Swap exceed & Child not responding to CLI, killing it

[Hans Schou]

Hi

I have just recently upgraded to Varnish 3.0.2 64bit Debian Squeeze on a 
new server an encounter some problems Ive not seen before.

There is 32G physical RAM and I use a 50G backend storage file.

# ps -eo rss,vsz,comm,cmd | grep varnishd
32439516 96968568 varnishd   /usr/sbin/varnishd -P /var/run/varnishd.pid 
-a :80,:8080 -T localhost:6082 -f /etc/varnish/default.vcl -s 
file,/var/lib/varnish/v02/varnish_storage.bin,50G -t 0 -p sess_workspace 
262144 -p thread_pool_min 200 -p thread_pool_max 4000 -p 
thread_pool_add_delay 2 -p session_linger 100

After running 8-10 hours the server is starting to using swap. When 
swapping increases to about 40% the server becomes slow and it ends up 
with a kill in syslog:

varnishd[7632]: Child (31696) not responding to CLI, killing it.
varnishd[7632]: Child (31696) died signal=3
varnishd[7632]: Child cleanup complete
varnishd[7632]: child (9203) Started
varnishd[7632]: Child (9203) said Child starts
varnishd[7632]: Child (9203) said SMF.s0 mmap'ed 53687091200 bytes of 
53687091200

My expectation was that varnish would use ~ 50G virtual RAM (vsz) and 
that Linux would use 32G RAM mapped with the file but not using swap. 
How can I prevent varnish from using more than the 50G vRAM and not 
swapping.

Any hint or help much appreciated.

-- 
Venlig hilsen
Hans Schou
tel:46923438