Jack,

Quick update: I was able to run a (relatively quick) test on a configuration
that hopefully later becomes part of the testbench (I am working towards that
goal and thought doing a manual check at that stage would not hurt). It is
not based your config, and it is relatively simple and straightforward.
Still, it uses TLS, and uses it in anon mode like you did. I used 4.4.2 on
the server. I processed around 500,000 message (not kept track of the actual
number). I did three such runs, all runing under the excellent valgrind[1]
memory debugger.

For none of the runs, valgrind reported any memory leaks. While this may not
be an ultimate indication, valgrind is *very* effective in finding leaks and
based on what you wrote I would have expected a small chunk of memory to be
lost per message. 

So the bottom line is that I currently cannot reproduce the bug. This may
change when I finally import your config. However, it would be useful for me
if you could run valgrind on rsyslogd in your environment and let me know if
valgrind reports any memory leaks. Doing so considerably slows down rsyslogd,
but given your load, I'd expect that it would be acceptable.

To run under valgrind is relatively simply. Valgrind is available as a
package on almost all distros. All you need to do is run valgrind and specify
your usual rsyslogd command line as the parameter. It is recommended to do
this in the foreground (see rsyslog troubleshooting doc).

So, for example, if you start rsyslogd usually by

/sbin/rsyslogd -c4 -...

you simply do

valgrind /sbin/rsyslogd -c4 -...

instead. By default, valgrind message go to stdout. At the end of the run
(kill `rsyslog.pid`) valgrind prints out memory leaks it detects. During the
run, it prints out any anomalies it sees.

Would be great if you could give that a try.

Rainer

[1] http://www.valgrind.org


> -----Original Message-----
> From: [email protected] [mailto:rsyslog-
> [email protected]] On Behalf Of Mr. Demeanour
> Sent: Sunday, November 08, 2009 6:24 PM
> To: rsyslog-users
> Subject: Re: [rsyslog] Rsyslog 4.4.2: server out-of-memory with gnutls
> 
> Rainer Gerhards wrote:
> > Thanks, but I wasn't specific enough. For TLS, I also need to client
> > config, because I need two machines to reproduce any issues (these
> > two instances are also the challenge for the current testbench, what
> > requires hopefully fewer than I expect changes ;)).
> 
> Perhaps someone could advise me on an alternative to resolving this
> problem by means of bug-hunting. While 4.4.2 is the version of rsyslog
> currently shipping in Debian squeeze, it's obviously a bit out-of-date
> from the point of view of the developers.
> 
> Well, the reason I'm using that version is precisely that it ships with
> Debian. Or to be more precise, I upgraded to squeeze because it ships
> with a version of rsyslog that includes encryption support. So perhaps
> I
> should just cut free, and install 5.x stable?
> 
> --
> Jack.
> _______________________________________________
> rsyslog mailing list
> http://lists.adiscon.net/mailman/listinfo/rsyslog
> http://www.rsyslog.com
_______________________________________________
rsyslog mailing list
http://lists.adiscon.net/mailman/listinfo/rsyslog
http://www.rsyslog.com

Reply via email to