HDB Index Corruption

[Alejandro Imass <aimass@yabarana.com>]

Hi,

I searched the archives of openldap-software and openldap-technical
for the terms "hdb index corruption", "hdb index", and "idex" and got
zero results.

I have a simple master-slave setup and my indexes are getting
corrupted. We have software that ties RDBMS entries with the entryUUID
because the DN can change. Of course, we use moddn to avoid changing
the entryUUID. So here is the set-up:


old_ldap    ---->  new_ldap_master  ------> new_ldap_slave

We run a cron that scans the old_ldap and updates information on all
the records on the new_ldap_master. The information is mostly the
same, but the structures of the DIT differ a lot. Ok, so the first
cron updates information (attributes) that have changed in the old
ldap and updates the ldap master entries. Another cron searches some
info in a database and may perform many moddns if people have moved
from departments, etc. The new_ldap_slave is just a replica of the
first, using the methods described in the docs for MirrorMode
replication. All writes AFAIK are always being done in the master.

Anyway, the problem is that the HDB indexes seem to be getting
corrupted. That is, if I query for and entryUUID I get a DN and if I
query the DN I may get another entryUUID. If I reconstruct the indexes
the problem is gone. Has anyone had a similar problem? What may be
causing these problems? Is it too many wirtes? Is it common pratice to
reconstruct the indexes after the massive updates? Any comments
greatly appreciated !

Thanks,
--
Alejandro Imass