Comment 12 for bug 1367941

Somebody correct me if I'm wrong, but I think the exception happening while we're holding the lock means the releasing log message gets skipped because it breaks out of the context before we get to do that. It should still be doing the right thing, but we can't log it because the semaphore's context manager is taking care of cleanup.

I think this is supported by the fact that the first acquisition is _never_ released, according to the log, which almost certainly isn't true because we then re-create the semaphore, which means nobody had a reference to it so it couldn't have been held through that whole time.

The direct link to the relevant section of the logs is http://logs.openstack.org/34/120934/1/check/check-tempest-dsvm-neutron-full/4c35ad6/logs/screen-n-cpu.txt.gz#_2014-09-11_23_50_26_641