold storage for affordable money, so I tried this myself, and only after found out that I can't access them from another machine. And I even promoted this feature to friend, before trying it myself. So that's why I'm about to create the PR.