Andreas Haubod wrote:

Hello,

at the moment, I have running two LDAP-Server (2.4.23) on two sites (AP and

Europe) with multi master replication. All is running very well.

On both sites will be written from applications to the local LDAP-server and

the complete database is replicated to the other site (syncrepl/push).

A third LDAP-server (read-only replica) on one site is used to make a backup

hourly. The LDAP is stopped an the database is saved with slapcat.

Now I want replace one of the LDAP-Server because of better Hardware.

So I know, I have two choices:

1) Only setup the new LDAP-Server with a empty Database and wait till the replication is finished
2) Fill the Database with one of the hourly backup and start the server and wait till the replication is finished.

What will happen, when I use method 1) and the application write during the

(initial) replication to the (for explamle only 2% filled) Database?

(I checked the initial replication. It needs about one hour because of the

old slow hardware.) Does this work without expecting any problems or should I avoid the writing till the initial replication has finished?

Assuming a correct configuration, this will work fine. However, it is not the fastest way to go.

When I use method 2) and for example the saved database is 10h old and

during this time are many (>10000) changes made in the productive system? Is then this method then recommended or is method 1) better in this situation?

It's always faster to start with a recent snapshot of the database and then let it catch up from there, than to start from zero.

Are there any comparison or recommendations for backup/recover and migration

in such multi master setups?

A second Question I have. Is it possible, to mix the openldap (minor)

versions (eg. 2.4.23 and 2.4.25)? I ask this, because I want update the servers on different times.

Yes.

Thanks for help and the great openLDAP !!!

Andreas