On Thu, Sep 8, 2016 at 2:06 AM, Gregory Farnum <gfarnum@xxxxxxxxxx> wrote: > On Wed, Sep 7, 2016 at 7:44 AM, Michael Sudnick > <michael.sudnick@xxxxxxxxx> wrote: >> I've had to force recreate some PGs on my cephfs data pool due to some >> cascading disk failures in my homelab cluster. Is there a way to easily >> determine which files I need to restore from backup? My metadata pool is >> completely intact. > > Assuming you're on Jewel, run a recursive "scrub" on the MDS root via > the admin socket, and all the missing files should get logged in the > local MDS log. This isn't quite accurate -- the forward scrub is only checking for the first object in the file (which contains the backtrace), so it won't identify any files where other objects may have been in the lost PGs. Also, it turns out that the MDS doesn't actually log anything in this case; the issue is noted in the scrub result object for the inode, but that doesn't go anywhere unless you were explicitly doing "scrub_path /<the file>" in which case you get the detailed results on the command line. Anyway -- currently there isn't an efficient tool for answering the question "which files have objects in this PG?". The only way to work it out is to scan through every possible object in every file in the system. You can sort of do this by writing a script, but it'll be very slow if you have to call into "ceph osd map" for each object ID. It may well be faster to do a full restore from your backup. John > (I'm surprised at this point to discover we don't seem to have any > documentation about how scrubbing works. It's a regular admin socket > command and "ceph daemon mds.<name> help" should get you going where > you need.) > -Greg > _______________________________________________ > 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
 |