I'm seeing something very similar except on creates. Specifically I have a script that fires off 10 threads, Each does a cinder create, gets the ID from the create call and then waits for the volume to become available. Normally all 10 call to 'create' return within the same second or 2. But every once and awhile there's a 50 second delay and I then also see the deadlock message in the API log, so it's not strictly related to a delete.
I'm seeing something very similar except on creates. Specifically I have a script that fires off 10 threads, Each does a cinder create, gets the ID from the create call and then waits for the volume to become available. Normally all 10 call to 'create' return within the same second or 2. But every once and awhile there's a 50 second delay and I then also see the deadlock message in the API log, so it's not strictly related to a delete.