[cddlm] CDDLM notes from 4/26

[Milojicic, Dejan S (HP Labs)]

Attendees: Ayla, Guilherme, Steve, Satish, Jun, Dejan

Interoperability:

Ayla: Problems with the component model. Guilherme performed other
testing with Jun's endpoint. Trying to initialize endpoint, but he could
not initialize the system. Satish noticed some access and had some
syntax errors on NEC side. XML was not really conforming to his
implementation. Satish sent some email about how to create files. They
have updated their own version, but they continuously do that. Steve
said that he had some problems regarding test wrt Deployment API.
Destroy the system, then try to access it, and it should not access it
any more, but on UFCG side these activities are not synchronous, it will
not happen immediately. Steve: The question is what WSRF considers a
right behavior, not addressed in WSRF spec, probably. Neither do we say
that destroy must be instantaneous. Should we modify the test?

Steve: running his client against two public endpoints and rudimentary
conversation with OurGrid, a lot of service side errors with NEC. Will
run his endpoint before the end of the week. He will publish his outcome
so that others can see what is going on. Once he puts his end point you
can see what is the problem with Satish's end point. Probably related to
WS-Addressing. So next 2 days, bring his endpoint and see what his
client is not talking to others. His client a the moment can not deploy
anything to his server. However, you can get properties and
create/destroy the system, however sending the deployment descriptor is
failing right now, will worry about this later.

Jun/Satish: Satish worked on component model testing, found bugs and
fixing them. Worked on Ayla's endpoints and have a few questions. When
we create systems and issue lookup request. If the system is not there,
the Ayla's implementation returns the stack, whereas NEC returns the app
does not exist (return not the error but error string). What is the
standard for returning errors? Steve: there is no normative standards.
It seemed to early to mandate for each possible implementation. NI the
deployment API spec, there is a bit of no-normative stuff and he has a
sort of faults with some error code underneath, but it is not required
for various faults. We have a very minimal set of faults. It would be
good to have a consistent set of error codes. Satish will continue
component testing and on the Ayla's endpoint.

Quick checkpoint on where we are: Ayla: she feels that approach is good
that they will continue to discover many interoperability problems.
Everybody will continue with feedback. Not sure where we are wrt
completion. At least 80% completed of deployment API, 60% of component
model. Steve: quite enjoying interoperability tests. If he can get his
endpoint up he feels he is 50%, the other 50% is how does he measure
progress. One by one test, he will checking them off. There are issues
of the underlying implementation of WSRF, NEC and OurGrid have the same
WSRF implementation, they will not have interoperability with. Satish:
NEC and OurGrid have different implementations because of the lack of
the standard CDL implementation configuration file. Everything else,
from the deployment API perspective it is ok. 60% component model
completed and deployment APIs 75%.

The goal is to have all tests passed between endpoints. 

Update on the specs, Deployment APIs and Component model are official
GGF specs. CDL a few weeks away. 

Update on Wiki: it up, need to enable others.

Jun: updates on the OGSA meeting, explained our position, Jun and Hiro
will discuss the component model design for OGSA, in Tokyo. First Hiro
and Jun will discuss it and then later on they will discuss it with
Stuart. 

Next week, we will discuss the presentation at GGF. Especially
interoperability. 

Thanks,

Dejan.