If you try to delete a cinder volume that only exists in the database and doesn't refer to a real SAN volume, the volume is deleted from the database, but no AMQ notification is sent. Therefore third party GUI's that watch for notification events do not know that they should refresh, so the user must refresh the GUI manually.
There are situations where volume creation fails during the scheduling phase, which results in a volume getting created in the database, but no real volume is created on the SAN. The volume in the database is left in Error state.
The reason for this issue is that in cinder/volume/api.delete(), it finds that there is no volume['host'], and so it just removes the volume from the database, but never attempts to delete the actual volume (because there is no actual volume). The notifications are only sent when it deletes the actual volume.
It seems like it should be ok to send an AMQ notification so that the GUI can catch the fact that a delete had started and then handle the fact that there is no host further down the line. Perhaps there is an easier way to just send a notification.
It looks like other places in the code they do a 'notify_about_volume_usage' with a delete.start and a delete.end . Is there a reason that isn't done before and after the db.volume_destroy in volume.api.delete() ? It appears adding the notifications there could resolve the user's problem.
Fix proposed to branch: master
Review: https:/