Re: [ecf-dev] Discovery race condition?

[Bryan Hunt <bhunt@xxxxxxx>]

On Jun 9, 2009, at 1:19 PM, Scott Lewis wrote:

> Hi Bryan,
>
> Bryan Hunt wrote:
> > <stuff deleted>
> > >
> >
> > You are on the right track here, but consider the following:
> >
> >   1. The org.eclipse.ecf.osgi.services.discovery bundle is started
> >   2. The org.eclipse.ecf.discovery bundle is started
> >   3. A bundle containing a discovery provider is started
> >   4. CompositeDiscoveryContainer.addContainer() is called
> >   5. The discovery provider connect() is called and services are
> >      discovered
> >   6. The org.eclipse.ecf.osgi.serives.discovery service tracker is
> >      called and the listener is added
> >
> >
> > The services discovered in step 5 will not be seen in step 6.
>
> I believe this will only happen if the discovery provider...when  
> it's started (3) does *not* call connect before it registers itself  
> as a IDiscoveryLocator.  That is...if the discovery provider calls  
> connect prior to/before it's publication as an IDiscoveryLocator  
> service, then the ServiceTracker created and opened in step 1, will  
> be notified (via addService) when the IDiscoveryLocator is  
> registered...and the addService implementation will then add a  
> service listener synchronously as part of the IDiscoveryLocator  
> service registration.

What if connect() discovers services, or connect starts a thread that  
discovers services before the listener is added?

> So the upshot is that if the connect call must be made for a given  
> discovery provider to initialize itself properly (i.e. begin  
> discovery) then it must be made before the IDiscoveryLocator is  
> registered as a service (as that will trigger usage...i.e. service  
> listeners will be added via service trackers).
>
> > This same problem happens when you do:
> >
> > bundleContext.addServiceListener() ... you don't get the previously  
> > added services.  I think the solution is to call:
> >
> > IDiscoveryLocator.getServiceTypes() and getServices() after adding  
> > the listener.
>
> I believe this can/could be done...to deal with the case of things  
> 'previously' being discovered.  I'll defer to Markus for comment  
> about the existing implementation and/or comments about problems  
> that could/might arise.
> Scott
>
>
> _______________________________________________
> ecf-dev mailing list
> ecf-dev@xxxxxxxxxxx
> https://dev.eclipse.org/mailman/listinfo/ecf-dev