src/mds: fix MDSMap upgrade decoding #12097
Conversation
There was a problem hiding this comment.
Reviewed-by: Greg Farnum gfarnum@redhat.com
Not sure how we should test this, though.
Hammer MDSMonitors did not validate the state in beacons and would apply anything to the mds_info_t for a standby, such as setting it to down:dne. We must handle this case during upgrade. Fixes: http://tracker.ceph.com/issues/17837 Signed-off-by: John Spray <john.spray@redhat.com>
|
Pushed a rebase to see if that kicks jenkins into proceeding here |
|
@gregsfortytwo yeah, this made me notice how the ceph-object-corpus stuff seems to have gone stale, this is the kind of thing that could have easily slotted in there :-/ |
|
Hmm, we can probably push our own stuff into the object corpus whenever we want to. @liewegas, was there ever anything formal for building that up, or was it just something that people did and gradually got forgotten? |
|
@jcsp In http://tracker.ceph.com/issues/17837#note-18 you mentioned that you backported this to jewel. If you still have that backport around, could you open a PR and link it to the backport tracker issue http://tracker.ceph.com/issues/18100 ? |
Hammer MDSMonitors did not validate the state in beacons
and would apply anything to the mds_info_t for a standby,
such as setting it to down:dne. We must handle this
case during upgrade.
Fixes: http://tracker.ceph.com/issues/17837
Signed-off-by: John Spray john.spray@redhat.com