This should tell us exactly what connections were opened, if they were closed etc.
strace -f -s0 -v -o /tmp/syscalls <rsyslogd cmd> Can you check if this shows too many FDs being opened? May be /tmp/syscalls would be worth sharing? On Fri, Nov 14, 2014 at 12:20 PM, Rainer Gerhards <[email protected]> wrote: > 2014-11-13 23:05 GMT+01:00 David Lang <[email protected]>: > > > I would say that if the test is using hundreds of filehandles at one > time, > > something is going wrong, let alone thousands. > > > > > I have just checked the tests. Something is wrong here. It should use 20 > connections at maximum. Maybe connections cached by the OS during close > count against the ulimit as well? > > Note that there are some tests that explicitly use many connections -- > because that is what they test. manyptcp.sh is a sample of this. > > In any case, I'll add a valgrind test for imptcp_conndrop.sh, so that we > can see if there is an actual leak. That would be an explanation. > > Thanks, > Rainer > > David Lang > > > > On Thu, 13 Nov 2014, Thomas D. wrote: > > > > Hi, > >> > >> On 2014-11-13 21:28, David Lang wrote: > >> > >>> What else is going on? or is this one test hitting the 1024 open file > >>> limit? > >>> > >>> I suspect that your user is doing other things that are making it so > >>> that the > >>> rsyslog test is going over the limit. > >>> > >> > >> It is the "imptcp_conndrop.sh" test. I did some testing: > >> > >> I changed the Makefile so that only the "imptcp_conndrop.sh" test will > >> run (https://bpaste.net/show/af9eac014278). > >> > >> # ps -u thomas > >> PID TTY TIME CMD > >> > >> # lsof -u thomas /var/tmp/portage | grep -i thomas | wc -l > >> 0 > >> > >> # su thomas -s "/bin/sh" -c "id && \ > >> ulimit -a && \ > >> cd /var/tmp/portage/app-admin/rsyslog-8.9999/work/rsyslog-8.9999/ && > >> make check" > >> > >> It is very strange... I cannot find the EXACT value. > >> > >> It is sometimes passing the test with "ulimit -n = 1024"... but more > >> often it is failing. Same with 2048... sometimes it will pass, but > >> sometimes the test will fail. > >> > >> Higher value -> lower failure rate; Haven't seen a failure with 3072 > yet. > >> > >> > >> -Thomas > >> > > _______________________________________________ > > rsyslog mailing list > > http://lists.adiscon.net/mailman/listinfo/rsyslog > > http://www.rsyslog.com/professional-services/ > > What's up with rsyslog? Follow https://twitter.com/rgerhards > > NOTE WELL: This is a PUBLIC mailing list, posts are ARCHIVED by a myriad > > of sites beyond our control. PLEASE UNSUBSCRIBE and DO NOT POST if you > > DON'T LIKE THAT. > > > _______________________________________________ > rsyslog mailing list > http://lists.adiscon.net/mailman/listinfo/rsyslog > http://www.rsyslog.com/professional-services/ > What's up with rsyslog? Follow https://twitter.com/rgerhards > NOTE WELL: This is a PUBLIC mailing list, posts are ARCHIVED by a myriad > of sites beyond our control. PLEASE UNSUBSCRIBE and DO NOT POST if you > DON'T LIKE THAT. > -- Regards, Janmejay http://codehunk.wordpress.com _______________________________________________ rsyslog mailing list http://lists.adiscon.net/mailman/listinfo/rsyslog http://www.rsyslog.com/professional-services/ What's up with rsyslog? Follow https://twitter.com/rgerhards NOTE WELL: This is a PUBLIC mailing list, posts are ARCHIVED by a myriad of sites beyond our control. PLEASE UNSUBSCRIBE and DO NOT POST if you DON'T LIKE THAT.
