Remove ceph reference in Storage Considerations

[swills]

This page:

https://cloudnative-pg.io/documentation/1.20/storage/#block-storage-considerations-ceph-longhorn

suggests using a pool with 1x replication with Ceph as backing for DB PVCs.

But due to the way Ceph works, all blocks for all PVCs using a given pool are distributed across all OSDs, meaning that with 1x replication, if you lose a single OSD, you lose the entire pool (all DB replicas). So while this advice may be fine for Longhorn (I think it is based on what I'm told and have read, but I'm no expert on it), from what I know of Ceph this is not good advice for that use case.

So I think the best thing to do in this doc would be to not mention Ceph or explicitly exclude Ceph from this suggestion.

[fhoekstra]

You are linking to a very old version of the docs.

In the current docs I see that the Ceph section lists CRUSH rules to ensure one volume per node:

In Ceph, this can be configured through CRUSH rules. The documentation for configuring CRUSH rules is available here. These rules aim to ensure one volume per pod per node. You can also relocate volumes by importing them into a different pool.

https://cloudnative-pg.io/documentation/current/storage/#block-storage-considerations-ceph-longhorn

(it happens all the time, search engines don't seem to like current yet)

[uhthomas]

I still think the Ceph recommendation is strange and inherently dangerous. I don't think there's even any way to guarantee data locality. I frankly have no idea why those rook docs even exist as the administrative burden and risk is so high.

[jsilvela]

This is not the right repo to fix documentation.
The top-level https://github.com/cloudnative-pg/cloudnative-pg repo is.
Please open there.
Also, I might add, the document to fix, in that repo, is in docs/src/storage.md. You might directly make a linked PR with your proposal.
Closing this Issue here, encouraging you to open in the right repo.