Den ons 22 jan. 2020 kl 18:01 skrev Wesley Dillingham <w...@wesdillingham.com >:
> After upgrading to Nautilus 14.2.6 from Luminous 12.2.12 we are seeing the > following behavior on OSDs which were created with "ceph-volume lvm create > --filestore --osd-id <osd> --data <device> --journal <journal>" > > Upon restart of the server containing these OSDs they fail to start with > the following error in the logs: > > 2020-01-21 13:36:11.635 7fee633e8a80 -1 filestore(/var/lib/ceph/osd/ceph-199) > mount(1928): failed to open journal /var/lib/ceph/osd/ceph-199/journal: (13) > Permission denied > > /var/lib/ceph/osd/ceph-199/journal symlinks to /dev/sdc5 in our case and > inspecting the ownership on /dev/sdc5 it is root:root, chowning that to > ceph:ceph causes the osd to start and come back up and in near instantly. > > As a note these OSDs we experience this with are OSDs which have > previously failed and been replaced using the above ceph-volume, longer > running OSDs in the same server created with ceph-disk or ceph-volume > simple (that have a corresponding .json in /etc/ceph/osd) start up fine and > get ceph:ceph on their journal partition. Bluestore OSDs also do not have > any issue. > > My hope is that I can preemptively fix these OSDs before shutting them > down so that reboots happen seamlessly. Thanks for any insight. > > Our workaround (not on Nautilus but still) is to add to the pre-run systemd unit file pointed out like this: more /usr/lib/systemd/system/ceph-osd\@.service ... ExecStartPre=/usr/lib/ceph/ceph-osd-prestart.sh <more args here> then in that file, after it figures out what your journal should be (even if it is a symlink), do a chown to ceph:ceph more /usr/lib/ceph/ceph-osd-prestart.sh ... journal="$data/journal" chown --dereference ceph:ceph $journal so it has the correct perms before the filestore OSD gets started. -- May the most significant bit of your life be positive.
_______________________________________________ ceph-users mailing list -- ceph-users@ceph.io To unsubscribe send an email to ceph-users-le...@ceph.io