Re: [lvs-users] Heartbeat and ldirector taking a long time to change ove

To: " users mailing list." <lvs-users@xxxxxxxxxxxxxxxxxxxxxx>
Subject: Re: [lvs-users] Heartbeat and ldirector taking a long time to change over.
From: Eric Renfro <erenfro@xxxxxxxxxxx>
Date: Mon, 21 Dec 2009 13:23:32 -0500
L.S. Keijser wrote:
> On Mon, 2009-12-21 at 11:21 -0500, Eric Renfro wrote:
>> Hello,
>> I'm trying to resolve a current problem I have setting up a pair of LVS
>> load balancing servers using heartbeat and ldirector under Gentoo.
>> I am using heartbeat 2.0.8 on two servers and the heartbeat and
>> ldirector setup is not very extensive but should be working better than
>> it is. I will provide complete configurations, minus IP's themselves,
>> but to explain the problem up front, the issues I'm having is rather
>> strange.
>> Our servers are named simply, network1, and network2, which I will use
>> to explain the issue.
>> How I am discovering these issues is when I shut down either network1 or
>> network2's heartbeat process, it successfully releases the IP and passes
>> it on to the other to take over. It does this rather quickly as
>> expected, however, when it brings up ldirector, that is when the
>> problems begin. We have two clusters of three webservers each, on both
>> http and https ports. On network1, it immediately brings up the first
>> cluster that was setup with all three RIP nodes active but inaccessible.
>> All the others are weighted to 0 under a weighted-based setup, otherwise
>> they are non-existent and going to the fallback server RIP initially.
>> For about 5-10 minutes the replaced heartbeat+ldirector server has heavy
>> CPU load with ksoftirq/0 and ksoftirq/1 being the culprits of the active
>> CPU load, atop confirms this by having 3 irq's showing at 200%, 100%,
>> and 100%, last 5-10 minutes.
> Just an idea, but why don't you run ldirectord on both network1 and 2?
> That way the LVS table will already set up when the failover occurs. If
> you run the sync daemons as well, chances are that nobody will notice a
> thing.

That is exactly what I am doing. ldirector and heartbeat are running on
both network1 and network2.

Heartbeat just decides whether and when to take over the VIP's when the
other node stops responding.

*Eric Renfro*
Software Developer, Inc
125 Excelsior Pkwy
Winter Springs, FL 32708

407-629-0900 ext 832

*_Join us for the 2010 World Tour_

-FITUR Madrid, Spain, January 20-24 Pavilion 8, Stand 8B29A
-Sabre Hospitality Solutions Customer Forum, Americas, Dallas, TX,
February 23-24
-ITB Berlin, Germany, March 10-14 Hall 10.1, Booth 111
-SoftBrands Hospitality User Forum, Scottsdale, AZ, March 16-18
-Sabre Hospitality Solutions Customer Forum, EMEA, London, UK, March 23-24
-Sabre Hospitality Solutions Customer Forum, APAC, Singapore, April 26-27
-HSMAI Revenue Management & Internet Marketing Strategy Conference,
Orlando, FL, June 21
-HITEC 2010, Orlando, FL, June 21-24
-World Travel Market, London, UK, November 8-11*

Attachment: signature.asc
Description: OpenPGP digital signature

Please read the documentation before posting - it's available at: mailing list - lvs-users@xxxxxxxxxxxxxxxxxxxxxx
Send requests to lvs-users-request@xxxxxxxxxxxxxxxxxxxxxx
or go to
<Prev in Thread] Current Thread [Next in Thread>