Comment 8 for bug 1648885

Here we see both volume and snapshot deletion operations begin, and then we see snapshot deletion fail before volume deletion completes. Given the current logic, this is expected behaviour. So we have two immediate choices:

1. Postpone the snapshot deletion in the driver until the final volume finishes deletion
2. Change the tempest test to not submit both operations concurrently.

The third options would be:

3. Change the cinder API to reject busy snapshot delete requests.

This one has a larger impact and will require project (cinder) agreement. I think it all depends on how many patches are being effected by this issue.