Re: OpenLDAP Cluster+Alteon

[John Morrissey <jwm@horde.net>]

On Mon, Feb 11, 2002 at 03:31:46PM +0800, Syahrul Sazli Shaharir wrote:
% Single master with chaining is not available in OpenLDAP, so the only
% alternative is single master with referrals:-
% 
%          LDAP clients
%            |        |
%         ___|________|___
%        |_____Alteon_____|
%          |   |    ||||
%          |   |    ||||
%          |   |    ||||
%          o   o    oooo
%         l1  l2    slaves
% 
% l2 is cold standby master, eg l1 and l2 maintains heartbeat, if l1 down,
% l2 automatically wakes up and takes over, and vice versa. Both runs slurpd
% and updates all the slaves as required. Not as pretty as I want [ Masters
% not scalable, and LDAP clients need to handle write referrals correctly ],
% but should work.

I'm doing the same thing with a couple AceSwitch 184s. It's been working
well in a production environment for about six months now.

One gotcha: be prepared to write your own health check for the LDAP service.
My memory is a bit fuzzy since it's been a while since we used a stock
health check on our LDAP servers, but our experience prior to that with the
built-in "TCP connection" health check was really poor. When we used the TCP
connection health check on our LDAP servers, we found that the AceSwitches
would churn real LDAP servers out of the pool on an alarmingly regular
basis. I heard an offhand remark from a co-worker that the TCP connection
health check wants an immediate response (e.g., a welcome banner) from the
host when it connects, not sure if that's totally true.

Other than that, it works well.

john
-- 
John Morrissey          _o            /\         ----  __o
jwm@horde.net        _-< \_          /  \       ----  <  \,
www.horde.net/    __(_)/_(_)________/    \_______(_) /_(_)__