[omniORB-dev] ImR idl proposal

Thomas Lockhart lockhart@fourpalms.org
Wed, 22 Jan 2003 16:35:29 -0800


More comments:

The ImplementationRepositoryAdmin interface is used by meta-clients to 
manipulate the ImR. The ImplementationRepository interface is used by 
servers to interact with the ImR. The ProcessManager interface is used 
by the ImR to interact with a server.

4b) The ImplementationRepositoryAdmin interface is not used by simple 
servers, so should be split out to allow servers to ignore the definitions.

4c) The ImplementationRepository and ProcessManager interfaces are used 
between ImR and server, so could be mixed into the same IDL file; no 
need to decouple them. Unless one could imagine a very simple server 
which does not bother to implement any ProcessManager functionality, 
which would then force the ImR to use more brute-force methods 
(suggested by Duncan earlier) to manipulate the server. So these should 
be separated also.

4d) Same with the LoadManager interface. I'm not certain where the 
server object for load information is required to live; I could imagine 
that one could get fine-grained statistics by distributing some server 
functions into application servers then asking the ImR to collate these 
for an integrated report.

5) registerProcess() could return a "process object", which can be used 
subsequently for operations like addProgramToProcess(). Seems to be 
cleaner and more CORBA-like than stringifying everything.

6) There is no corresponding unregisterProcess(). The IDL as-written 
allows "programs" to be added and removed, but not "processes".

More comments later :)

                   - Tom