[lvs-users] is it possible to have ldirector and real cluster server on same physical machine?

Darren.Mansell at opengi.co.uk Darren.Mansell at opengi.co.uk
Fri Dec 3 11:33:52 GMT 2010


We have around 20 nodes of SLES 11 HAE servers running in 2 or 3 node clusters.

All of them are configured with local resources and ldirectord to load-balance to the local nodes with a highly available ldirectord resource being able to move on failover.

I can give more info if required.

-----Original Message-----
From: lvs-users-bounces at linuxvirtualserver.org [mailto:lvs-users-bounces at linuxvirtualserver.org] On Behalf Of Mrvka Andreas
Sent: 03 December 2010 10:43
To: 'lvs-users at linuxvirtualserver.org'
Subject: Re: [lvs-users] is it possible to have ldirector and real cluster server on same physical machine?


I've misstped the arp_announce entries. It should be 2 instead of 1.
After a reboot it looks quite good having only one ldirector.

I am satisfied at the moment.
Novell HAE (High Availability Extension) shipps pacemaker.
I will try to configure it managing ldirector service on both machines but only be explicitly active on one node.
And in the backup-case pacemaker should switch.

This sounds clear for me. But in fact, that the sysctl.cf file differs - how should this be updated?

Thank you very much till now.

-----Ursprüngliche Nachricht-----
Von: Mrvka Andreas
Gesendet: Freitag, 3. Dezember 2010 11:09
An: 'lvs-users at linuxvirtualserver.org'
Betreff: AW: [lvs-users] is it possible to have ldirector and real cluster server on same physical machine?

Hello Simon,

thank you for your fast response! I am happy for every feedback.
Actually I thought I can have more than one active ldirector (for failover and ldirector load balancing).

Well, okay let's do ldirector with one node.
I refer to your Ultramonkey Howto. 

----- On my first node I run ldirector and cluster node ---
eth0 is real IP and lo:0 is virtual IP. So I do have to arp_announce on lo:0 butnot arp_ignore.
net.ipv4.conf.all.arp_announce = 1 # enable the feature net.ipv4.conf.lo.arp_anounce = 1

----- On my second and n+1 node I run only cluster nodes ---
eth0 is real IP and lo:0 is virtual IP. So I do have to arp_ignore on lo:0 but not arp_announce.
net.ipv4.conf.all.arp_ignore = 1 # enable the feature net.ipv4.conf.lo.arp_ignore = 1

Is this correct?

Thank for directing me into right direction.

-----Ursprüngliche Nachricht-----
Von: Simon Horman [mailto:horms at verge.net.au]
Gesendet: Freitag, 3. Dezember 2010 10:39
An: Mrvka Andreas
Cc: 'lvs-users at linuxvirtualserver.org'
Betreff: Re: [lvs-users] is it possible to have ldirector and real cluster server on same physical machine?

On Fri, Dec 03, 2010 at 09:31:19AM +0100, Mrvka Andreas wrote:
> Hello list,
> I hope you can help me.
> I am using a simple 2 node cluster environment on SUSE Linux Enterprise Server 11 SP1 HAE.
> I only use ldiretord/ipvs and an apache service on both machines.
> My virtual IP address on interface lo:0 and real cluster IP on eth0.
> Regarding to the documentation of Novell it should be possible to have ldirectord etc. on the same machines.
> Do you agree?
> I ask because if I do a tcpdump I see lots of traffic from the client 
> browser to the server (http port) and the tcp session will never end.
> ( client side has finished the transfer but on the server side tcpdump 
> runs and runs....)

Hi Andreas,

if you want a pair of machines and you want one of them to be the active linux director + a real server, and you want the other to be the backup-linux director + a real server then yes it is possible, but you need to take special care.

1. On the active linux director the VIP should be on lo and
   On the backup-linux director the VIP should be on ethX
   - That is, you need to move the VIP when a machine moves
     from being active to backup and vice versa.
2. The backup-linux director must not have LVS rules in place for the
   service being handled by the active linux director.
   - That is, you need to have ldirectord running on only the active
     linux director. Or in other words, you need to start it
     when a machine becomes active and stop it when it becomes a backup.
3. You have to make sure that arp_ignore and arp_announce are set correctly.

Its a bit old, but I have some notes on this at http://www.ultramonkey.org/3/topologies/sl-ha-lb-eg.html

(In relation to ARP you want the "Debian" notes unless  you have a really ancient kernel.)

Please read the documentation before posting - it's available at:

LinuxVirtualServer.org mailing list - lvs-users at LinuxVirtualServer.org Send requests to lvs-users-request at LinuxVirtualServer.org
or go to http://lists.graemef.net/mailman/listinfo/lvs-users

More information about the lvs-users mailing list