IBM team raised the following concern. The related support case is 04278369.

https://access.redhat.com/support/cases/#/case/04278369
 
Quote Unquote
============
I want to summarize the multipath/volume issues and our progress so far.
Right now we are looking at https://access.redhat.com/support/cases/#/case/04278369
also this article explains the issue, but does not provide a solution other than
"Cinder DB surgery" - https://access.redhat.com/solutions/7131782
 
The issue comes down to a large number of volumes with duplicate records in Cinder DB.

First we started with a clean Cinder DB, no duplicates, we also ensured that there are no actual duplicate mappings
between the SAN and the Compute nodes. Over time, due to certain activities (I suspect mostly due to VM migrations) these duplicate records
are being accrued in Cinder. Once a VM has got these duplicate records, it seems operations with it(server hard restart, migrate etc) start to fail.