Re: syncrepl questions

[Jonghyuk Choi <jongchoi@us.ibm.com>]

>It would be a problem with atomic 'update and sync' operations, but a
>separate 'sync this DN' LDAP operation after the update could return
>failure if one or more updates failed.  Then the user will have to
>handle the problem by rebuilding the databases or something if they have
>become inconsistent.  Thats what we do today when an update fails.  It's
>doesn't happen often.

If we consider syncrepl extension to perform such 'update and sync' operation,
a possible scenario is :
Suppose a master and N slave servers that are connected to the matser by
the refreshAndPersist mode sync search. The master can be made to wait
slave's update notification (as extended operation ?) after sending N persistent
search entry responses to the slaves. If all succeeded, the client will get success.
If failed, the master either retries the update, or terminates the failed persistent
connection and return failure to the client, according to the error type.
The disconnected server would reissue a persistent search when
it becomes healthy again. The LDAP content sync protocol will take care of the
changes made during the disconnected period.

------------------------
Jong Hyuk Choi
IBM Thomas J. Watson Research Center - Enterprise Linux Group
P. O. Box 218, Yorktown Heights, NY 10598
email: jongchoi@us.ibm.com
(phone) 914-945-3979    (fax) 914-945-4425   TL: 862-3979