|
second attempt at getting some help. sorry, but I'm desperate....
For some reason the connection table just keeps growing. I just rebooted a
box when memory used by kernel >400MB large, most of it by ip_vs_conn:
# grep ip_vs /proc/slabinfo
ip_vs_conn 3212718 3302988 128 31 1 : tunables 120 60
8 : slabdata 106548 106548 0
Last weekend one crashed when the full 1Gig was used up completely by the
ipvs connection table...
I'm balancing 4 DNS servers and setup my virtual services as follows:
dns_vips= [9 IPs]
dns_realservers= [4 IPs]
for service in $dns_vips
do
$IPVSADM -A -t ${service}:53 -s wrr
$IPVSADM -A -u ${service}:53 -s wrr
for real in $dns_realservers
do
$IPVSADM -a -t ${service}:53 -r ${real} -g -w 10
$IPVSADM -a -u ${service}:53 -r ${real} -g -w 10
done
done
So that's 9 VIPs served by 4 realservers.
Kernel is 2.6.10 (kernel.org), all ipvs stuff is compiled in. ipvsadm is
version 1.24. Running in a dual xeon...
Incomming packet(/query) rate is approx 2800/s
Other than the fact that kernel memory useage just keeps growing & growing
& gowing things are working fine...
Does anybody hava any ideas on what could be the problem?
TIA,
Mark.
|
