I see. That clears it up for me. Thanks.

----- Original Message ----
From: Madison Pruet <mpruet@tx.rr.com>
To: DL Redden <redden96@yahoo.com>
Cc: Informix List <informix-list@iiug.org>
Sent: Tuesday, February 20, 2007 10:40:12 AM
Subject: Re: Update anywhere column level ER

DLRedden wrote:
> Madison (or anyone else who wishes to chime in),
>
> I have a 4-way update anywhere ER topology established that has been
>working fine for about a year now. We have just recently found that the
> row level Informix replication has started causing some issues where one
> branch offices updates were being overwritten by some other offices
> changes. The two locations modified different columns of the same record
> so if we had been replicating column changes instead of rows this would
> not have been an issue.
>
> In the docs, regarding column level changes, it says that "You cannot
> use update-anywhere replication; doing so can result in inconsistent
> conflict resolution."
>
> My company wants changed column, update anywhere replication so I need
> to know the following:
>
> 1) Why will changed column replication result in "inconsistent conflict
> resolution"?

I assume that you are talking about "partial row replication" in which
only the changed columnsget propagated. In order to provide
convergence, each column would need to have it's own conflict resolution
columns. Let's just take a simple example where we have two columns
(col1 and col2) with timestamp conflict resolution.

As you are aware, by turning on conflict resolution, you get a couple of
extra hidden columns on the table, but those columns reflect the time of
the row, not of the columns within the row.

Time Serv1 Serv2
start (10,20) (10,20)
1 update col1 to 11

2 update col2 to 21

3Arrival of data from Serv1
Since TS 1 < TS 2, update rejected
4 Arrival from Serv2
Since TS 2 > TS 1, update accepted

Final
Row (11,21) (10,21)


> 2) Can update-anywhere changed column replication beimplemented or will
> IDS prevent it?
> 3) If it can be done what would the implementation look like, I'm
> assuming a bunch of primary target replicates?
>
> Thanks,
>
> DL