[Date Prev][Date Next] [Chronological] [Thread] [Top]

Re: syncrepl questions

At 11:55 AM 9/24/2003, Hallvard B Furuseth wrote:
>Kurt D. Zeilenga writes:
>> Second, we just started this Syncrepl experiment.  I suspect
>> Syncrepl will prove to be better than the slurpd approach
>> in almost all deployments.  I also suspect that some deployments
>> will need a far more sophisticated replication engine than either
>> syncrepl and slurpd provide.
>For example, the University of Oslo needs update operations which do not
>return until all the servers have been updated, or maybe a 'sync this
>particular DN' operation which is performed after an update operation
>and only returns when that entry has been synced to all the servers.

Transactional consistent replication, yuk.

>Lacking that, we simply use independent master servers and run each
>update operation against all of them.

Of course, to ensure consistency between the masters would require
use of a two-phase commit transaction capabilities.