river-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Peter <j...@zeus.net.au>
Subject Re: Seasonal djinn blues...
Date Mon, 11 Dec 2017 21:51:55 GMT
Thanks Bill,

Sounds interesting, let us know how you get on, maybe you could send us 
some screen shots and we could feature it on our website?

There's an updated version of Browser available at the following link, 
it uses the River 3.0 Startable interface, which was introduced to 
support safe publication of services after construction, so it might 
make a better starting point, don't mind the pun :).




On 11/12/2017 8:03 PM, bill pickup wrote:
>   Dear Peter,
> Thanks for offering a patch, but that won't be necessary, as I can work around the shutdown
problem by just dumping the JVM through System#exit(..), and the sluggishness problem by just
being patient.
>      I checked the shutdown bug on 3.0.1 and it still persists in a modified form, in
that the enigmatic thread, 'Thread-1', is left in LockSupport#parkNanos(..), so in locking.
I ran the Mickey Mouse shutdown test a few times to make sure it wasn't a racy problem.
>      To give some context to the source of these slight concerns, I'm trying to build
a replacement for the now tired 'browser' that comes with the distribution. The new stuff
uses JavaFX, so will run on Java 7 and 8, hopefully. I've got as far as event-driven controls
for registrars and group membership, and will look to other functionality of the 'browser'
in fleshing out the candidate. It looks like it's going to be quite a simple application,
so easily modifiable, which can't be said for the 'browser' any more, unfortunately.
> Kind regards,
> Bill
>      On Sunday, December 10, 2017, 10:35:21 PM GMT, Peter<jini@zeus.net.au>  wrote:
>   Thanks Bill,
> I suspect the thread bug was fixed in 3.0.1, are you able to confirm?
> I think I can provide a fix for sluggish discovery.
> One of the changes with River 3.0 was to remove instances of unsafe
> publication, that is allowing a reference to "this" to escape during
> object construction.
> LookupDiscovery was changed to implement Startable (which allows threads
> to start and safe publication to occur after construction), and
> LookupDiscoveryManager was also changed, to ensure safe publication.
> LookupDiscoveryManager utilises LookupDiscovery to perform discovery.
> I suspect the problem is LookupDiscovery starts before the safe
> publication of LookupDiscoveryManager, so that LDM is not notified
> immediately, becuase LookupDiscovery doesn't have a reference to it yet.
> It will be a simple fix to delay starting of LookupDiscovery until after
> LookupDiscoveryManager publication is complete.
> Would you like me to provide a patch?
> Regards,
> Peter.
> On 11/12/2017 12:53 AM, bill pickup wrote:
>>    Hi Peter,
>> More from theseasonal djinn…
>> So, I had a littlemore patience with River 3.0’s unadorned LookupDiscoveryManager,and
it did indeed recognised the existence of an already running 3.0djinn, but only after a fashion.
River 2.3.3’sLookupDiscoveryManager recognises the same djinn immediately.
>> Also noted was atermination concern with 3.0’s LookupDiscoveryManager, which keepthreads
>>        sun.misc.Launcher$AppClassLoader@blah_thread-0,and    Thread-1 (?)
>> running permanently.
>> The former is a bitof a bind, but the latter can be worked around with a system exit
of0, much as must be done in QA.
>> Kind regards,
>> Bill
>>        On Sunday, December 10, 2017, 7:01:49 AM GMT, Peter<jini@zeus.net.au>
>>    Hi Bill,
>> Are you trying to discover a lookup service provided by River 2.2.X?
>> Which version of discovery are you using?  V2 is now enabled by default
>> in 3.0.X, you won't discover any V1 lookup services unless you enable it
>> by passing in a ConstrainableLookupLocator using InvocationConstraint
>> with DiscoveryProtocolVersion.ONE to LookupDiscoveryManager's constructor.
>> If you're still using V1, I'd suggest using LookupLocators to support
>> both discovery versions 1 and 2, which should allow you to migrate off
>> version 1.
>> Version 1 discovery has been deprecated and will likely be removed in a
>> future version of River, for now at least, it can still be re-enabled.
>> If this isn't the cause, can you enable the loggers for LookupDiscover
>> and LookupDiscoveryManager and send us the output?
>> Regards,
>> Peter.
>> On 9/12/2017 11:42 PM, bill pickup wrote:
>>> Hi all. Season’s greetings… well, almost.
>>> Apologies for coming over the dev@ wire, but the user@ wire seems to be defunct.
Anyway, I’m designing something for River, so I don’t feel too bad.
>>> I’m having problems with LookupDiscoveryManager in 3.0.0 and 3.0.1. My call
is simple,
>>>          new LookupDiscoveryManager(LookupDiscovery.ALL_GROUPS, null, (DiscoveryListener)
>>> but this refuses to discover anything. However, it works in 2.2.3, for the same
>>> Is something broken?
>>> Kind regards,
>>> Bill

View raw message