|
Hi,
I prefer to analyse a little bit more the problem before updating the kernel
for this production cluster of LVS.
I've done the commands :
_____________________________________________
lvs1:~# ipvsadm
IP Virtual Server version 1.2.0 (size=4096)
Prot LocalAddress:Port Scheduler Flags
-> RemoteAddress:Port Forward Weight ActiveConn InActConn
TCP vip_squid:www wlc
-> realsquid17:www Route 100 23081 36519
-> realsquid16:www Route 100 23076 36654
We are saturday a low traffic day, the 27 Nov 2005 the number of active
connections was of 19988 and now it's 23081, the number of active connections
increase every day.
_____________________________________________
lvs1:~# ipvsadm -Lcn > result.txt
________________________________________________________________
The result.txt starts with :
IPVS connection entries
pro expire state source virtual destination
TCP 00:55 CLOSE xx.xx.119.192:1289 192.168.0.100:80 192.168.0.16:80
TCP 00:51 ESTABLISHED xx.xx.69.175:2163 192.168.0.100:80 192.168.0.16:80
TCP 00:48 ESTABLISHED xx.xx.20.154:1154 192.168.0.100:80 192.168.0.16:80
TCP 00:46 FIN_WAIT xx.xx.191.10:1612 192.168.0.100:80 192.168.0.17:80
TCP 00:42 SYN_RECV xx.xx.8.143:1114 192.168.0.100:80 192.168.0.16:80
TCP 00:38 FIN_WAIT xx.xx.89.3:3410 192.168.0.100:80 192.168.0.16:80
TCP 00:35 ESTABLISHED xx.xx.18.91:1085 192.168.0.100:80 192.168.0.17:80
TCP 00:34 ESTABLISHED xx.xx.40.17:40241 192.168.0.100:80 192.168.0.17:80
TCP 00:31 ESTABLISHED xx.xx.125.118:1315 192.168.0.100:80 192.168.0.16:80
TCP 00:31 ESTABLISHED xx.xx.196.181:4338 192.168.0.100:80 192.168.0.16:80
TCP 00:26 FIN_WAIT xx.xx.200.47:60364 192.168.0.100:80 192.168.0.16:80
...
_____________________________________________
lvs1:~# cat result.txt | grep -c TCP
119318
lvs1:~# cat result.txt | grep -c ESTABLISHED
46119
lvs1:~# cat result.txt | grep -c CLOSE
25776
lvs1:~# cat result.txt | grep -c FIN_WAIT
40583
lvs1:~# cat result.txt | grep -c SYN_RECV
6813
lvs1:~# cat result.txt | grep -c TIME_WAIT
27
_____________________________________________
lvs1:~# cat result.txt | cut -f2 -d" " | cut -f1 -d":" | sort | uniq -c
108248 00
4926 01
51 02
62 03
70 04
53 05
57 06
68 07
84 08
77 09
56 10
114 11
145 12
145 13
5161 14
That means :
108248 connections with an expire of 00:xx
4926 connections with an expire of 01:xx
51 connections with an expire of 02:xx
62 connections with an expire of 03:xx
70 connections with an expire of 04:xx
53 connections with an expire of 05:xx
57 connections with an expire of 06:xx
68 connections with an expire of 07:xx
84 connections with an expire of 08:xx
77 connections with an expire of 09:xx
56 connections with an expire of 10:xx
114 connections with an expire of 11:xx
145 connections with an expire of 12:xx
145 connections with an expire of 13:xx
5161 connections with an expire of 14:xx
Most of connections have less than 1 minute of expire if this value of expire
is minutes:seconds.
________________________________________________________________
Have you got an idea of the problem ?
Thanks for your work for LVS.
Marc
Horms <horms@xxxxxxxxxxxx> a écrit :
Kerscaven Marc wrote:
> Hello
>
> thanks for your answer.
>
> When I stop 1 real server : I switch off this real server.
>
> 30 minutes later I've again some active connections to this real server :
>
> TCP vip_squid:www wlc
> -> realsquid17:www Route 100 20034 69179
> -> realsquid16:www Route 0 15499 24624
That does sound like a problem. I know that there a lot of connections,
but some analysis of ipvsadm -Lcn might shed some light on the problem.
> I've a kernel 2.6.14.2 for real servers, I will test it on the director.
An excellent idea. 2.6.8 is a bit old, and there have been some LVS
fixes lately, at least one relating to connection expiry. I'm not
exactly sure if it was post 2.6.8 or not, but testing the latest
at least allows us to see if the problem is still present.
--
Horms
---------------------------------
Appel audio GRATUIT partout dans le monde avec le nouveau Yahoo! Messenger
Téléchargez le ici !
|
