PSA: The Ceph Rook update to 1.20 is a bit more involved due to the introduction of the ceph-csi-drivers Helm chart. Have a look at the upgrade docs: https://rook.github.io/docs/rook/v1.20/Upgrade/rook-upgrade/
Also have a look at this GitHub issue for a few things which might be helpful but haven't made their way into the docs yet: https://github.com/rook/rook/issues/17696
I've already trying leaving out the capacity definition entirely, as well as setting it to 0 in the PVC, but k8s accepts neither.
Next thing to try is whether I can somehow manually create the CephFS subvolume and then create a PV/PVC for it. But then again, those PVs and PVCs would invariably have capacities as well.
I'm currently trying to figure out a storage issue: In my baremetal Ceph cluster, I've got a CephFS subvolume which doesn't have a quota - it gets to write to the entirety of my storage. It contains stuff like my Linux ISO collection, and because it doesn't get any automated writes at all, I'm pretty comfortable with it.
And I would like to repeat that for my k8s Ceph Rook cluster, but cannot figure out how to set up a CephFS PVC without a capacity/quota.
1/2
Today's service update round is really containing some bangers. Just read the release notes for Ceph Rook's 1.16 release (https://github.com/rook/rook/releases/tag/v1.16.0), and with it, ObjectBucketClaims can now provide a bucket policy, so I won't have to add one manually to e.g. give my backup user access to the bucket.
I'm just updating Ceph Rook to 1.14, and there's some nice features in there. For one, you can now set the application (rgw,rbd,cephfs) on the pool, which is nice for pool reuse! And they've also implemented sharing of pools for multiple object stores.
The update would have been extremely smooth if I hadn't forgotten to update the Cluster chart before the operator chart. 🙄
But recovery was automatic and fast. I'm always impressed how reliable Ceph and Rook are.
Michael