Re: Jetty + JNDI

[Miroslav.Halas@xxxxxxxxxxxxxx]

Just let me know how can I help. We have 2cm of ice on all streets so life
slowed down here a bit.

Miro



                                                                              
                                           
                    Guillaume.riviere@in                                      
                                           
                    rialpes.fr                 To:     
Miroslav.Halas@xxxxxxxxxxxxxx                                     
                                               cc:     carol@xxxxxxxxxxxxx    
                                           
                    02/25/2003 05:33 AM        Subject:     Re: Jetty + JNDI  
                                           
                                                                              
                                           
                                                                              
                                           




Miroslav.Halas@xxxxxxxxxxxxxx wote :
> Hi Guillaume
>

Hi Miro,

Note that there is a new carol version 1.3.2 (another one !)
This version is just a traces correction including all your remarks on
this subjects

>
>
>
> >> It looks like I need to register all datasources and other
> >> resource-ref, resource-env-ref, ejb-ref, ejb-local-ref,
security-domain
> >> under java:comp/env/
>
> >Yes, I think that Jetty have to register resources inside java:com and
> >java:comp/env.
> >We may have to develop a JNDI SPI for this. This is not very long.
> >May be JOTM or Jetty should provide is JNDI implementation of
> >java:comp and java:comp/env.
> >We can use the one for Jetty and Tyrex or redevelop a JNDI using The
> >Jonas one, for example (org.objectweb.jonas.naming). It's depends on
> >the needed of each projects.
> >For me there is no problems with integration of a JNDI ENC "java:comp"
> >with the LMI environment and context.
> >ENC can be can be just an indirection for ressources name classification
> >in J2EE environment.
>
> I looked through the Jonas clases and they seems to be pretty
> straightforward. It seems there is only 3-4 classes with no other
> dependency on Objectweb packages than Monolog. I think (very uneducated
> guess though ;-) it would be possible to just reuse those. Do you think
> they should be part of Carol on Jetty integration?

May be ... This is probably a very good features for carol to provide
a simple JNDI management of java ENC (with java:/comp and
java:comp/env)

I can work on this stuff, in two weeks, maybe you or Jeff can help me
to integrate/adapte jonas jndi to carol jndi ?


> Also why would you make it available only for LMI and not for JRMI...?

It'a a mistake, I think that JNDI ENC is RMI implementation independant
(so, if it works with LMI it work also with IIOP or JRMP)
In any cases, the jonas one is independant.

> So what would be the scenario? I would parse the web.xml, retrieve the
> names such as jdbc/MyDataSource, then construct the full name such as
> java:comp/env/jdbc/MyDataSource, get the appropriate DataSource for that
> name and then just bind it there and the JNDI implementation would know
how
> to handle names like that?

Yes, this is, I think,  the mechanism: after, JNDI do the right conversion
for those contexts names. I need to check the JNDI/J2EE Specifications to
search for exact mechanism.

> >> AN another thing, I have changed carol.properties to
> >> carol.rmi.activated=lmi
> >> to use just the in-memory  context. What should I put into
> jndi.properties?
> >> Does it matter. THis is what Carol prints out when it starts, could
you
> >> please look over it if it look OK with you? Basically I am just trying
> to
> >> start lmi.
> >Yes, it's ok for me, the confusion is that there is many defaults for
> >carol properties. So all those defaults properties are written inside
> >the carol traces. Do you think that carol should trace only the used
> >properties ?
> >This is probably a good idea, with more readable traces.
> >I can set this features in the next version.
> Yes I think this is needed, because what everybody is mainly interested
in
> is what is active. Maybe you can handle it by two trace levels (in java
> logging I would use FINER and FINEST)
>
> > So the lmi seam to work fine ?
> > Nice ! This is, I think, experimental and non standard rmi and jdni
> > features but, without java serialization, we can get a big performance
> > enhancement for J2EE features !
> Well, so far I know that it starts without any erros and when Jeff's
little
> test tries to access the UserTransaction I get NamingException so it
seems
> to work.

Nice !
I'm going to check for JNDI specifications before the end of this week.

Best Regards,
Guillaume

>
> Thanks again for all the help,
>
> Miiro
>
>
>
>

--
Guillaume Riviere
INRIA -- SARDES/IMPACT
ObjectWeb Consortium
+(33)476615525