This is not a patch suggestion but may be it helps to better see the problem's origin :
I applied the following patch to syncprov.c and I got good results. The concerned objects are sent with a NULL cookie but they are replicated well. The contextCSN is made up by the subsequent objects.
Thanks and best regards Sylvain Thomas.
============================================
--- syncprov.c.orig 2008-09-29 11:08:36.000000000 +0200 +++ syncprov.c 2008-09-29 11:10:33.000000000 +0200 @@ -1642,8 +1642,6 @@ } } else { /* internal ops that aren't meant to be replicated */ - ldap_pvt_thread_rdwr_wunlock( &si->si_csn_rwlock ); - return SLAP_CB_CONTINUE; }
/* Don't do any processing for consumer contextCSN updates */