Kerscaven Marc <mkerscaven@xxxxxxxx> wrote:
> I use 2 directors with heartbeat in LVS-DR front of 2 real servers (squid).
> I use the Debian Sarge default kernel 2.6.8 :
> lvs1:~# uname -a
> Linux velvet 2.6.8-2-686-smp #1 SMP Thu May 19 17:27:55 JST 2005 i686
> and ultramonkey3 with ipvsadm 1.24+1.21-1.
> This cluster works well for 1 month, but I've seen I've dead connections on
> the director with lvs-rrd :
> specially in the month view we can see I've 60 000 active connections but in
> reality I've just 5300 active connections :
> The director show 19988+19991=21979 active connections.
> 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 19988 31468
> -> realsquid16:www Route 100 19991 31739
> But in the first real server I've just 2621 active connections.
> realsquid17:~# cat /proc/net/sockstat
> sockets: used 2623
> TCP: inuse 2621 orphan 97 tw 1932 alloc 2623 mem 668
> and in the second real server I've just 2653 active connections.
> realsquid16:~# cat /proc/net/sockstat
> sockets: used 2581
> TCP: inuse 2563 orphan 74 tw 2060 alloc 2563 mem 1036
> I can reboot the director, but I think it's not the solution
Are you sure its a problem? Does the number of active connections on the
linux-director grow without bound. If so its probably a bug, and I'd
recommend a newer kernel, like the 2.6.12 backport I made for sarge.
If not, it seems likely to me that LVS and the real servers are counting
connections in a different way.