geronimo-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Rick McGuire <rick...@gmail.com>
Subject Re: Resolving a GBean reference
Date Sat, 09 Sep 2006 12:42:29 GMT
Dain Sundstrom wrote:
> This is starting to make more sense to me. Can you point me at the 
> implementation we currently have for the sun orb?  Also, what is the 
> information the socket factory needs from the orb?
The current implementation is 
org.openejb.corba.sunorb.OrbSocketFactory.  The property/argument setup 
is done in org.openejb.corba.sunorb.SunOrbConfigAdapter. 

For the current version, the required information is passed along by 
parsing the CSSConfig/TSSConfig information and cramming the settings 
into some system properties that are picked up by the socket factory 
once it's instantiated.  Not exactly the most thread-safe of methods.  
However, in addition to the Yoko work, I'm also reworking things for 
GERONIMO-2002.  To implement this, the socket factories (for either ORB) 
is going to need a reference to a configured KeystoreManager.  This is 
not easily encoded in string form, and also involves many more pieces of 
information.  So, rather than expand the existing processing, it seemed 
better to just allow the socket factory to obtain a reference to its 
instantiating GBean (either a CorbaBean or a CSSBean) and have it pull 
the configuration information from the bean.  This seriously simplified 
things, but I still had the problem of how the socket factory manages to 
obtain the object reference when the GBean does not directly instantiate 
the object.  I'd discussed this with David Jencks back at JavaOne, and 
he recommended using the AbstractName and getBean() method.  Things 
finally reached the point this week where I could put this to the 
test....that's when I bumped into the problem with getGBean() not 
working because it was called before the doStart() method completed.  
I've managed to bypass the problem by storing the GBean reference in a 
static HashMap in the configurator class so that the socket factory can 
request it when it's started.
Rick

>
> -dain
>
> On Sep 8, 2006, at 4:02 PM, Rick McGuire wrote:
>
>> Dain Sundstrom wrote:
>>> I'm more confused now... is the ORB aware of the geronimo kernel?  
>>> If not, what code is talking to the registry?
>> Only the socket factory is aware of the geronimo kernel.  This is a 
>> pluggable bit of code that is configured by specifying a special 
>> argument on ORB.init().  Both Yoko and the Sun ORB have a similar 
>> facility in that regard.  The ORB instantiates an instance of the 
>> specified class and plugs it in as the provider that creates the 
>> sockets used for the connections.  The socket factory, once created, 
>> needs some configuration info that's held in the GBean that created 
>> the ORB instance.  Unfortunately, there's no way to pass object 
>> references through the interface used to create an ORB 
>> instance...we're limited to string arguments and properties (and 
>> again, I have the same problem with the Sun ORB as well).
>> Rick
>>
>>
>>>
>>> -dain
>>>
>>> On Sep 8, 2006, at 3:44 PM, Rick McGuire wrote:
>>>
>>>> David Jencks wrote:
>>>>> I don't follow this at all.  What gbean is creating the orb?  Do I 
>>>>> understand correctly that the ssl info can be in either a 
>>>>> CorbaBean or a CSSBean?  Can these gbeans create the socket 
>>>>> factory and whatever needs it get it from them using a reference 
>>>>> from that "X" bean to the CorbaBean/CSSBean?  It sounds like 
>>>>> there's a circular dependency but I don't see what the dependency 
>>>>> of the CorbaBean/CSSBean is on.
>>>> The ORB instantiates the socket factory, based on arguments passed 
>>>> in to ORB.init(), so there's no way to get a reference to the 
>>>> factory after the ORB is created.  The ORB instances are created by 
>>>> the doStart() methods of either a CorbaBean or CSSBean, which tells 
>>>> the ORB to create a socket factory instance.  The only information 
>>>> I'm able to pass into the socket factory instance is a single 
>>>> string value.  I was passing the abstract name of the instantiating 
>>>> GBean, which I was then attempting to retrieve using 
>>>> kernel.getGBean().  Unfortunately, since the ORB is getting 
>>>> initialized by the bean's doStart() method, the bean is not yet 
>>>> active, and not retrievable.
>>>>
>>>> I managed to get around this problem by creating a registry in the 
>>>> class that configures and creates the ORBs.  The socket factory 
>>>> calls back to that registry to retrieve the reference, and 
>>>> everything works fine from there.  This afternoon, I managed to 
>>>> successfully start the j2ee-corba module using the Yoko ORB.  That 
>>>> means we're successfully launching the transient name service plus 
>>>> 2 ORBs (one using plain sockets, one using SSL).  I immediately 
>>>> found a problem with stopping the module, but it's nice to actually 
>>>> report progress!
>>>>
>>>> Rick
>>>>
>>>>
>>>>
>>>>>
>>>>> thanks
>>>>> david jencks
>>>>>
>>>>> On Sep 8, 2006, at 2:25 PM, Dain Sundstrom wrote:
>>>>>
>>>>>> I'm not sure I totally follow this, but is sounds like a classic

