cayenne-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Eric Lazarus <ericllaza...@yahoo.com>
Subject Re: Correctly implementing optimistic concurrency management in Cayenne for web-based computing
Date Fri, 05 May 2006 15:13:30 GMT
And one more question: 

Should access to the "changed elsewhere" set be
synchronized? Will different threads be invoking the
data source delegate? Right now it is not synchronized
and we do not seem to be having problems. 

Thanks!

Eric

--- Eric Lazarus <ericllazarus@yahoo.com> wrote:

> Bryan, I am not sure I know the answer to your
> question. I need to explore the answer to that.  
> 
> Ok, as of today�s small changes, we now clear the
> changed elsewhere set at the very start of the page
> cycle, which is before any updates are made to the
> application domain objects. Today�s testing seemed
> to
> suggest that we still have some unexpected
> concurrency
> conflicts but the window is much smaller so we
> believe
> we are getting fewer of them. 
> 
> We determine if we want to commit or roll back just
> before rendering the HTML, after the main changes
> have
> been made to the object model. 
> 
> One concern that we have is that the rendering
> process
> itself can be responsible for changes to the object
> model, for example, lazy initiation of an instance
> variable will sometimes happen as a result of
> rendering. What would happen to those changes? Will
> they hang around in a session and cause problems
> somehow? What problems? 
> 
> Seems to me that we should consider committing after
> render also, not just before. Does that sound right?
> What happens if we hit an update conflict at that
> point? I will be hard to display the message in the
> browser if the rendering is complete so I am not
> sure
> what to do to inform the user that there was a
> problem
> at that point.  
> 
> Eric
> 
> --- Bryan Lewis <bryan@maine.rr.com> wrote:
> 
> > I'm missing something.  Why not use Cayenne's
> > built-in optimistic
> > locking support, with a timestamp or other
> > versioning attribute in each
> > important entity?
> > 
> > 
> > Eric Lazarus wrote:
> > 
> > >Anyone want to write an article on: Correctly
> > >implementing optimistic concurrency management in
> > >Cayenne for web-based computing? We will soon
> have
> > a
> > >commercial application doing this! I would think
> it
> > >would be something that lots of web-developers
> > should
> > >want to do if they are building a web-based
> > >application with a domain object model. 
> > >
> > >Anyway, we still have a few small problems with
> it.
> > >We obtain a set of ObjectIds for all objects
> > changed
> > >in our session, and another set of ObjectIds for
> > all
> > >objects changed in other sessions. When these
> sets
> > >have a non-empty intersection we have an update
> > >conflict
> > >and rollback our changes, and issue a message
> > >containing the offending objects.
> > >
> > >The Ids changed in our session come from
> > >DataContext.modifiedObjects()
> > >The Ids changed in other sessions come from a set
> > >maintained by our
> > >DataContextDelegate, which adds the ids of
> objects
> > >passed to its
> > >"shouldMergeChanges() method. This set is cleared
> > each
> > >time we commit
> > >or rollback our changes.
> > >
> > >Perhaps we should shorten the period of time when
> > >conflicts can occur by clearing the
> > changed-elsewhere
> > >set at the beginning of the page submit cycle
> > rather
> > >than at the end of the pervious submit. How can
> we
> > >tell if that will give us the correct semantics?
> We
> > >want to make sure that we are not overwriting new
> > data
> > >with 
> > >old data. Would we be messing that up if we clear
> > he
> > >changed-elsewhere set just before
> > >we begin updating objects in the object model?
> > >
> > >What is most troubling for us right now is that
> we
> > are
> > >hitting conflicts on objects that we are not
> > >intentionally modifying, and are having trouble
> > >tracking down the code that is causing these
> > objects
> > >to wind up in the "modified"
> > >collections.
> > >
> > >All of our persistent objects descend from the
> > class
> > >PAXPersistentObject, which extends
> > CayenneDataObject.
> > >I used this to try to track down the 
> > >updates by overriding writeProperty(),
> > >addToManyTarget(), removeToManyTarget(),
> > >setToOneDependentTarget(), and
> > setPersistenceState()
> > >produce a log of ObjectIds
> > >and their modified fields (or PersistenceState).
> > >
> > >Am I logging in on the correct methods? What else
> > >might be called that would cause an object to be
> > put
> > >into modifiedObjects and/or passed to the data
> > context
> > >delegate? 
> > >
> > >The problem is that objects are winding up
> causing
> > >update conflicts without showing up in this log.
> > >Strangely we never see setPersistanceState set
> the
> > >state to 'Modified' - the only values we see are
> > >Hollow (5) and Committed (3).
> > >
> > >Is there some better way to find which fields and
> > >records in our database are
> > >being modified ? It would be great if we could
> find
> > a
> > >spot to put a breakpoint
> > >where a stack trace could lead to the code that
> is
> > >doing the modification, and
> > >where the objectid and property name are
> availible
> > so
> > >that we can filter out the many fields that we
> are
> > not
> > >conserned with.
> > >
> > >To summarize we have 2 questions.
> > >(1) What is the best way to find places in our
> code
> > >that
> > >     are causing un-intended updates?
> > >(2) Are we constructing the update conflict set
> > >correctly?
> > >
> > >Below you will find the relevent portions of our
> > code.
> > >Can you help ?
> > >
> > >If notice we are missusing Cayenne in some way,
> > please
> > >let us know. We love it and want to use it right.
> > >
> > >Dan and Eric
> > >
> > >
> > >/* This is called after all updates are complete,
> > just
> > >before the page
> > >   is rendered. It will rollback when an update
> > >conflict is found, and
> > >   commit otherwise.                             
>  
> >   
> > >             */
> > >
> > >public void preRenderNotification(EditorContext
> > >aContext) {
> > >  try {
> > >    Set aSet= getUpdateConflictSet(aContext) ;
> > >    if(aSet.size()>0)
> > rollbackChanges(aContext,aSet) ;
> > >    else getDataContext(aContext).commitChanges()
> ;
> > >  } catch(Exception e) {
> > >    getAppModel(aContext).addMessage(e, "Error In
> > >Database Update") ;
> > >    e.printStackTrace() ;
> > >getDataContext(aContext).rollbackChanges() ;
> 
=== message truncated ===


__________________________________________________
Do You Yahoo!?
Tired of spam?  Yahoo! Mail has the best spam protection around 
http://mail.yahoo.com 

Mime
View raw message