On 2 January 2013 13:02, Peter Firmstone <[email protected]> wrote:
> Dan Creswell wrote:
>>>>>
>>>>> These methods could easily return jini://host:port for standard Jini
>>>>> unicast
>>>>> discovery, this allows a lot more freedom for future expansion of
>>>>> discovery
>>>>> methods, for very little effort.
>>>>>
>>>>>
>>>>>
>>>>
>>>> How would you see these being used in a real system? Would I want
>>>> these URL's on the Registrar? See, to get them, I need to have got
>>>> hold of the Registrar in the first place. Dunno, this feels like
>>>> something that needs to be on an Admin interface somewhere.
>>>>
>>>>
>>>
>>> Got time for an example?
>>>
>>>
>>
>>
>> You offering or asking? Can't tell.
>>
>> The thing about getting these locators from the registrar itself is
>> that you have to have located the registrar in some other way
>> previously (e.g. via multicast).
>>
>> Question then would be, who uses unicast lookup locators and why?
>> Developers, those messing with firewalls?
>>
>> And once we've agreed on that a little, what's the best way for those
>> people to get and use those URLs? Perhaps they have to be shoved into
>> a config file or passed on a command-line to some client? So we'd need
>> a way for people to get ahold of these URL's for sure and we could do
>> as you suggest but is that the best way? As I say, feels to me like an
>> Admin/Config task best handled via a tool operating on an Admin
>> interface or similar.
>>
>>
>
>
> I suppose the use case for asking the ServiceRegistrar for a URL (similar to
> LookupLocator) is weak, the client has a copy via multicast discovery or
> configuration anyway, future methods might include dns-srv records.
> I was thinking about protocols other than jini://host:port for unicast
> lookup, tunnelling for example, yes messing with firewalls and client
> mobility.  LookupLocator only supports the jini://host:port scheme.
>

Yeah, that's kinda where my mind is going. I'd like to see a real-need
emerge to drive the change and I can't see it at the moment. S'why I'm
asking you and anyone else to wade in...

> Are there any other methods a next gen ServiceRegistrar could benefit from?
> Once it's released, were stuck with the interface.

Mmmm, well I hadn't thought about it much at all. Was just doing my
usual architecture hat feedback. I guess we should look back through
the archives and see what others think too.

>
>
>>>> I would certainly say StreamServiceRegistrar is the wrong place to put
>>>> them. They have nothing to do with Streaming at all. That's okay,
>>>> StreamRegistrar is becoming a next-gen ServiceRegistrar interface but,
>>>> if it is, we need a different name.
>>>>
>>>>
>>>>
>>>
>>> Sounds like you've a better name in mind?
>>>
>>>
>>
>>
>> Heh, hell no. When we did the JavaSpace extensions we ended up with
>> JavaSpace05 because we had such a mixed bag of bits.
>>
>> If we're gonna follow that style though we'd have ServiceRegistrar13
>> or 14 for the superstitious :)
>>
>
> Hmm, maybe ServiceRegistrar2?   Perhaps it should be moved into the same
> package as ServiceRegistrar?
>

re: package, yes. re: name, that's fine with me, although someone,
somewhere must surely be able to come up with something better ;)

> I figure it's about time we added JavaSpace05 to the Jini Specifications as
> well, I've written it up locally using wording from javadocs.
>

Yes, I was around at the time and, to all intents and purposes, the
JavaDoc was seen as the spec so there'd be no problem with that.

> Findbugs identified some concurrency issues with Outrigger.  I've been
> wondering if we should drop Outrigger and promote Blitz and Rio instead, hey
> they've got dedicated maintainers ;)  It would be relatively straightforward
> to set up a JavaSpace TCK from the qa suite.

I've never actually run findbugs against Blitz, it may be worse :)

>
> I was also thinking about the benefits of dropping Phoenix, we could support
> other JVM's.  Phoenix ties us to Sun JVM proprietary implementation classes.
> Activation isn't part of the Jini spec anyway.
>

>From recollection only, Phoenix will do basic security type stuff that
rmid doesn't do. Probably some thinking to do there. A quick google
reveals: http://river.apache.org/doc/release-notes/execpolicy.html


> I'd also like to remove TaskManager and other implementation classes that
> could be replaced by Java concurrent libraries.
>

No reason why not.

> Slim things down to reduce the maintenance burden somewhat.
>

Smaller codebase is always good.

> Cheers,
>
> Peter.
>

Reply via email to