Howard Chu wrote:
Is there a doc somewhere that describes the replication protocol?
Here you are :
http://cwiki.apache.org/confluence/display/DIRxSRVx11/Mitosis
These pages are really very rough.
We encountered similar issues when developing the multimaster support
in OpenLDAP. Our solution was simply to assign unique server IDs to
each master, and record the last known modification for each known
server ID. This way local modifications don't overwrite the (lack of)
state from other servers.
Another issue to consider when bringing up a new replica - you may
want to disallow all queries completely, until the replica has fully
sync'd up. If you allow searches during this initial phase, a client
is likely to get a partial set of results. In some cases, that may be
worse than getting no result...
The pb is that if you have a lot of replication, it may not be a viable
approach. Also if you want to implement the 'paging control', as it's up
to the client to request the next results, you have no good solutions...
At some point, I think we should consider that a LDAP server is not like
a RDBMS system with a full transaction system, and you must assume that
dirty read is a standard mode. That does not mean we shouldn't consider
blocking incoming requests until a sync-up is done when a server is
restarted.
This is not a simple problem...
--
--
cordialement, regards,
Emmanuel Lécharny
www.iktek.com
directory.apache.org
