LVS
lvs-users
Google
 
Web LinuxVirtualServer.org

IPVS-2.4: weird state timers

To: "LinuxVirtualServer.org users mailing list." <lvs-users@xxxxxxxxxxxxxxxxxxxxxx>, Horms <horms@xxxxxxxxxxxx>, Julian Anastasov <ja@xxxxxx>
Subject: IPVS-2.4: weird state timers
From: Roberto Nibali <ratz@xxxxxxxxxxxx>
Date: Mon, 09 Jan 2006 15:32:27 +0100
Hello,

I've got a really strange case (which is very hard to reproduce) where
the state timer gets reset to 15 Minutes, which means the timeout for
IP_VS_S_ESTABLISHED. However the connection was in FIN_WAIT1 state or in
CLOSING state. I haven't been able to gather IPVS debug output regarding
this but wanted to ask if you guys have seen something similar in the
wild and if there could be a buglet in vs_tcp_state()?

I'll try to reproduce but it's really hard. You need a couple of hundred
clients that connect via HTTP 1.1 non-pipelined (with 2 socket caches)
to an apache2 server with a very low keepalive timeout and a low max
keepalive connection setting. If you have a couple of hundred active
connections and a couple of ten-thousand inactive connections and then
re-start apache with different keepalive settings, the clients will have
"lingering" sockets and IPVS kind of transforms some of the connections
into the IP_VS_S_ESTABLISHED setting the timer higher than the
persistency timer, which is definitely wrong!

Friends, we still have issues with the timers :). It's a productive
setup, so reproducing it is a bit tricky. It's not reproduceable in the
lab, unfortunately.

Regards,
Roberto Nibali, ratz
-- 
-------------------------------------------------------------
addr://Kasinostrasse 30, CH-5001 Aarau tel://++41 62 823 9355
http://www.terreactive.com             fax://++41 62 823 9356
-------------------------------------------------------------
terreActive AG                       Wir sichern Ihren Erfolg
-------------------------------------------------------------

<Prev in Thread] Current Thread [Next in Thread>
  • IPVS-2.4: weird state timers, Roberto Nibali <=