pthreads leaks handles and threads when threads use sockets

[Mark Pizzolato]

I've been using clamav's clamd under cygwin and noticed that over time the 
handle count as viewed with TaskManager seems to grow to arbitrary values. 
I used clamd's option IdleTimeout set to 600 seconds which dramatically 
reduced the growth rate of the Handle Count.  Of course clamd has many 
things going on that could contribute to handle leakags, so I tried to write 
a simple program to demonstrate the problem.

The attached program demonstrates the problem when sockets are used and that 
things look pretty clean when they are not.

There  seems to be both a thread leakage issue and a separate handle leakage 
issue.

  Invoking the program as:
      threadtest -sockets 0
 creates groups of 5 threads simultaneously.  Each thread merely prints 
something and sleeps, prints something else and exits.  This is repeated 10 
times displaying the process handle count between each iteration.  While 
running and watching with Task Manager, the process thread counts seems to 
start at 2 and reach 7 at times and then return to 2.  The handle count 
grows during the first iteration but stays flat thereafter.


  Invoking the program as:
      threadtest -sockets 3
 creates groups of 5 threads simultaneously.  Each thread merely prints 
something and sleeps, connects a socket to the main thread, passes a little 
data and closes the socket.  This socket business is repeated 3 times after 
which the thread prints something else and exits.  This is repeated 10 times 
displaying the process handle count between each iteration.  While running 
and watching with Task Manager, the process thread counts seems to start at 
2 and reach 14 during the each iteration and drops back to 9 between 
iterations.  The handle count grows significantly during the first iteration 
but seems to grow by 10 or 11 between each subsequent iteration. 
 The -sockets 3 argument controls the number of sockets each thread creates 
during its life.  The amount of thread and handle leakage seems to be 
independent of the number of sockets the thread uses during its lifetime (as 
long as the number of sockets used is 1 or greater).

The number of threads created simultaneously can be controlled by 
specifying -threads n as command arguments.  The number of threads leaked 
seems to be directly related to the number of thread using sockets 
concurrently.  running the program with -sockets 3 and -threads 10 causes 
the thread count to jump to 24 during each iteration and drop back to 14 
between iterations, while the handle count seems to increase by 10 or 11 
between each iteration identical to the case described in the previous 
paragraph.

I hope this test can help someone familar enough with cygwin internals to 
help get this problem under control.

- Mark Pizzolato 
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: threadtest.c
URL: <http://cygwin.com/pipermail/cygwin/attachments/20050128/b9576f70/attachment.c>
-------------- next part --------------
--
Unsubscribe info:      http://cygwin.com/ml/#unsubscribe-simple
Problem reports:       http://cygwin.com/problems.html
Documentation:         http://cygwin.com/docs.html
FAQ:                   http://cygwin.com/faq/