mds: fix FSMap upgrade with daemons in the map #8073
Conversation
This would trigger a sanity() assertion because you'd have standbys in a Filesystem::mds_map, instead of in standby_daemons. Also, fix handling the enabled flag. Signed-off-by: John Spray <john.spray@redhat.com>
…ent FS Signed-off-by: Greg Farnum <gfarnum@redhat.com>
The upgrade path was asserting that FSMap::filesystems was empty. This would be violated if FSMap was initially loaded from disk in old format, and then also received off the network (but with a newer epoch) in old format. Signed-off-by: John Spray <john.spray@redhat.com>
jcsp
force-pushed
the
wip-fsmap-upgrade-jewel
branch
from
cf51f8f
to
bdcff15
Compare
jcsp
changed the title
|
Tested this on my home cluster while upgrading, worked there. |
|
what is your home cluster? On Sat, Mar 12, 2016 at 3:44 PM, John Spray notifications@github.com
|
|
@jcsp: Nice little cluster! |
|
edit: n/m, wrong PR |
|
@jcsp this looks better than my simple upgrade switch, but we also need gregsfortytwo@09fce2d or similar to make "fs new" an idempotent command. We can throw that in here if the patch looks good to you. :) |
…ation If we check for creating an FS with one already existing first, we will reject (with EINVAL) duplicate attempts to create the same FS with the same params. That makes it non-idempotent, which is bad! Signed-off-by: Greg Farnum <gfarnum@redhat.com>
|
Rebased and cherry-picked in greg's idempotency fix |
|
I had the pre-rebase version in the (all-green!) http://pulpito.ceph.com/gregf-2016-03-15_04:07:43-fs-greg-fs-testing-314---basic-mira/ And in upgrade testing with ceph/ceph-qa-suite#875 it seems to be doing as well as current master (http://pulpito.ceph.com/gregf-2016-03-15_12:49:59-upgrade:infernalis-x-greg-fs-testing-314---basic-mira). |
mds: fix FSMap upgrade with daemons in the map Reviewed-by: Greg Farnum <gfarnum@redhat.com>
This would trigger a sanity() assertion because
you'd have standbys in a Filesystem::mds_map, instead
of in standby_daemons.
Also, fix handling the enabled flag.
Signed-off-by: John Spray john.spray@redhat.com