OK, by editing zentyal-core.postinst script, i've found the problem.
I was trying to do the final migration on my servers, but it didn't work again.
I still must understand why on the cloned server it's working, maybe it has something to do with LDAP sync.
So I've created the custom log file and here is the error:
usermod --home /var/lib/zentyal/ ebox
usermod: user ebox is currently logged in
Obviously I've manually modified the passwd file, and commented the line in the script to let the it continue, and it worked.
But, even if my servers are all up and running, when this server (slave mail) is going to update ldap configuration, it give me the error:
Upgrading ldap attributes (quotas from mail and users)...
Cannot restore slave machine when master is down: xxx.xxx.xxx.xxx
Here I cannot guess what is not working: maybe I have to launch the script on slaves before than master?
If someone need a log file, I will redo the migration when I can. This is a production server, and I cannot test on it.