Re: Federating haproxy servers

Hi Jeff,

On Fri, Jun 26, 2009 at 12:03:07PM -0400, Jeff Buchbinder wrote:
> Are there any plans to allow federating of haproxy servers, so that they
> can (for example) cut down on the number of duplicate health checks?
> Currently we run two carp'd OpenBSD firewalls running haproxy, and the
> second firewall currently duplicates all health checks for our
> applications, which is less than ideal.

I had some plans to externalize health checks so that a central process may inform several components about server status. But it's not that easy and it has some drawbacks :

• it can only work when both confs are in sync. This means that it will not be possible anymore to use the backup for pre-prod config testing.
• I want haproxy to interact with the checker, for instance, speed it up when it detects an error 500. It becomes difficult with such an architecture.
• the checker must be replicated so that it does not become a SPOF. How to keep them in sync and to ensure only one is running the tests ? And how to ensure the second one will start from the same state ?
• how do you know that all the servers are down and that it's not in fact the only instance running checks which is having trouble ?

If you think about it longer, you'll find a lot of issues when trying to share the servers' state. In the end, a health-check must be extremely cheap by design so it must not be a problem to have them duplicated. At least, it's generally less of a problem than all the potential problems you would face if they were not duplicated.

Regards,
Willy Received on 2009/06/28 16:08