Bug? Barage of hits leads to failure creating worker threads / stats tracking

Ray Barnes tical.net at gmail.com
Mon Apr 13 07:57:59 CEST 2009

On Sat, Apr 11, 2009 at 5:48 PM, Artur Bergman <sky at crucially.net> wrote:

> I've never seen it do worker threads not created.
> Are there any limits on number of threads?

Apparently there are; thanks for pointing me in the right direction.  I
found a C program that attempts to spawn threads and lets you know at what
point it hits an error -
http://people.redhat.com/alikins/tuning_utils/thread-limit.c - it reports
that I can't open more than 383 threads.  The question is why.  Here's what
I've done thus far:

1) Recompiled glibc per
http://people.redhat.com/alikins/system_tuning.html#threads - the definition
of  PTHREAD_THREADS_MAX is tied to the value in /usr/include/linux/limits.h
so I adjusted that value, installed the source RPM, rebuilt all glibc RPMs
and installed using 'rpm -Uvh --force' to overcome pre/post installation
errors within the RPM (hopefully that did what it was supposed to).

2) Set /proc/sys/kernel/threads-max to 65535 (was 3000ish before), no change

3) Set /etc/security/limits.conf to "* soft nofile 1024" and "* hard nofile
10240" and added "session required /lib/security/pam_limits.so" to
/etc/pam.d/login with no change, per the advice at
http://www.mail-archive.com/java-linux@java.blackdown.org/msg15247.html where
the poster indicates he did not have to recompile glibc to do this

I've tried the same C program above on a few other Linux boxes and they all
seem to be somewhere between 200 and 383 allowed threads.  The first obvious
solution would be to dump Linux and use FBSD - a direction i'll look into in
the future.  But for now we're stuck on Linux.  Any ideas?

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://www.varnish-cache.org/lists/pipermail/varnish-dev/attachments/20090413/614aba9f/attachment-0002.html>

More information about the varnish-dev mailing list