Comment 4 for bug 1432387

I've did some research on how to implement that and I think this is impossible now. I'll try to explain this in more details.

First of all you need to realize that oslo_concurrency file locks are using fcntl (https://github.com/openstack/oslo.concurrency/blob/master/oslo_concurrency/lockutils.py#L261). This means that existence of lock file doesn't indicate that lock is taken. When lock is acquired fcntl.LOCK_EX is set on it and only this is the indicator of the fact that someones own the lock.

In the process of acquiring the lock file is opened in append mode by the process (https://github.com/openstack/oslo.concurrency/blob/master/oslo_concurrency/lockutils.py#L200). This creates the file if it doesn't exist. Lock file *isn't* removed when lock is released because there can be other processes waiting for the lock and removing the file would make it impossible to set fcntl.LOCK_EX on the file descriptor they own, because it would lead to a removed file. In my tests processes simply hanged.

This means that to remove the file we need to be sure that there are no other processes waiting. We can do that iterating through the processes and looking for their opened files. Unfortunately still a new process can open the file between our check and the removal and it will result in it hanging.

Therefore I think that it is impossible to safely remove the lock file in current implementation of oslo_concurrency. A possible solution would be to make oslo's FileLock open the file again on each try to acquire the lock so if file get's removed a new one is created.