frank.offermanns@caseris.de wrote:
Multimaster only works when slapdadd for initial filling is used with
-w
Is this a bug or does it work as designed?
Can't confirm your statement. You need to provide more information about
the
steps you followed that lead you to this conclusion.
In the thread "delta-syncrepl and mirrormode problem (2.4.29 and 2.4.30)" I already described how I came to the conclusion, that multimaster only works when using slapdadd with -w to initially fill the database. http://www.openldap.org/lists/openldap-technical/201203/msg00010.html and http://www.openldap.org/lists/openldap-technical/201202/msg00212.html
But to sum up: I have 2 Masters configured in mirrormode.
- both databases empty.
- I use slapadd without -w to fill one database. (Ldif has only 1400 bytes
)
- then do nothing else but start the first master and the second master
(which database is still empty).
This won't work. Both servers will initialize their accesslogs separately from their main DB since they are empty, and the two logs will have different timestamps. You need to make sure that the main DB and log DB are initialized together on at least one of the servers.
- Each slapd process now has 40 to 50 % CPU usage, although nobody is
doing anything with LDAP and this won't reduce no matter how long I wait. Configuration files, logs, etc are in the above threads.
Anyway I uploaded a zipfile with the data/aceesdata directories, a logfile, the configfile and my schema. Name is usage of slapadd with replication option -w.zip
If I use -w with the initial slapadd everything works fine.
are there cases in which this option could be harmfull?
Yes.
Could you please tell us, what are these cases?
For example, if you take a slapcat of a running database, and load it on another server, you should not use -w.