What do you think S3 or other such immutability protection do here (more than PBS does already)?
Note that the content addressable storage already gives you strong protection, and more important the possible to safely detect any wrong doing; which then allows you to restore from syncs and/or tapes - see our
chapter in the docs for how to protect against ransomware or other data mutation in PBS.
Ceph, S3 and their lock features are just software with a "please don't delete me" on a bit of steroids and on a separate instances (i.e., decoupled from the backup server host), certainly nice and can bring some protection, but I don't see how they "magically" protect more than our content integrity verification combined with syncs or tape backup do.
Please also note that with Proxmox Backup Server 2.4 we got support for WORM tapes (pro-tip: not the best terms to google without some additional terms like LTO to avoid getting a nice view of some parasites
) - Write Once Read Many tapes.
Those (and naturally normal tapes too) you can lock in a safe or otherwise secure location, giving one of the safest and most secure "if all else fails" protections there is.
That said, we still plan to implement S3, but due to its architecture it will need some special handling over standard syncs – that's why we also hinted that it will work more like tape does now. But I don't think it will have the huge safety benefits from some immutable flag on the (S3) storage over our existing architecture; but sure, I can understand wanting it, especially if one already depends on that already (i.e., has it evaluated and heavily tested to meet their requirements) and/or has it on the (often not so mutable) "must have" checklist from higher ups. When exactly I cannot say, that would just create wrong expectations (our devs normally don't work by hard deadlines, but rather ship stuff when it is ready, to avoid half-baked things getting pushed through), but we started working on some prerequisites in the sync framework (local-to-local and push based sync) which should both make it a bit easier to implement a s3 target then.
