Description of problem:


Version-Release number of selected component (if applicable):
12.2.0-2.el7cp

How reproducible:


Steps to Reproduce:
1. A file is written to master, with a delay set on slave(10 mins).
2. Before the delay the master goes down abruptly (stopped the mirror service).
3. The data written on master is not yet synced to slave.

image-2:
  global_id:   97551f32-2ac3-4df7-93a4-95a73d93b3e8
  state:       up+replaying
  description: replaying, master_position=[object_number=7, tag_tid=2, entry_tid=25599], mirror_position=[], entries_behind_master=25602
  last_update: 2017-09-23 16:15:02

4. Force promoted the slave to become primary.

5. After 10 minutes, data got synced on slave(entries_behind_master became 0)   
    During this time it was in up+replaying' state , after the delay state changed to 'up+stopped'(description: force promoted)

image-2:
  global_id:   97551f32-2ac3-4df7-93a4-95a73d93b3e8
  state:       up+replaying
  description: replaying, master_position=[object_number=7, tag_tid=2, entry_tid=25599], mirror_position=[object_number=7, tag_tid=2, entry_tid=25599], entries_behind_master=0
  last_update: 2017-09-23 16:25:54
  
6. Brought back the master, demoted it and did a re-sync.

$rbd mirror image status data/image-2 --cluster master
image-2:
  global_id:   f84a899f-909e-4e23-9428-ee31c5ca14fa
  state:       up+replaying
  description: replaying, master_position=[object_number=3, tag_tid=3, entry_tid=3], mirror_position=[object_number=3, tag_tid=3, entry_tid=3], entries_behind_master=0
  last_update: 2017-09-23 16:27:58

7. Checked the size of the images on both clusters.(wired snap was created)

Master:
$sudo rbd du -p data --cluster master
warning: fast-diff map is not enabled for image-2. operation may be slow.
NAME                                                                                                                                                               PROVISIONED USED 
image-2@.rbd-mirror.bc392583-5662-4721-8726-55573351bd8f.a96731e9-d9a0-4c7b-8b96-e844e8502421                                       1024M 152M 
image-2                                                                                                                                                                    
                                                         1024M    0 
<TOTAL>                                                                                                                                                               
                                                          1024M 152M 

slave:
$sudo rbd du -p data --cluster slave
warning: fast-diff map is not enabled for image-2. operation may be slow.
NAME                                                                                                                                                           PROVISIONED USED 
image-2@.rbd-mirror.bc392583-5662-4721-8726-55573351bd8f.a96731e9-d9a0-4c7b-8b96-e844e8502421                                           1024M    0 
image-2                                                                                                                                                                      
                                                            1024M    0 
<TOTAL>                                                                                                                                                                    
                                                             1024M    0

Actual results:
After re-sync on "master" cluster it failed delete the image and sync the image from the "slave" cluster

Expected results:
primary and secondary images should have of same size.

Additional info: