Multi-Master Replication

[Jon Miner <miner@doit.wisc.edu>]

So, I've gotten MM replication set up and running in test (yes, I know,
it's experimental.)  In our stress testing (about 50,000 changes to the
same object talking to the two servers) we managed to wind up in a
situation that we thought should be impossible: LDAP1 had the last
update from LDAP2, and LDAP2 had the last update from LDAP1 (They were
different, and had processed eachother's last change.)

In thinking about it, this is an easy situation to get in:
    1. Change to LDAP1
    2. LDAP1 puts it in replog
    3. Change to LDAP2
    4. LDAP2 puts it in replog
    5. slurpd on LDAP1 applies change to LDAP2
    6. slurpd on LDAP2 applies change to LDAP1

End result: All changes processed, LDAP servers out-of-sync.

I've been wracking my brain trying to figure out an elegant solution to
this, but I'm at a loss.  Since neither slapd knows about the other
replication agreement (in a manner of speaking -- they know that they
have replication agreements, but not who is replicating to them), there
seems to be no way to serialize the changes such that this situation
won't happen.

It seems that an external clock or sequence is required.o

Anybody?

jon

-- 
.Jonathan J. Miner------------------Division of Information Technology.
|miner@doit.wisc.edu                 University Of Wisconsin - Madison|
|608/262.9655                               Room 3149 Computer Science|
`---------------------------------------------------------------------'

Just put in another goto, and then it'll be readable.  :-)
             -- Larry Wall in <199804161810.LAA18902@wall.org>
                                                                 (251)