Infrastructure » Python GMN

Dave Vieglais wrote:

However, the implementation is incorrect and does not provide the expected functionality.

Oops...

The file descriptor should be held for the life of the application or alternatively, simply create a file, then treat existence of the file as the existence of a lock.

Adding global f at the top of the function should do the trick.

Just treating the existence of the file as a lock can be problematic if something happens that causes the script to bomb out without deleting the file, e.g., if it gets killed by the OOM.

Can't imagine what could go wrong declaring "f" as a global...

Better to declare the variable within the class, and give it a real name.

See http://blog.vmfarms.com/2011/03/cross-process-locking-and.html for a simple example that provides a reusable Lock class

I added the Lock class from the link Dave included and used that to lock the resource while the queue is processed::

def abort_if_other_instance_is_running(self):
single_path = os.path.join(tempfile.gettempdir(),
os.path.splitext(file_)[0] + '.single')
try:
lock = Lock(single_path)
lock.acquire()
# fcntl.lockf(f, fcntl.LOCK_EX | fcntl.LOCK_NB)
except IOError:
logging.info('Aborted: Another instance is still running')
exit(0)
return lock

def handle_noargs(self, **options):
try:
verbosity = int(options.get('verbosity', 1))
self._log_setup(verbosity)
logging.debug('Running management command: process_replication_queue')
lock = self._abort_if_other_instance_is_running()
self._abort_if_stand_alone_instance()
p = ReplicationQueueProcessor()
p.process_replication_queue()
finally:
lock.release()

I'm trying write to functional tests, but it requires makes calls using cnclient to a CN. I can verify that the lock works using the debugger. When I pause execution after the lock has been acquired and run another instance of process_replication_queue, the second instance pauses until the first is done.

I have written unit tests that test the basic functionality of the file locks and those work. However, I'm not sure how to functionally test process_replication_queue. I don't see where an instance of class Command is created so that _abort_if_other_instance_is_running() can lock the .single file

I see that 'Command' is a django requirement for writing custom django commands.

I've finished fixing the bug and posted the code and tests to subversion. I'll work with Roger to get it tagged and released to pypi.