On 04/20/2015 04:47 PM, Jan Engelhardt wrote:
On Monday 2015-04-20 04:45, Tim Serong wrote:
As long as process credentials are not 64/128-bit UIDs similar to NTSIDs, services ought to cope with different numerical UIDs. NFS made the jump with idmapd, so I am sure ceph can come up with something too. Perhaps even reuse idmapd.
I agree. There are still some use cases where same uids are desirable nevertheless. For example to be able to reuse a file system with data on it on different installations.
That's exactly why the ceph project wants a static UID - so you can easily hot swap a disk from one node in the cluster to another node.
That seems bogus.
The data that people generally share in networked fashion is user data (rather than, say, operating system data).
These user files (hereby by definition) already have consistent UID allocation, e.g. because the participating import nodes all used a centrally-administered database.
Not in this case. I'm not talking about directly user-accessible files, I'm talking about ceph OSDs. A ceph OSD is (approximately) a single disk with an xfs/btrfs filesystem on top. The ceph-osd daemon manages everything on this filesystem; objects stored by clients in a ceph cluster map back (somehow, by magic) to files on an OSD, but the client doesn't see them at this level in the stack. Rather, all files on an OSD are are created and owned by the user ceph-osd runs as (presently root, but in future, the unprivileged user "ceph"). In order to be able to hotswap an OSD from one node to another, the UID for ceph thus needs to be the same across all nodes in the cluster.
The OS files are naturally a jumble, but that is ok, because they are generally not shared. If, for whatever reason, they are to be shared, then one would have to make sure that *all* system user UIDs are also consistently allocated — which is not something that will evaluate to `true` in most installations however. So, giving _just ceph_ a fixed static UID is nothing but a drop on a hotstone.
We're not worried about OS files, just the files owned by the ceph-osd process on the OSD disks, which is to say, all the files backing all the objects stored in the ceph cluster :) Regards, Tim -- Tim Serong Senior Clustering Engineer SUSE tserong@suse.com -- To unsubscribe, e-mail: opensuse-factory+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse-factory+owner@opensuse.org