Re: status of multi-master replication?

[Elliot Metsger <emetsger@jhu.edu>]

Nir,

> You can try using a mechanism like heartbeat, using the multimaster
> capabality, coupled with a heartbeat mechanism and IP takeover will
> work, solve the conflict resolution, and rid you of the pain of
> attempting to promote the slave to the master (and the ensuing 'what
> hasn't been replicated to me' debate).

The problem with that implementation is that a slave slapd cannot accept 
writes, and must be promoted to a master slapd in some fashion 
(manually? programatically?).  One way to implement this (perhaps) would 
be that a missed heartbeat would trigger the stoppage of the slave slapd 
on the remaining cluster member, and start a different instance of slapd 
which is configured to act as the master for the same part of the DIT. 
I think the Linux HA project has the ability to run scripts when a 
heartbeat is missed.  In many large enterprise environments, this HA 
implementation would be passed over in favor of multimaster capabilities.

Elliot