Hi there, Image gurus.

I have a client who wants to work with dynamic transactions
(DBXBegin/DBXEnd/DBXUndo) that span multiple databases. (Just naturally
masochistic, I guess!) We have discussed the strategy for sequencing calls,
and I said I would run our tentative decisions by the list.

We have a variable number of databases, which will be accessed via a table
of BaseIDs. The table allows us to indicate whether or not a particular
BaseId is participating in the gruesome activity.

All bases are actively logging, independently, to individual log files.
There could be between 3 and 6 databases participating at any one time.

For each BaseID we have a published lock sequence strategy for locks at the
set level, although we expect to be able to get by with predicate level
locking for the most part.

We plan (tentatively) on using the following sequence of operations, where
the [] surrounding the name of an intrinsic imply that that operation is
done on each BaseId in the table, as needed, in sequence:

     [DBXBEGIN]
     [DBLOCK - (predicate level lock, varying by BaseId) ]
     [DBMashTheDataUsing DBPUT/DBUPDATE/DBDELETE]
     [DBUNLOCK]
     [DBXEND]

I have some questions to satisfy my own curiosity:

1.  Does the relative sequence of [DBXBEGIN]/[DBLOCK] matter? The manual
implies that it doe NOT, but it would be good to hear from somebody who has
engaged in this sort of behaviour!

2.  We are thinking of using conditional locking, just in case someone
somewhere has different ideas about sequences. Any strong thoughts either way?

3.  In the event that we roll back one of these mega transactions, will a
simple sequence of:

     [DBUNLOCK]
     [DBXUNDO]
 be enough, or should we be 'clever' in situations like this?

As usual, any help and comments will be appreciated.

Tony