Hi Dustin,
Yes, I do require the ability to be able to manually set transaction
boundaries from controller code. The main reason is because Hibernate
updates database records when a persistent object changes state during a
transaction. It picks up on the changes made to the object when the
transaction is committed and persists them to the database. Using the
service layer to do that wouldn't really make sense because you would
have to create methods for every set of changed properties you require.
And even if you did that, you still wouldn't be able to combine the
transaction with any other controller logic.
Basically, I try to use service layer calls as much as possible without
an explicit transaction created in the controller if not necessary.
Updates to persistent instances virtually always require the controller
to start a transaction though, for the reason outlined above. So for
example, I have a service for products which has a method to grab a
product instance by ID. If all I need is the product instance to read
some stuff from, I'll do:
Product p = productService.getProduct(myProductId);
Notice that I didn't create a transaction in the controller code, so the
service method will create one for itself, and that's fine. When I
update a product however, I'll do this:
txr.execute(new TransactionalExecution(){
public void execute() {
Product p = productService.getProduct(myProductId);
if(p == null) throw new MyException("No such product exists!");
p.setName(name);
p.setPrice(price);
p.setColor(color);
}
});
The reason here is that I need the product instance to be updated with
my Struts 2 parameter data (name, price and color), so I need to change
its properties during a transaction. I think you'll agree that it really
doesn't make sense to create a service layer method updateProduct(name,
price, color) or something like that for every possible combination of
properties you want to change. As I mentioned before, the controller
logic that updates the instance's properties can use arbitrary service
layer functionality all within the same transaction, courtesy of
propagation=REQUIRED. I think this makes for pretty clear and easy to
understand code.
Does this answer your question?
Jeroen,
This setup is so that you can initiate and control the properties of the
transaction from the controller, if that is a pattern you require?
Do you do this for all your calls to the service layer from controllers, and
how is it better/different from a calling a service method annotated with
@Transactional(propagation=Propagation.REQUIRED) normally from the
controller? Is it just so you can control the propagation characteristics?
It seems like an interesting pattern, I am just wondering how it is used.
Jeroen De Ridder wrote:
I'll agree that a service layer alone won't cut it, simply because of
the way JPA/Hibernate works. Updating an instance for example is just
something that doesn't belong in a service. I'm by no means an expert of
best practices in JPA/Hibernate and Spring, but I've found a combination
of services and anonymous runner interface instances to work quite well.
Basically, the idea is that you create a bunch of services to do routine
stuff to improve code clarity and avoid code duplication in your
actions. You'd mark these services with propagation=REQUIRED, so that
they can run by themselves if needed as well as run along with any
existing transactions. For logic that needs more than a single call to a
service, I then do something like this:
txr.execute(new TransactionalExecution(){
public void execute() {
Foo foo = fooService.getFoo(id);
if(foo != null) throw new FooException("No such foo exists!");
foo.setName(name);
}
});
TransactionalExecution is just an interface with a single method
execute() that exists just so we can create anonymous instances of it to
pass to txr, which would be an instance of TransactionalExecutionRunner:
public class JpaSpringTransactionalExecutionRunner implements
TransactionalExecutionRunner {
@Transactional(propagation=Propagation.REQUIRED)
public void execute(TransactionalExecution t) {
t.execute();
}
@Transactional(propagation=Propagation.REQUIRES_NEW)
public void executeRequiresNew(TransactionalExecution t) {
t.execute();
}
@Transactional(propagation=Propagation.MANDATORY)
public void executeMandatory(TransactionalExecution t) {
t.execute();
}
}
(I'm sure you can figure out what the TransactionalExecutionRunner
interface says). You'd then declare the transactionalExecutionRunner
bean in your Spring context and have it injected into every action
created by the Spring object factory through autowiring for example, and
you're good to go. The cool thing about this is that your controller
code stays very clear and to the point with minimal persistence bloat,
and that any call to a service method from within a
TransactionalExecution will automatically run within the ongoing
transaction.
As for your configuration, other than your applicationContext.xml file
you shouldn't have to do anything other than include the spring plugin
jar in your classpath. The jar comes with a struts-default.xml file that
sets Spring as the default object factory. Of course, it can never hurt
to explicitly set the objectFactory; I'm using
struts.objectFactory=org.apache.struts2.spring.StrutsSpringObjectFactory,
but struts.objectFactory=spring should work equally well.
-- Jeroen
Hi Jeroen,
The problem is that I am not a big fan of services layer. Sometimes it
looks
very anemic to me. But I totally agree with you when you say the action
should not know about persistence problems, and that's why I want to do
it
via AOP.
I had the same thought about the problem: the Spring proxy does not work
properly with all the magic Struts2 and Reflection do!
I tried to open a bug in the Struts2 JIRA, but they closed it and said
that
it works. I think it should be some kind of spring or struts
configuration I
am not doing right.
Thanks in advance,
Mauricio
On Thu, Apr 30, 2009 at 11:22 PM, Jeroen De Ridder
<voetsjo...@gmail.com>wrote:
You really shouldn't be making your Struts 2 actions @Transactional.
Doing
that causes Spring to create a proxy so it can put some extra
transaction-handling logic between the method call and the actual
method.
The thing is, Struts 2 and OGNL rely heavily on reflection on the action
classes which simply does not work at all with the proxies created by
Spring.
Regardless, making your actions @Transactional means mixing persistence
concerns with controller logic in the same class. You should consider
keeping the two separated. For example, the service approach is a good
start:
http://struts.apache.org/2.0.14/docs/struts-2-spring-2-jpa-ajax.html.
Yes, I am. Everything works fine when I don't try to use Spring
transactional AOP!
Mauricio
On Thu, Apr 30, 2009 at 9:43 PM, Dave Newton <newton.d...@yahoo.com>
wrote:
Mauricio Aniche wrote:
I am using Struts2+Spring+JPA/Hibernate. When I use the
@Transactional
to
mark an execute() method in a Struts2 Action, the action stops
working
properly (i.e. the attributes in the action are not automatically
setted).
It does not work with Spring AOP transactions as well.
In my struts.config I setted the following constant:
----
<constant name="struts.objectFactory" value="spring" />
You're using the Spring plugin, correct?
Dave
---------------------------------------------------------------------
To unsubscribe, e-mail: user-unsubscr...@struts.apache.org
For additional commands, e-mail: user-h...@struts.apache.org
---------------------------------------------------------------------
To unsubscribe, e-mail: user-unsubscr...@struts.apache.org
For additional commands, e-mail: user-h...@struts.apache.org
