Re: distributed transactions

From: Andrus Adamchik (andru..bjectstyle.org)
Date: Sun Nov 09 2003 - 15:27:39 EST

  • Next message: Andrus Adamchik: "[Implemented in 1.1] Was: distributed transactions"

    On Nov 7, 2003, at 9:28 AM, Michael Gentry wrote:

    > I haven't really looked at the code or tried to use multiple DBs, but
    > it seems to me a practical (although not failsafe) approach would be
    > to do all of the INSERT/UPDATE/DELETE statements across all of the DBs
    > and then do the COMMITs across all of the DBs. You can detect when
    > there is an error updating the data and rollback everything as one
    > combined transaction. If all of the INSERT/UPDATE/DELETE statements
    > succeed, then do all of the COMMITs, which will most likely succeed
    > (no real reason why it shouldn't at that point). Of course, there is
    > a chance that something will happen before you can get all the COMMIT
    > statements processed (you process the first and then try to process
    > the second, but you have lost the second DB connection, etc), but that
    > window is pretty small. It might be a fair trade-off if you know the
    > risks involved.
    >
    > Comments?

    Yes, this is what I was trying to say too. I opened a JIRA ticket for
    it:

    http://objectstyle.org/jira/secure/ViewIssue.jspa?key=CAY-28

    It looks like the changes required to fix it (though relatively
    trivial), require API modifications, and are generally pretty invasive.
    So I am targeting this for 1.1, not 1.0. Feel free to post comments on
    the ticket above. Also you can "watch" the issue, i.e. receive updates
    from JIRA whenever it is updated/fixed. To configure a "watch", you
    must login to JIRA, go to the link above, and click "Watch It" link on
    the left.

    Andrus



    This archive was generated by hypermail 2.0.0 : Sun Nov 09 2003 - 15:27:43 EST