Memory leak in on Windows?

Mark Hammond mhammond at
Thu Aug 22 04:26:38 CEST 2002

Mark Moales wrote:

> Today I'm running a test in which my client issues 6000 calls instead of
> 600 and has a delay between calls of a half second.  In perfmon (I'm
> monitoring private and virtual bytes for the process), I see a steady
> increase in the number of private bytes used by the process up until the
> process terminates.

Well this is a leak :)  The best thing you could do would be to code the 
smallest, tightest loop that leaks.  If you can reduce it down to one or 
2 calls that must be in the loop to demonstrate the leak, then it should 
be fairly easy to track down.

If you have access to a debug build of Python, sys.gettotalrefcount() 
will help determine if it is simply a Python reference count bug, or a 
deeper leak.

> My next step is to write my own Transport that just uses sockets to send
> the XML instead of HTTP to see if that makes any difference.

Almost certainly will (I hope ;)  sockets have been around for too long 
and used for too many important things for an obvious leak to still exist.


More information about the Python-list mailing list