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

Re: slurpd vs ldapsync

flo@redflo.de wrote:

Just yesterday i set up slurpd for replication. The openldap
documentation points one to slurpd at many places. Example:
No word of an alternative. Could someone give some hints in that chapter
and problably in the chapter "14. Replication with slurpd"?

"Someone" probably could. Maybe that someone should be you. File an ITS.

The FAQ-O-Matic says that "multimaster replication is harmful". Is this
still true as you say that syncrepl works with that?

At a basic level, multi-master replication introduces a lot of uncertainty/potential inconsistency into the database. The X.500 service model makes some consistency guarantees that break in a multi-master environment, and such inconsistencies can certainly cause problems in many situations.

On the other hand, even single-master isn't perfectly consistent 100% of the time, due to the delay time in propagating changes from a master to all of the slaves. And it's always possible for a rapid sequence of changes to create the appearance of conflicting data. (E.g., a bunch of modifies in rapid succession, that toggle a particular attribute between a few values. The value that a reading client sees is really unpredictable, it just depends on luck, at what instant its search request is processed, relative to the modifications.)

The syncrepl protocol only offers loose consistency to begin with - it only guarantees eventual convergence. But what's important is that the convergence will actually occur. The multi-master support is the same - it's possible for conflicting views of the data to appear within small time intervals, but eventually it will all resolve.

Thanks for the nice piece of software.

-- -- Howard Chu Chief Architect, Symas Corp. http://www.symas.com Director, Highland Sun http://highlandsun.com/hyc Chief Architect, OpenLDAP http://www.openldap.org/project/