I do not believe the file is locked using the code above because the
second open closes the previous file pointer and unlocks the file. Can
you show the traceback of the problem you are having?
On Mar 11, 3:30 am, Corne <i...@cobra-scripters.nl> wrote:
> Hello,
>
> We were running into the following:
>
> (os: Linux - Apache with mod_wsgi)
>
> We have client code which does a number of requests:
> what happened is that some of these requests got 401-not authorized as
> response.
>
> The problem is that in globals.py the session file is opened (rb+) and
> the file is locked.
> The second request (which is handled at the same time from a different
> process) tries to do the same; but the file is still locked at that
> time.
> Expected result would be that we have to wait for the lock to free,
> but instead open throws an exception.
>
> As solution we now first open the file for reading, than acquire the
> lock; and finally open the file for reading/appending (rb+)
>
> Opening the file for reading will succeed, than the lock is acquired
> (or we wait until we can get the lock). When we have the lock there is
> no problem opening the file for reading and writing anymore..
>
> Patch:
> --- globals.py 2011-03-11 10:13:22.634740461 +0100
> +++ globals.py_new 2011-03-10 15:43:55.000000000 +0100
> @@ -270,9 +270,11 @@
> if response.session_id:
> try:
> response.session_file = \
> - open(response.session_filename, 'rb+')
> + open(response.session_filename, 'r')
> portalocker.lock(response.session_file,
> portalocker.LOCK_EX)
> + response.session_file = \
> + open(response.session_filename, 'rb+')
> self.update(cPickle.load(response.session_file))
> response.session_file.seek(0)
> oc = response.session_filename.split('/')
> [-1].split('-')[0]