>>>>>> circular dependency problem.
>>>>>>
>>>>>> To start with using KernelRegistry.getSingleKernel() or looking 
>>>>>> up references via getGBean are both highly discouraged.  It is 
>>>>>> better use either a GBean reference or to pass a reference to 
>>>>>> 'this' to the corba socket factory in the doStart method (you 
>>>>>> should never publish a public 'this' reference from a constructor).
>>>>>>
>>>>>> In this case can we change the socket factory so it has a create

>>>>>> method taking the orb or a create method that takes the 
>>>>>> information it extracts from the orb?
>>>>>>
>>>>>> -dain
>>>>>>
>>>>>> On Sep 8, 2006, at 6:42 AM, Rick McGuire wrote:
>>>>>>
>>>>>>> I think I've run into a classic chicken-and-egg problem trying

>>>>>>> to get yoko hooked into Geronimo.
>>>>>>>
>>>>>>> Let me recap the situation first, and the consensus solution
we 
>>>>>>> came up with earlier.  To fully enable the SSL transport for

>>>>>>> Yoko, it's necessary to cause Yoko to load a socket factory 
>>>>>>> object that will be used to sort out what sort of sockets need

>>>>>>> to get created for a given connection situation.  This socket

>>>>>>> factory needs access to information held by the creating 
>>>>>>> CORBABean or CSSBean object.  The socket factory is enabled by

>>>>>>> setting an argument on the call to ORB.init() to identify the

>>>>>>> name of the socket factory class, and there's an additional 
>>>>>>> optional argument that can contain a single string value that's

>>>>>>> passed to the socket factory's init() method.
>>>>>>>
>>>>>>> So, to hook the socket factory back to the launching CORBABean

>>>>>>> or CSSBean object, the socket factory argument value is set to

>>>>>>> the abstract name for the associated bean instance, and the 
>>>>>>> socket factor init() method retrieves the reference using:
>>>>>>>             Kernel kernel = KernelRegistry.getSingleKernel();
>>>>>>>             config = (ORBConfiguration)kernel.getGBean(new 
>>>>>>> AbstractName(new URI(configName)));
>>>>>>>
>>>>>>> I'm getting the ORB launched, the socket factory class is 
>>>>>>> getting loaded and the init() is failing with the following 
>>>>>>> exception:
>>>>>>>
>>>>>>>    java.lang.IllegalStateException: GBean is not running:   
 
>>>>>>> org.apache.geronimo.configs/j2ee-corba/1.2-SNAPSHOT/car?ServiceModule=org.apache.geronimo.configs/j2ee-corba/1.2-SNAPSHOT/car,j2eeType=CORBABean,name=Server

>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>> In the calling sequence, the ORB is initialized in the CORBABean

>>>>>>> or CSSBean doStart() method, so these beans are not yet in a

>>>>>>> started state when the socket factory gets initialized.  I'm

>>>>>>> unable to delay resolution of the bean because the first actions

>>>>>>> required of the socket factory occur before the doStart() 
>>>>>>> methods return.
>>>>>>> The best solution I've come up thus far would be to keep a 
>>>>>>> private registry of the beans in the ORBConfigAdapter class as

>>>>>>> static methods, and have the SocketFactory use that to retrieve

>>>>>>> the reference.  I'd rather not resort to that if there is a 
>>>>>>> better solution.
>>>>>>>
>>>>>>> Rick
>>>>>>>
>>>>>>
>>>>>
>>>>>
>>>
>>>
>
>


Mime
View raw message