Re: (ITS#3859) Strange replication and search problems

[Ludovic Drolez <ldrolez@linbox.com>]

Pierangelo Masarati wrote:
> Ludovic Drolez wrote:
>
> > Version: 2.2.23
> > OS: Debian 3.1 sarge
> >
> > Hi !
> >
> > We are currently replicating an openldap 2.0.23 directory to a 2.2.23 
> > one, and
> > modified entries have a corrupt index on 'objectClass' (we are using 
> > the ldbm
> > backend).
> >
> > Steps to reproduce:
> > 1- make an initial slapcat;slapadd to synchronize the two servers.
> > 2- 'ldapsearch -x -h slave objectClass=posixAccount|grep toto' returns 
> > the entry
> > (there are less than 500 entries)
> > 3- modify something on the master
> > 4- the same ldapsearch now returns nothing (but a 'ldapsearch -x' 
> > without a
> > filter on the objectClass is ok).
> > 5- if I remove 'index objectClass eq' from the slave server, then the
> > 'ldapsearch' is ok.
> >
> > So it seems that the replication corrupts the index. I can also 
> > reproduce the
> > bug with slapd 2.2.26.
> >
> > Is this bug related to 'Software Bugs/3406' ? Or is it only a replication
> > problem ?
>
>
> I note that replication from 2.0 to 2.2 should be impossible because the 
> 2.2 slave expects stuff like structuralObjectClass, entryUUID, entryCSN.
> I don't see how ITS/3406 may be related to this.
I thought that only the contrary was impossible (2.2 to 2.0), and I don't get 
syntax errors during the replication. How could a replication corrupt the index?

Cheers,

PS: I'm not on the list anymore.

--
Ludovic DROLEZ                              Linbox / Free&ALter Soft
www.linbox.com lrs.linbox.org	              tel: +33 3 87 50 87 90
152 rue de Grigy - Technopole Metz 2000                   57070 METZ