Re: Best Practices Question

From: Michael Gentry (mgentr..asslight.net)
Date: Tue Apr 21 2009 - 17:00:42 EDT

  • Next message: Ylan Segal: "Re: Best Practices Question"

    On Tue, Apr 21, 2009 at 4:40 PM, Ylan Segal <ylan.sega..mail.com> wrote:
    > Ok. That makes sense. I can see that for a web application you would want to
    > have each user manipulate data on it's own, without affecting what other
    > users are doing. As far as junit test cases go, I can just create a new
    > DataContext without much consequence. Right?

    Yes. A DataContext (or ObjectContext) is fairly cheap to create, too.
     Create them as it makes sense. Some people like to create a new DC
    for every "transaction" they wish to do. Some like to have a
    session-based DC (in a web application). Some use a mixture or add in
    nested DCs. It really depends on what you are trying to accomplish.
    It is also safe in Cayenne to reuse a DC across request/response loops
    in a web application. Or even within the same request/response cycle.
     Cayenne does not disconnect objects from their database channel after
    a commit is done (unlike Hibernate). You typically do not care about
    the database channel with Cayenne, either -- the DC manages that.

    > In a web app where each session has a DataContext, do I need to worry about
    > data caching in different DataContexts? I don't know if that is clear, so
    > let me explain. Suppose we are talking about a pet store. User A and B are
    > both separately
    > in their browser shopping for a cat. There is only 1 left in inventory. Both
    > user A and B see that there is one left. User A makes up his mind first and
    > buys the cat. The changes are committed. In the DB the inventory correctly
    > has the number of cats at 0. If User B's DataContext caches the data, it
    > might think that the inventory is still 1 and also allow User B to buy the
    > cat. Should I worry about this or is it all taken care of by Cayenne?

    This actually brings up many issues. One of the issues is data
    freshness (and you can force a refresh if if you want). However,
    there is always some latency involved between two requests, so even
    that could result in a missed read. I think what you are really
    asking about here is optimistic locking:

    http://cwiki.apache.org/CAY/optimistic-locking-explained.html

    Let's say user A and B both read from the "inventory" table a value of
    1 for "number_left". You are really wanting Cayenne to generate, as
    part of a transaction -- dataContext.commitChanges() -- an update
    similar to this:

    UPDATE inventory SET number_left = 0 WHERE primary_key = 587375 and
    number_left = 1;

    When user A commits, all is well. When user B commits, an exception
    will be thrown by Cayenne because there WHERE clause fails (the
    "number_left" value is now at 0 due to user A's commit).

    > Thanks for the answers. In the meantime I have been trying out the modeler
    > and it seems to do a great job of creating the db schema, relationship
    > mapping and Java code generation. So far I am very impressed!
    >
    > --
    > Ylan.



    This archive was generated by hypermail 2.0.0 : Tue Apr 21 2009 - 17:01:16 EDT