This happens when there is significant memory pressure (heavy network or block 
I/O exacerbates it) when you have swap disabled.  This is a long-standing 
kernel behavior issue, from my experience.  I've seen it with MySQL and other 
servers, in addition to Varnish.

That being said, perhaps Varnish is initiating a huge object expiry pass that's 
triggering this problem.  If you increase the cli_timeout, it may help prevent 
the child from being reaped, to probably dubious benefit.

Increasing /proc/sys/vm/min_free_kbytes (by 2-3x) may also help, but it's a 
dangerous beast.  I use 131072 for all of my Varnish instances, for this reason 
(16-64GB boxes).

You may also find that adding a few hundred meg of swap is the lesser of two 
evils.

Hope it helps,
-- 
Ken

On May 20, 2010, at 3:45 PM, Augusto Becciu wrote:

> We don't have swap enabled on these servers.


_______________________________________________
varnish-misc mailing list
[email protected]
http://lists.varnish-cache.org/mailman/listinfo/varnish-misc

Reply via email to