Re: How to do a HA redundant

On Sun, Feb 07, 2010 at 10:36:24AM +0100, Malte Geierhos wrote:
> Hi,
>
> There's a OCF Agent for Linux Pacemaker / Heartbeat available.
> Just add your IP address and HaProxy as a ressource to your cluster.
>
> By using OCF Agents for the IP address and Haproxy the process
> can be integrated with a monitor action - and thus will failover if
> its not running on
> one node.
> You can add some sort of availability test by defining ping(d)
> ressources to monitor
> the connectivity of your LB nodes.
>
> However its necessary to have redundant connectivity between your nodes,
> to have reliable fault dedection and to avoid split-brain situations.

Keepalived has the advantage of using the traffic links for monitoring, so you don't have the split brain problem. I really don't like fail-over solutions which require additional links and which don't really monitor what is essential. Heartbeat is good for clusters with shared FS and such, but for just sharing a stateless service such as a proxy, the VRRP provided by keepalived is a lot better, simpler and more reliable (the same is true for ucarp BTW). The real advantage of keepalived over ucarp is that it can adjust its priority depending on interface states, process states, and that it can even ping haproxy to check if it's doing well, and/or if it sees some servers alive. All those information can be combined as you like to decide how they affect priorities, thus who's the master and who's the backup.

I'd say that I really like how haproxy and keepalived interact together!

Regards,
Willy Received on 2010/02/09 18:40