On Thu, Sep 29, 2016 at 12:59 PM, Dan van der Ster <dan@xxxxxxxxxxxxxx> wrote: > Hi all, > > We currently have a CephFS cluster dedicated to a single user running > an HPC cluster. The client machines use a single keyring which > accesses the whole cephfs filesystem (i.e. there are no mds prefix or > namespace limitations in their client caps). > > We want to transition this user (and the cluster) to OpenStack Manila, > using the CephFS Native driver. So we're wondering what would be the > best way to make this transition. I think we have these options: > > 1. Create a new fs for Manila, give the HPC users Manila quota then > let them migrate their data and mounts. I'm not sure if multi-fs is > stable in 10.2.3 ?? It's still protected by an experimental feature flag, but it was always one of the less risky items, it's had some bug fixes in the jewel branch and I would be fairly comfortable switching it on (Manila itself probably more unstable than multi-fs is). Remember you'll need to spin up an extra MDS for each filesystem. Come to think of it, while you can control which filesystem Manila uses in its ceph.conf (client_mds_namespace), the cloud tenants won't have a way to know which filesystem they're meant to be connecting to, so you would have to set the Manila filesystem as the "legacy" one in the FSMap so that the clients would connect to it by default. > 2. Configure Manila to use the existing fs, with Manila data ending up > in the /volumes directory. HPC users would again create a new Manila > share and migrate their data. This would be easier and is for sure > stable -- but it's not clear to me if the Manila share namespaces > would break the existing native usage of the HPC machines. This should be fine, Manila doesn't do anything to the filesystem outside of its own directory. Note that when you migrate data, you'll find that a "mv" probably won't work, because the files won't have the right RADOS namespace for the share (Manila clients are only given capabilities for their share's namespace), you'll need to do a "cp". John > > We'd prefer (2), but just wanted to check if there is something we missed. > > Best Regards, > > Dan van der Ster > CERN IT > _______________________________________________ > ceph-users mailing list > ceph-users@xxxxxxxxxxxxxx > http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com _______________________________________________ ceph-users mailing list ceph-users@xxxxxxxxxxxxxx http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
 |