[Date Prev][Date Next] [Chronological] [Thread] [Top]

Re: Syncrepl not replicating entire tree

>>My other question is: since you counted the DNs in both slapcats, can you
>>check if any of the entries you cannot see has "glue" objectClass?
> Yes, they do. Considering the same ex.:  'dn:
> ou=EE,ou=People,dc=iitb,dc=ac,dc=in' on my consumer has a "glue"
> objectClass.
> [ldif]
> dn: ou=EE,ou=People,dc=iitb,dc=ac,dc=in
> structuralObjectClass: glue
> objectClass: top
> objectClass: glue
> [\ldif]
> But the same entry on the provider does not have a "glue" objectClass.
> [ldif]
> dn: ou=EE,ou=People,dc=iitb,dc=ac,dc=in
> ou: EE
> objectClass: top
> objectClass: organizationalUnit
> structuralObjectClass: organizationalUnit
> entryUUID: 9ea7320a-ec3b-1028-8a25-eb9297806c56
> creatorsName: cn=Manager,dc=iitb,dc=ac,dc=in
> createTimestamp: 20041227101255Z
> entryCSN: 20041227101255Z#00000a#00#000000
> modifiersName: cn=Manager,dc=iitb,dc=ac,dc=in
> modifyTimestamp: 20041227101255Z
> [\ldif]

OK, that's the issue.  I do not know the reason, but your consumer is
receiving the children before the parent; as such, it's creating the
parent as a "glue" entry, i.e. a placeholder for the real object in case
it comes in.  Usually, this only happens if some entries are intentionally
not replicated (e.g. by specifying a filter in the syncrepl directive, or
by means of ACLs, or because a sizelimit was hit and the entries, for some
reason, were returned in a non-consecutive manner or so).  In any case,
you should investigate in that direction.  Do entries listed by slapcat
appear in a consecutive manner (e.g. parents before children)?

I recall there was an issue with this earlier in 2.2, because in some
cases "glue" entries were not replaced by the actual entry as soon as this
became available.  This was fixed in 2.2.21, so it should not be an issue
for you, since you're using more recent software.


Pierangelo Masarati

    SysNet - via Dossi,8 27100 Pavia Tel: +390382573859 Fax: +390382476497