2009/4/15 Andre Merzky
Quoting [Steve Fisher] (Apr 15 2009):
1: with URL specified:
When you do
discoverer d ("glite://cern.ch"); list = d.discover (...);
and you have 5 adaptors (made up list)
glite mds local unicore condor
then you get a valid response from the glite adaptor. You get no errors from the other adaptors, and no time overhead to speak off.
Thanks for the concrete example. The gLite information system is based on the "BDII" and accepts ldap calls (ldap://...). According to http://globus.org/toolkit/mds/ the mds (in GT4) is WSRF based - so you need to talk SOAP not LDAP. You are sure to get a rude reply if you make an LDAP call to a WS or a SOAP call to an LDAP server. In some cases the adaptor could choose to ignore a URL with the wrong scheme (first part of the URL). gLite does not use gLite as the scheme part of the url but "ldap" and most services use "http" or "https" so there is no easy way for an adapter to recognise the validity of an info system URL.
I hope this is clear now.
Yes, a lot clearer, thanks! At last I think we are getting on the same page - thanks for your patience! :-)
So, a given URL can lead to problems if more than one adaptor feel able to handle it.
Yes!
Removing the URL surely would solve the problem. But, as said, the URL is optional anyway. What is the gain?
OK - I am to keep it in the spec
Also, the glite adaptor is free to define the schema glite://, and translate that internally to ldap://. We do that, for example, with gridftp, which is translated to gsiftp. Also, we do that for ssh:// URLs, which are translated into local file URLS (for sshfs mounted file systems).
I think that SAGA adapters should not require changes to the underlying middleware so I don't like this idea.
I am not saying that you should not remove the URL - just want to make sure I understand why the URL is a problem, really.
Having had this discussion I now feel that the problem with the SD is not serious however making proper use of the returned URLs is a problem for all the other SAGA components. It is however an implementation problem so I will move it to the developers list. Nobody commented on the CamelCased attribute names - I would propose to change all out attribute names in the spec to be CamelCased like the rest of SAGA and to change the Java and C++ now to match. We want to get the code released rather soon into gLite. Steve
Best, Andre.
-- Nothing is ever easy.