[Date Prev][Date Next] [Chronological] [Thread] [Top]

Re: 1.2.4 memory leak

At 11:20 AM 7/22/99 -0400, wesley.craig@umich.edu wrote:
>I'm running openldap 1.2.4 with sleepcat db 2.7.5 under Solaris 2.6
>using Solaris lwp.  I have a few performance tests that I run, one is
>~40 processes on another machine simulating the traffic generated in a
>typical day of using mail500 for the University of Michigan.  This test
>reliably exhausts all available memory on a machine with 1G of RAM.

>From previous reports I've read, I believe there is a problem
with detached threads not being automatically reaped under LWP
(I might be confusing this with SunOS LWP).  Besides leaking large
amounts of memory, it leaks thread handles.  You'll eventually run
out of threads.  You might scan -bugs and/or -devel for postings
on this subject.

>I'm sort of committed to fixing this and a few other little problems.

We gladly accept all the help we can get.  Please feel free to
contribute patches against 1.2 or -devel.  If you have questions,
please feel free to post to our -devel mailing list or to contact
<project@openldap.org>.   Guidelines for contributing can be found
at: <http://www.openldap.org/devel/contributing.html>.

>I was wondering if perhaps there are known memory leaks in 1.2.4, or if
>anyone has likely suspects.

Yes.  Many have been fixed in our devel code.  We haven't brought
many of the minor (and no so minor) fixes into the release engineer
as not to destabilize the 1.x.   One not so minor leak is in
ber_scanf.  It doesn't free everything upon error.  However, the
fix (rescanning the arguments) was believed might cause some
problems on some platforms.  Hence, we held the fix back for
further testing.