LVS
lvs-users
Google
 
Web LinuxVirtualServer.org

Re: ldirector and heartbeat

To: "LinuxVirtualServer.org users mailing list." <lvs-users@xxxxxxxxxxxxxxxxxxxxxx>
Subject: Re: ldirector and heartbeat
From: Horms <horms@xxxxxxxxxxxx>
Date: Wed, 28 Jun 2006 12:50:45 +0900 (JST)
In article <449DA02F.8090608@xxxxxxxxxxxxx> you wrote:
> I see that ldirector docs recommend starting it by heartbeat.
> 
> I run two LVS boxes, in active-passive mode that serve as a 
> load-balancer.  I've been running ldirectord on both boxes all the time 
> instead of starting it with heartbeat because it take about 2 minutes 
> for ldirector to test all my services and come to a fully functional 
> state.  So it does failover much faster if the secondary box already has 
> ldirectord running.
> 
> It was a way to start ldirectord  where it assumes the services are good 
> until  they are tested?  That would reduce the failover time if 
> ldriectord is started it on failover.  And the services all almost alway 
> there anyway.

No but that does sound like a good idea. If I prepare a patch are
you able to test it to see if it helps your problem?

> Is there a compelling reason for not running ldirector on both the LVS 
> boxes all the time?  I realize I have all those extra service checks 
> happening, but that seems worth it for the fast cut-over times.

For simple setups, the answer is usually that it saves network traffic
and load on the real servers. My personal oppinion is if that
ldirectord's checks are loading your network and real-servers, then
you have big problems, but hey.

For more complex setups, by which I am refering to ones where
real-servers and linux-directors are the same machines, then I'm
pretty sure you need to have heartbeat manage ldirectord or the whole
thing will fall over in a bit mess.

In short, if your linux-directors and real-servers are separate
machines, and you want to run ldirectord on both of them from boot,
then you are likely (unless I've forgotten something) to resolve
your problem.

> I'm trying to have heartbeat handle as little as possible to keep the 
> cut-over time as short as possible.  So I run all the services I can on 
> the secondary, so I don't have to wait for them to start.  Is there a 
> good argument for not doing this? 

There are arguments, on a case by case basis, but in general I agree
with your point there. Its just easier to document that you should
have heartbeat manage resources, as that always works.

> The truth of the matter is it only cuts over when I'm testing it any 
> way, but I'm sure the day will come...
 
Probably at a very invonvenient moment...

-- 
Horms                                           
H: http://www.vergenet.net/~horms/          W: http://www.valinux.co.jp/en/


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