Completion and Exception Events

I wouldn't rely on load balancing your service object to 'fix'

the problem that you're seeing. I had much the same problem

on a program that I'm working on and have fixed it by instantiating

a unique object for each service request, invoking the asynchronous

method call on that object and waiting for the completion event for that

unique instance. This will prevent the problem of one session

'interfering' with another. Another solution would be to add

information

to the result of the asynchronous method call that would distinguish

it from other method calls. You could then check this value upon

receipt

of the completion event and determine whether or not it is truly the

correct completion event... but I think that's much less elegant than

the aforementioned solution.

-

From: kkroeger@csc.com

To: kamranamin@yahoo.com

Sent: 1/19/98 4:01:46 AM

Subject: Completion and Exception Events

Our application contains an environment visible service object which

instantiates several member attribute objects (call them POMs) used to

retrieve information from a persistent data store. Many client

partitions

will access this service object to retrieve information. It is likely

that

two clients will use the same POM to retrieve data at the same time.

For

each client request, the application performs a start task where

completion=event is specified to execute a method on the POM.

If the service object IS NOT load balanced and we use completion events

to

determine when the method that is 'started' completes by 'listening' for

a

POM.completionevent or a POM.exceptionevent, the following occurs:

Client 1 performs request.

Service object starts task for client 1.

Client 2 performs request.

Service object starts task for client 2.

task completes for client 1.

Service object receives event for completion for client 1.

Service object responds to completion event and reacts for client

2.

task completes for client 2.

Can this still happen if the Service object IS load balanced? From the

statistics we're seeing, our architect believes that messages are not

sent

to a service object until that service object is done with a previous

request. I believe that the requests are simply forwarded in a

round-robin

fashion.

Specifics:

Forte vsn:3.0.F.2

Server:AIX or NT 4.0

Client:NT 4.0

Thank you for your help

at 2007-6-29 9:21:19 >