LVS
lvs-users
Google
 
Web LinuxVirtualServer.org

Re: ActiveConn problem

To: "LinuxVirtualServer.org users mailing list." <lvs-users@xxxxxxxxxxxxxxxxxxxxxx>
Subject: Re: ActiveConn problem
From: Horms <horms@xxxxxxxxxxxx>
Date: Thu, 4 Dec 2003 12:47:12 +0900
On Wed, Dec 03, 2003 at 06:30:30PM -0700, Micah Abrams wrote:
> I have a very simple LVS setup using two Load balancers, two real servers
> and keepalived for the failover/monitoring.  The LVS topology is using
> Direct Routing.  The real servers are running a web server called Resin to
> serve up .jsp pages.  The problem is that ipvsadm is reporting hundreds
> (sometimes thousands) of connections when in reality each server is only
> recieving 10-15 simultaneous connections (20-30 accross the cluster).
> Keepalived is constantantly adding/removing the servers from the cluster and
> my mailbox is fillin up with notices.  Below is the contents of my
> keepalived.conf as well as the output from ipvsadm -L.  Please feel free to
> email me off the list should you need any other details.  Any help would be
> greatly appreciated.

It sounds a lot like your real servers are badly overloaded
and can't keep up with incoming connections. Keepalived is inteprating
this as a failure and thus takes the server out of the pool.
As no new connections are being recieved by the given real server
its load subsides, keepalived things it is alive again and
puts it back into the pool. 

LVS's connection count reflects connections that it routes from
end-users. If your machine can only accept 10-15 simultaneous
connections then the numbers are going to get out of sync.

Can you check the load on the real servers to see if they
are indeed overloaded. If so I suggest optimising whatever
code they are running or buying more hardware.

-- 
Horms
<Prev in Thread] Current Thread [Next in Thread>