yes. it is user level modify. reading data from any of the mirror mode providers, modifying the data and then write back to any of the mirror mode providers.
I can now understand this model can cause inconsistency problem when the data is not in sync between the providers. This means we can not use Loadbalancers in-front of write endpoints? Or it suggests there is no way we can have multiple providers in a ldap cluster without the possibility of corrupting the data.
Better solution from the suggestions is to use two LBs. one for read only operation which uses round robin to select any of the servers(good option for this is HA Proxy). another LB for read+write operations, which uses ldap-backend to select one of the mirror mode provider until it fails.