Please excuse if I am misinterpreting your reply, but I think you are
referring to the fact that ThreadLocals can only be accessed by the owning
thread. This is thought to work in my example as
AsyncTaskHook.wrapQueueTask() is executed on the same thread as
AsyncTaskService.queueTask(), and thus has access to the ThreadLocals. The
shown hook implementation is incomplete (forgot to add that to the example,
sorry...) as there should really be a constructor on the wrapper Runnable
that extracts and saves the ThreadLocal data and keeps it until run() is
called.
 
There is one thing I guess I can't get in this design; knowledge that my
Hook has been registered. In a startup scenario where all three bundles have
just started, and MyBundle.myMethod() executes, I can't be sure that
AsyncTaskService has registered my Hook and thus started considering the
contextual workaround?
 
If I want to be sure about this, I guess I need to switch to explicit
registration with AsyncTaskService (not whiteboard), and delay publishing my
ContextualService until I know my Hook has been registered. Or are there
ways to make this work with a whiteboard design?
 
Thanks
Mike
 
BJ Hargrave wrote:

Conceptually yes. But in the specific example, the queueTask method impl
would need to do the wrapping since it know what thread's threadlocals to
use. 
-- 



BJ Hargrave
Senior Technical Staff Member, IBM
OSGi Fellow and CTO of the  <http://www.osgi.org/> OSGi Alliance
 <mailto:[email protected]> [email protected]       


office: +1 386 848 1781
mobile: +1 386 848 3788





From:        Mike Wilson <[email protected]> 
To:        "'OSGi Developer Mail List'" <[email protected]> 
Date:        2014/06/05 08:25 
Subject:        Re: [osgi-dev] cooperation between optional bundles 
Sent by:        [email protected] 


  _____  




So based on the first alternative I made up the below example as an
illustration. 

Let's say I have two services in my system:

 ContextualService (stores data in ThreadLocal)
 AsyncTaskService (runs tasks in separate threads)

that are used by one of my other bundles:

 MyBundle.myMethod() {
   contextualService.doStuff();
   asyncTaskService.queueTask(new Runnable() {
     public void run() {
       contextualService.doMoreStuff(); // fails
     }
   });
 }

then the call to doMoreStuff() will fail as the ThreadLocal is not available
in the async thread. To solve this, AsyncTaskService could offer a Hook
interface that ContextualService implements:

 interface AsyncTaskHook {
   Runnable wrapQueueTask(Runnable r);
 }

 class ContextualServiceAsyncTaskHook 
     implements AsyncTaskHook {
   public Runnable wrapQueueTask(Runnable r) {
     return new Runnable() {
       public void run() {
         <set up ThreadLocals>
         <invoke original runnable>
         <clear ThreadLocals>
       }
     });
   }
 }
 registerService(new ContextualServiceAsyncTaskHook());

that AsyncTaskService discovers whiteboard-style and invokes on every call
to its queueTask method.

Does this example make sense?
Thanks
Mike 

________________________________

On Tue, Jun 3, 2014 at 8:32 AM, Mike Wilson <[email protected]> wrote:

I'm looking at the generic problem of one optional bundle wanting to affect
the behaviour of another optional bundle's service, without hard-wiring
these bundles too much to each other. Each bundle should be deployable by
itself in a system, but when both are available they should be able to hook
into each other to affect the outcome of calls made on them by other
bundles.

As I have control over source code myself the first alternative that comes
to mind is the "cooperative" style where each bundle could provide a custom
Hook interface that others may implement. When executing calls from other
bundles these Hook implementations could then be consulted for modifying the
result of the operation. This seems to be in the spirit of many parts of
core OSGi.

Or, I could skip on the cooperative style and rely on framework ServiceHooks
and proxies, which seems to be more hackish but may have advantages?

Or are there other, better, alternatives?

Thanks
Mike Wilson

_______________________________________________
OSGi Developer Mail List
[email protected]
 <https://mail.osgi.org/mailman/listinfo/osgi-dev>
https://mail.osgi.org/mailman/listinfo/osgi-dev




_______________________________________________
OSGi Developer Mail List
[email protected]
https://mail.osgi.org/mailman/listinfo/osgi-dev

Reply via email to