Comment 4 for bug 1427060

So this behavior can have pretty bad consequences, if we fail to delete the created BDM.

If in normal operation once the BDM record is created reserving the device name on the compute host, we will attempt to mark it as 'attaching' in Cinder. This fails if the volume is already attached, and as part of the cleanup we delete the BDM record we created.

If something goes wrong with the reply message however (there is a problem with the network, or for some reason the API worker that accepted the attach request goes away in the meantime are some of the problems I can think of), the cleanup code never runs and we are stuck with an invalid BDM, which we have to clean up manually. This is not ideal obviously but is made worse, by the fact that when looking for the BDM record, we commonly select records only by instance and volume IDs which will not be unique in this case, and Nova may attempt to use the wrong one by chance which causes things like detach, migrate etc. to fail until manual cleanup