🚀 go-pugleaf

> I used ORBagentAddr in the java.util.Properties
> for ORB.init() and I had the services running on two different
> machines using the same port number.  I found that sometimes I bind to
> machine A while sometimes I bind to machine B.  The service names on
> both machines are the same but the logic of them are different so that
> I can differentiate.
>

Surely you need to differentiate between communicating with osagents
and communicating with the CORBA objects themselves. The ORBagentAddr
can be used to specify that a VisiBroker process communicates with an
osagent on a particular host (communicate directly with that agent
using p2p UDP instead of finding an agent using UDP broadcast) but
that doesn't say anything about the IORs that the agent will return.

If you run only one agent, then you will definitely round-robin
between the two services. For example, ORBagentAddr points to an
osagent on machine A. The osagent on this host holds references to
services with identical names and repIds on machines A and B. Each
time a reference is requested, it gives out A, B, A, B etc.

If you run an osagent on both hosts, then you might get what you need,
but that is less predictable (depends on startup order, network
latencies etc.) If the service on A is registered with an osagent on
A, the service on B is registered with an osagent on B, pointing
ORBagentAddr to the particular host will give you the server you need.

However, why register these 2 instances of the service with the
osagent under identical names if you want to distinguish between them?
Why not just register them with different names, and bind to the name
you want? When you do a "bind" you also have the option of specifying
that you want a reference to an object on a particular host (see the
overloaded versions of "bind" that allow you to specify a host address
and BindOptions).

davisglynne@hotmail.com (Glynne Davis) wrote in message news:<2a88fc0b.0208120359.102536b2@posting.google.com>...
> > I used ORBagentAddr in the java.util.Properties
> > for ORB.init() and I had the services running on two different
> > machines using the same port number.  I found that sometimes I bind to
> > machine A while sometimes I bind to machine B.  The service names on
> > both machines are the same but the logic of them are different so that
> > I can differentiate.
> >
> 
> Surely you need to differentiate between communicating with osagents
> and communicating with the CORBA objects themselves. The ORBagentAddr
> can be used to specify that a VisiBroker process communicates with an
> osagent on a particular host (communicate directly with that agent
> using p2p UDP instead of finding an agent using UDP broadcast) but
> that doesn't say anything about the IORs that the agent will return.
> 
> If you run only one agent, then you will definitely round-robin
> between the two services. For example, ORBagentAddr points to an
> osagent on machine A. The osagent on this host holds references to
> services with identical names and repIds on machines A and B. Each
> time a reference is requested, it gives out A, B, A, B etc.
> 
> If you run an osagent on both hosts, then you might get what you need,
> but that is less predictable (depends on startup order, network
> latencies etc.) If the service on A is registered with an osagent on
> A, the service on B is registered with an osagent on B, pointing
> ORBagentAddr to the particular host will give you the server you need.

This is exactly my question because it did not give me the server I
need sometimes.  I felt surprised.  This seems unanswered still.


> 
> However, why register these 2 instances of the service with the
> osagent under identical names if you want to distinguish between them?
> Why not just register them with different names, and bind to the name
> you want? When you do a "bind" you also have the option of specifying
> that you want a reference to an object on a particular host (see the
> overloaded versions of "bind" that allow you to specify a host address
> and BindOptions).

Maybe I did not mention clear enough about the whole picture.  This
was just a testing.  The reason for do this testing was because one
day, a developer by mistake started her osagent with the services
having the same names as in production environment using the
production port number on her local machine.  This caused issues on
production because sometimes it talked to the correct production
osagent but sometimes it talked to the osagent on her machine which
did not have production logic.  Until that happened, we thought that
by using ORBagentAddr, we can always talk to just a particular
osagent.  This seemed to be not true.  So I'm just wondering if this
is a know bug in visibroker.  Or I have missed something.

Thanks.