atom feed8 messages in org.apache.beehive.devRe: EJBControl for EJB 3.0
FromSent OnAttachments
Andrew McCullochJul 14, 2006 1:40 pm 
Chad SchoettgerJul 18, 2006 8:26 am 
Kenneth TamJul 18, 2006 3:59 pm 
Andrew McCullochJul 18, 2006 7:47 pm 
Chris HogueJul 19, 2006 3:25 pm 
Ken TamJul 19, 2006 7:00 pm 
Eddie O'NeilJul 19, 2006 8:46 pm 
Andrew McCullochJul 20, 2006 9:09 am 
Subject:Re: EJBControl for EJB 3.0
From:Andrew McCulloch (amcc@gmail.com)
Date:Jul 20, 2006 9:09:52 am
List:org.apache.beehive.dev

OK, these are great points that lead me to believe that and EJB3 control does not provide much value add for Beehive right now. I will still do some testing to make sure the current control works with EJB3 beans that use the 2.1 back compat annotations.

A generic injection method for pageflows that can inject EJBs as well as Controls and other arbitrary objects seems like the way to go from here and would provide far more benefit than an EJB3 control. It seems as if it would not be possible to use the JSR-220 @EJB annotation based on Ken's comment about cherry-picking JSR parts. As far as JSR-250... would it be possible to implement all 10 annotations in Beehive?

@Generated seems like mostly a no-op, except perhaps using it in any code-generated files Beehive apt processors create. @Resource(s) seems like exactly what we are discussing The 2 lifecycle based annotations appear to have some overlap with existing lifecycle events that Beehive Controls already deal with. The 5 security annotations are a bigger question since I am not familiar with what beehive controls currently do around security. This may be a lot of work that would preclude Beehive from implementing JSR-250... I just don't know.

Maybe this is a much bigger wish-list than is necessary, but, I would hate to have to invent/learn another set of dependency injection annotations if there is any way we can use the ones that are already defined.

--Andrew

On 7/19/06, Eddie O'Neil <ekon@gmail.com> wrote:

Ken--

To your point on injection into JPF, I totally agree with the direction here. It should be possible (and super simple!) to inject arbitrary objects into a Page Flow and any other Beehive source artifact without limiting the framework to a specific type that can be injected. Basically, this should be a plug point that could be extended by a developer to implement arbitrary injection mechanisms (including support for @EJB). It'd be even better if there was a common way to specify an injection style between Controls, Page Flow, and web services so there's only one configuration point to learn.

In general I agree that we could deprecate @Control in favor of other resource injection mechanisms as long as we're not tied to only injecting JNDI resources and private field injection is supported. Both of these have important uses that should come forward.

Eddie

On 7/19/06, Ken Tam <kent@gmail.com> wrote:

IANAEE5E (I am not an EE5 expert :) either, but let's keep talking and maybe we'll stumble our way to something useful..

Agree that the web container supports @EJB for injection of an EJB3 client proxy -- I can't find it in the official specs, but I've seen a Sun demo that did this. Yes, we'd definitely need to do some work to support @EJB in pageflows, but in general it seems like it'd be a great idea to move pageflow towards supporting various injection mechanisms.

(There may be some legal issues around supporting @EJB -- I'm thinking of the prohibition on "cherry-picking" parts of a JSR, ie implementing parts you like but not others -- but that seems like it would be terribly stupid..)

On 7/19/06, Chris Hogue <csho@gmail.com> wrote:

I was wondering about Andrew's second case. I'm not an expert on the EE5 spec but I believe it does support injection into different containers, including the web container. Of course it knows nothing about page flows. So for example, if I wanted to use an EJB3 SLSB from a page flow, I assume I couldn't use @EJB injection to do that without work in the page flow runtime.

So this may be what Andrew is implying, but is the second case really about supporting @EJB injection (and possibly other JSR-250 annotations like @Resource) in page flows and controls?

I'm a little lost when it comes to references like "Andrew's second case" -- not sure exactly what you're asking/saying here, but I'll take a stab :)

Per my earlier mails on future directions, I really believe controls (and Beehive as a whole, including pageflows and wsm) ought to get out of the injection business. E.g. we should deprecate @Control (though still provide legacy support for it, outside of core), and support injection via standard Javabean accessors, JSR-250, and other standardized/popular mechanisms.

Part of where controls add value is in its ability to define injected values via annotations (that's what propertysets are all about) -- I think we should continue to emphasize that, but treat annotations it as yet another source of injection values, not as the way to define what is injectable (a trail well-blazed already by others). Some of the code in the hornet sandbox illustrates what that might look like.

