Re: CLUSTERIP

From: Caleb Anthony <caleb.anthony#gmail.com>
Date: Tue, 9 Jun 2009 07:58:53 -0600


Yes, the IPAddr2 resource agent in Heartbeat supports CLUSTERIP:

http://www.novell.com/documentation/sle_ha/book_sleha/index.html?page=/documentation/sle_ha/book_sleha/data/ref_ocfagent_ipaddr2.html

A small side-note on the CLUSTERIP target - Cisco routers (maybe others too?) don't believe that the mechanism that makes CLUSTERIP work (unicast IP address with a multicast MAC address) is RFC compliant. And therefore will ignore this traffic. The only way around this is to add a static ARP entry on your router that points to the CLUSTERIP address. Also, this isn't a problem if your traffic never crosses a router.

On Mon, Jun 8, 2009 at 2:42 PM, Holger Just <haproxy#meine-er.de> wrote:

> On 08.06.2009 22:25 Uhr, Infos E-Blokos wrote:
> > Hi Holger and thanks for your answer.
> > is VirtualIP the same role as CLUSTERIP ?
>
> Actually, i have not worked with the ClusterIP approach yet.
> What I learned from http://www.linux-ha.org/ClusterIP is, that it is an
> extension to what is commonly called a virtual IP in HA speak.
>
> What I (and most other people) mean by "virtual IP" is just an
> additional IP address which is under control of a resource agent like
> IPAddr or IPAddr2. That IP is shared in the cluster and is assigned to
> exactly one of the cluster nodes. If one node fails, the IP is switched
> over. That IP is distinguishable from the static IPs of each of the
> cluster nodes.
>
> The ClusterIP approach adds a loadbalancing technique on top of that by
> applying special iptables rules (see the link).
>
> I think, what we need here is just the virtual IP. Additionally you
> might want to observe the haproxy process. As I found it extremely
> stable by now, I have not managed that yet (read: have not found time to
> look into that...)
>
> However, I'm sure, the list can help here :)
>
> --Holger
>
>
Received on 2009/06/09 15:58

This archive was generated by hypermail 2.2.0 : 2009/06/09 16:00 CEST