LVS
lvs-users
Google
 
Web LinuxVirtualServer.org

Re: local node on failover server.

To: "LinuxVirtualServer.org users mailing list." <lvs-users@xxxxxxxxxxxxxxxxxxxxxx>
Subject: Re: local node on failover server.
From: Jan Klopper <janklopper@xxxxxxxxx>
Date: Wed, 02 Mar 2005 15:16:03 +0100
Hmm,

I think you'r right,

I might need to buy two new (via-eden????) servers then, maybe i can place two of these 1u servers back to back in the rack as load balancers.
Not really sure if i want to use via-eden's as servers though.

Anyway, would it realy be that error prone?
I agree that you'd want to make sure the systems stays as simple as possible, but hearthbeat could just be changed to add/remove noarp of the VIP on the failover server, instead of adding/removing the VIP at all right?

greets
Jan

PMilanese@xxxxxxxx wrote:

I'd strongly agree.
The system is designed for redundancy, and subjecting the load balancers
(backbone to the operation) to unnecessary tasks. In my experiences, LVS
does not need to be on a robust setup, so maybe you want to pull the
existing servers into the server pool and replace LVS with something
paired down. I use a couple of old Dell 350's for LVS (well.. about 3
years old). I have 5 6650's blown out on the backend, and had no downtime
on the frontend for 3 years and change (with the exception of that silly
NE power outage). Once in a while, I test the backup box just for
reassurance.
Split the roles. Keep it simple. It will pay off.

P

-----Original Message-----
From: lvs-users-bounces+pmilanese=nypl.org@xxxxxxxxxxxxxxxxxxxxxx
[mailto:lvs-users-bounces+pmilanese=nypl.org@xxxxxxxxxxxxxxxxxxxxxx] On
Behalf Of Mack.Joseph@xxxxxxxxxxxxxxx
Sent: Wednesday, March 02, 2005 8:31 AM
To: LinuxVirtualServer.org users mailing list.
Subject: Re: local node on failover server.

Joseph Mack PhD, High Performance Computing & Scientific Visualisation
LMIT, Supporting the EPA Research Triangle Park, NC 919-541-0007
Federal Contact - John B. Smith 919-541-1087 - smith.john@xxxxxxx

lvs-users-bounces@xxxxxxxxxxxxxxxxxxxxxx wrote on 03/02/2005 08:03:25
AM:


As all 5 machines are pretty beefy, and i use LVS-dr, i wanted the
balancers to also be realservers. I used the local node "feature"
(why on earth is this called a feature),
we're up for suggestions. I don't like all of the names used
in LVS but haven't been able to think of anything better myself.

If i add the VIP with noarp to the load balancer,
hearthbeat would not
be able to setup the VIP when it needed to become the loadbalancer,
so is there any way to tell hearthbeat to toggle the noarp
switch on the load balancers instead of adding/removing the VIP?
sure. It's a nightmare of scripts that will probably fail under some
condition you don't expect and will be impossible for someone else
to debug. An idle server is a small price to pay for robustity.
In the early days of LVS we thought it would be trivial to promote
a realserver to be a director when the director failed, by running
a few setup scripts. While simple in concept, it turned out that it
was simpler just to have a standby director just parked there
waiting to take over. Brute force and low tech usually win when
reliability is required.

Joe





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