Sorry about the confusion, I specified MirrorMode since that is a specific type of Multi-Master Replication and I wanted to be exact.
To answer Michael's questions:
1. Check whether cn=syncuser,dc=example,dc=com can really read *everything*
and no ACLs are preventing access to relevant CSN attributes etc.
My replication user can read everything. In fact, these deletes work when both instances of slapd are online when the delete happens. My problem only occurs when one of the instances of slapd is offline when the delete occurs.
2. Try to reproduce your issue with 2.4.40.
I will try this out and let you know what happens.
I am also looking for any resources that explain exactly how replication works and how conflicts are resolved in a multi-master configuration. For example, I know there are two phases of replication: the present phase and the delete phase. Could server1 be replicating the data for 'deletedUser' to server2 in the present phase before server2 has communicated to server1 that 'deletedUser' no longer exists in the delete phase? I don't believe this problem happens when I just delete 'deletedUser' (instead of deleting AND modifying). I know I am butchering the technical details a little bit, which is why I am curious how this situation is expected to be resolved.
Thanks,
Kevin