LVS
lvs-users
Google
 
Web LinuxVirtualServer.org

Re: LVS ignoring weights and schedulers

To: "LinuxVirtualServer.org users mailing list." <lvs-users@xxxxxxxxxxxxxxxxxxxxxx>
Subject: Re: LVS ignoring weights and schedulers
From: Joe Stump <joe@xxxxxxxxxxxx>
Date: Fri, 19 May 2006 13:38:03 -0700
I'm using this to balance traffic to my MySQL slaves. I've found that the problem with persistency in this case is that web nodes connect so frequently that eventually LVS is always sending them to the same slave on the backend. After dropping persistency everything is working as expected.

--Joe

On May 19, 2006, at 1:24 PM, Graeme Fowler wrote:

On Fri, 2006-05-19 at 10:30 -0700, Joe Stump wrote:
Anybody see anything wrong with this?

IP Virtual Server version 1.2.1 (size=4096)
Prot LocalAddress:Port Scheduler Flags
-> RemoteAddress:Port Forward Weight ActiveConn InActConn
TCP  192.168.10.51:mysql wrr persistent 1
   -> 192.168.11.41:mysql          Masq    1      7          886
   -> 192.168.11.40:mysql          Masq    1000   1          444

How can a server with a weight of 1000 have 1 connection while the
one with a weight of 1 has 7? I've tried wlc and rr with no luck.

1 *active* connection. All that it indicates is that it's got a single
active connection when the other server has seven; that could be because
the connections being handled by .40 are very short-lived (or it
responds more quickly through being a better spec server), or because of
your persistence setting.

A little background as to what the clients of this service are might be a help here - if they're webservers, the connections are very likely not
to be long-lived.

If you can leave it running for some time and then check the stats &
rates using:

ipvsadm -ln --stats
ipvsadm -ln --rate

That way you can view the history of your system since it was last
started (or counters were zeroed). You'll get a better view with that of how the balancing is happening, rather than using the snapshot view you
just posted.

My only question is: persistence timeout of 1s? That seems somewhat
short as it means when a session idles out it'll only be kept sticky for 1 second, which to me rather negates the point of sending it back to the
same server again.

Graeme

_______________________________________________
LinuxVirtualServer.org mailing list - lvs-users@xxxxxxxxxxxxxxxxxxxxxx
Send requests to lvs-users-request@xxxxxxxxxxxxxxxxxxxxxx
or go to http://www.in-addr.de/mailman/listinfo/lvs-users


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