rbd-mirror: force-promoted image will remain R/O until rbd-mirror daemon restarted

[dillaman]

http://tracker.ceph.com/issues/16974

[dillaman]

There is an unrelated memory corruption issue within rbd-mirror on master -- need to fix that first before testing and merging this PR

[trociny]

@dillaman Does #11104 fix the memory corruption issue you mention here and [DNM] prefix can be removed?

[dillaman]

@trociny Yes, so far so good.

[trociny]

@dillaman "force promote" functional test passes locally for me but regularly fails on teuthology:

http://pulpito.ceph.com/trociny-2016-09-17_15:04:54-rbd-wip-mgolub-testing---basic-smithi/
http://pulpito.ceph.com/trociny-2016-09-17_15:04:37-rbd-wip-mgolub-testing---basic-mira/
http://pulpito.ceph.com/trociny-2016-09-17_15:04:21-rbd-wip-mgolub-testing---basic-vps/
http://pulpito.ceph.com/trociny-2016-09-17_15:04:04-rbd-wip-mgolub-testing---basic-vps/

Note, I rose debugging level for rbd-mirror daemon in the qa-suite config, so you might find useful data in cluster1-client.mirror.*.log files.

It looks I don't see image replayer handle_promoted handler being fired on tethology.

[dillaman]

@trociny Looks like journal::JournalMetadata::allocate_tag doesn't send out an update notification, so a forced promotion on an idle image would never be detected.

[dillaman]

@trociny repushed

[trociny]

@dillaman Could you please rebase?

[dillaman]

@trociny rebase pushed

[trociny]

lgtm