Distributed transaction over rpc

Question

Hi,
I'm developing a distributed system configured as:
- an xml-rpc server that provide/store data into a database
- multiple Swing based client that display/modify the stored data invoking a server remote procedure
so clients don't access the database but simply extract datas from user inputs and send them as parameters of a remote call.
The server (remote) method invoked by clients is declared as syncronized and
I use a single JDBC connection to the database.
I need to give the client the possibility to ask the server to begin a database transaction, to make some (write) request, and finally to commit or rollback depending on requests's success.
Do I have to use JTA API?
Do I have to introduce connection pooling?
And what do you think about this implementation? any suggestion?
Thank's for the attention, sorry for my english

Answer 1

I will try to make a quick typical example: authors
and books.
Please, for the last time, tell me if I have to worry
about my health...
My swing interface show a tabbed pane to the client
who want to store a new author:
in the first tab client fill generic information
about the author (name,birth,...)
in the second tab client add a list of books
When client want to save this data I want to call two
distinct remote methods,doing two different
requests:
1.saveAuthor(<listOfParams>)
2.savePubblications(<listOfParams>)
'd like to keep this methods separate because if I
want to modify only author data or only pubblications
data (relative to a just saved author) I want to use
the same panels, collect data from panels in the same
way, and call the same remote method, calling
saveAuthor() in the first case and
savePubblications() in the second.Yes. In those cases saving author alone or publications alone are units of work.
I'd like to emphasize that the code to save either author or publications is NOT in the Swing UI. They should be separate Java classes that you can write, test, and debug separately, without any UI.
Do a search for the Data Access Object or DAO pattern. That's what you need here.
But when I create a new author I want save his data
and his pubblications all together. For this reason
I'd like to specify from client when a group of
requests are to be threatened as one.yes, but in that case you have to provide some kind of button to trigger an event that indicates that both author and publications are to be a combined unit of work. The thing that will demark the transaction is a service object, that in turn will use both the author and publication DAOs in a single transaction.
each request could be something like this:
connection.setAutocommit(false);
2..n write operations
connection.setAutocommit(true);
so I thoughtI can't use connection level transaction
management but something like JTA.No, you can use either JTA or Java hand-coded transactions. You just have to be able to pass the same connection to both DAOs.
Or implement my "request transaction" system if JTA
is not for this purpose.You need separate DAOs that know nothing about transactions.
You need a service layer that does know about transactions and uses DAOs to satisfy user requests.
%