Problem #2) If a MMR node is processing a change during which a slapd shutdown is initiated, it will update the contextCSN of the database but LOSE the related change (at least with a delete op), resulting in a database difference. This is reproducible in 2.4.47 as well (so this is not a regression).
Ok, this is just a timing issue, the ldapsearch to generate the DB data to compare was generated before all the nodes had finished processing the changes involved.
It might be better to do something like get the contextCSN from the initial master after the change has completed, and ensure that the remaining nodes have the same contextCSN value before generating the LDIF for comparison.
--Quanah -- Quanah Gibson-Mount Product Architect Symas Corporation Packaged, certified, and supported LDAP solutions powered by OpenLDAP: <http://www.symas.com>