-Chris

On 7/18/06, Andrew McCulloch <amcc@gmail.com> wrote:

Kenneth,

Thanks for the input! Yesterday when going through the possible use cases for this control I came to much the same thought... What would be the value add for an EJBControl on top of EJB3 Session Beans? Basically I have come up with a couple of minor points. First is consistency. Users who currently use controls for accessing almost every resources may want to continue that model with EJB3. This is a very minor point and by itself may not warrant this development effort. The second issue may just be a misunderstanding of the spec on my part, but, I will mention it anyway. Does the JEE 5 spec require the @EJB injection to work from every container? If not, it may be worth having the control to hide the jndi lookup users would be required to do if they access an EJB from the web container. This simply hides a few lines of code that could also be abstracted into helper classes by the user. This would mean that #4 in my original e-mail may be moot. So these two points by themselves are probably not worth creating a EJB3 control, but coupled with any work involving my first task below may make it a trivial amount of additional effort, I am not sure. I also need to investigate further to see if there are any additional featured introduced in EJB3 that would be worth exposing in an EJBControl, perhaps surrounding transactions or security.

As far as entity beans go... I have not gotten to the point where I understand the changes from 2.1 to 3.0 enough to comment on any possible value add here. I hope to get to a more knowledgeable point on the subject this week.

Thanks, --Andrew

On 7/18/06, Kenneth Tam <kent@gmail.com> wrote:

Hi Andrew,

On 7/14/06, Andrew McCulloch <amcc@gmail.com> wrote:

I have been reading up a bit on the EJB 3 spec and I would like

to look

into

updating / cloning the current EJBControl to support the EJB 3.0client contracts. I am looking for a few pointers on what features of

the new

spec

the community might be most interested in (it may be too new for this feedback). I would also be receptive to any other thoughts on

this

topic

that you may have.

My current plan: 1. Determine if the current controls works against EJB 3 beans

that use

the

back-compat annotations in the spec to produce the remote interfaces and other EJB 2.1 artifacts.

+1

2. Determine what would have to be modified on the Session Bean

Control

to

use only the Business interface through direct lookups. 3. Determine what would have to be modified on the Entity Bean

Control

to

use only EJB 3.0 artifacts through direct lookups. 4. Determine what would have to be modified on the 2 controls to

make

use of

EJB Dependency Injection instead of direct lookups.

Can you describe a few use-cases for these items? I'm not very knowledgeable about EJB 3, but it seems to me that the EJB 3 client model has essentially standardized much of the value-add that the EJB control offered in terms of simplifying the EJB 2.1 client model, to the point where I don't really understand how what you're proposing would make life easier for the developer.

Take a classic session bean use-case:

---

EJB 2.1:

Trader trader = null; try { InitialContext ic = new InitialContext(); TraderHome home = (TraderHome)ic.lookup("MyTraderBean"); trader = home.create(); TradeResult tradeResult = trader.buy(stock, shares); return tradeResult; } catch (NamingException e) { ... } catch (CreateException e) { ... } catch (RemoteException e) { ... } finally { if (trader != null) trader.remove(); }

EJB Control:

@Control TraderControlBean traderControl;

try { TradeResult tradeResult = traderControl.buy(stock, shares); return tradeResult; } catch (RemoteException re) { ... } finally { if (traderControl != null) traderControl.remove(); }

EJB 3:

@EJB Trader trader

TradeResult tradeResult = trader.buy(stock, shares); return tradeResult;

---

Between what the spec has done in switching to an injection model, removing the need for a home interface (and corresponding indirection), and removing the need to support RemoteExceptions on the business interface, I'm at a bit of a loss to see where there's real value to be added in evolving the current EJB control to make it easier to work with EJB 3 beans.

My sense is that the value of the EJB control today is in providing EJB3-style ease-of-use on top of EJB 2.1 beans -- and actually, this is an area where EJB3 didn't really make huge inroads. Even though the EJB 3 client model supports accessing 2.1 beans, there is still value in what the EJB control does wrt collapsing the remote/local & home interfaces and managing some of the exceptions.

An ideal result would be the ability to treat an EJB 2.1remote/local interface just like an EJB 3 business interface. Collapsing the remote/local & home interfaces is the easy part, but there's a tension between simplifying exception handling and ease of authoring.. since the remote interface will always throw RemoteException, it seems the only way to eliminate it would be to author a separate interface (which is a drag)...