Some workflow questions

Question

Hello All,
I have a couple of seemingly simple questions:
I have a session bean method with a void return type, and I have an EJB
Control for the session bean.
I can invoke this method only with a Control Send node, since a Control Send
and Return node does not see
this method, and when I drag the method from the EJB Control and drop it
onto a Control Send and Return
node, it gets converted to a Control Send node.
My questions is:
1. Is the method invoked synchronouly, meaning, does it wait until the
actual EJB method returns, or is
it just put into a message buffer?
2. Is this method participating in the implicit transaction the previous
(Control Send and Return) node is part of?
3. If the answer to either of the following two questions is no, what can I
do, to ensure the synchronous invocation and the participation in the
implicit transaction?
A fourth question:
When I start a workflow with a Client Request and Return node, and I have a
couple of
synchronous invocation nodes after the Client Request node, to a Worklist
Control and a
Session EJB, they form an implicit transaction before the Client Return
node.
If I throw an exception in any of these nodes, and don't catch the
exception, what happens?
1. The workflow aborts, all method calls on the previous nodes are rolled
back.
2. The method calls on the previous nodes are rolled back, the workflow does
not abort.
Which one? If the second, what can I do with that workflow?
A fifth question:
What exactly happens in the following case:
My workflow processing is blocking (waiting for, does not matter that
actively or passively) at a Control Receive node waiting for a callback.
What happens if an invocation of a Client Request related method or a
callback on another Control or another callback no the same Control comes
in?
1. Will it be buffered in a Message Buffer and be retrievable when the
processing leads to a place
where it will be read from the buffer (appropriate node for the method call,
or event choice containing the appropriate call) and calling process
returns?
2. Calling process will block until the method call can be processed at a
similar node?
3. Will it be silently dropped?
4. Will an exception be thrown in the calling process? If yes, what
Exception?
5. Will something weird happen, and the receiving process jump to some
totally improper place in the execution
of the workflow?
Thanks for the (hopefully fast) answers in advance,
Regards,
Robert Varga

Answer 1

Hello Steven,
Some of my questions was answered in the documentation in the web service
areas. Most of them were not.
When here I state that something is not covered in the documentation, that
means that I was not able to find it stated clearly for the thing I was
searching at. It may be cleared in other places of the documentation
regarding a similar thing. It may also be possible that I was simply
overlooked it, after days of browsing the documentation. I would quite
happily accept it if someone just bumped my nose into the particular place
in the docs that answered these questions mentioned here.
So let me just explain what I did not find in the documentation:
Unfortunately the documentation is more geared to a web-service viewpoint,
which, I agree, has much common with business processes (jpd files), but the
business process design view also contains some differences to a normal .jws
web-service.
The biggest difference is, that when you design a .jpd, you assume to have
something that behaves like an instruction pointer. That is supposed to keep
track which stateful node the processing is currently blocked at.
The .jws web-services do not have this in the design view.
Now the fourth and fifth question was specifically about the behaviour and
existence of this instruction pointer-like behaviour, and these questions
were not answered in the documentation at all.
It is nowhere specified what happens to a business process instance if it
throws an exception, and it has no exception paths handling that exception.
I assume that it remains in the state that it was in before the invocation
that triggered the operations that resulted in the exception thrown.
It is also not covered in the documentation whether this instruction pointer
is really there and how it exactly behaves. The only thing in the
documentation which I was able to find was a side note, that if a message
buffer is in place, it is not guaranteed, that the retrieval of the
invocations from the message buffer will be in the order of the invocations
themselves. It is even a bit vague, whether message buffers are per method
or per jws/jpd instance.
But it is definitely not covered, what happens in case of an out-of-order
invocation to a business process. In case of a jws, it is clear that you are
supposed to handle if an invocation comes, that is not supposed to come at
all.
But in case of a business process (jpd), it is not covered in the docs, what
would happen with the invocation, what would happen with the calling
process, and what would happen with the called process. And this is
important information for being able to plan for unexpected things, or even
expected things, for example when a callback is timed out, but it comes in
later, when we are blocking at a different control, or even at the same
control, but in a different iteration.
It is also not clearly covered, if and if yes, how one can create a
syncronous callback, and what will be in the same transaction with the
callback invocation (only the Control Receive node, receiving the callback,
or the following nodes as well?). Is disabling the message-buffer on the
Client Response node enough?
It is possibly covered somewhere what the case is with the void EJB methods,
but I was not able to find it at all. EJB method invocations are supposed to
be synchronous, but void methods are callable upon with Control Send only
which is asynchronous as far as I was able to determine. I was not able to
ascertain what is the real case, and neither was I able to find a properly
documented way to ensure synchronous invocation to a void method on an EJB
control (which explicitely states that this happens).
Regards,
Robert Varga
"Steven Ostrowski" <[email protected]> wrote in message
news:[email protected]...
I'm pretty sure the answer to each of these questions is in the WLI
documentation.
There is a section on how to enable message buffers to accomplish
exactly what you are talking about. There is also a section on what
creates an implicit transaction and what does not. I believe control
sends start one, but you should reference the docs.
Robert Varga wrote:
Hello All,
I have a couple of seemingly simple questions:
I have a session bean method with a void return type, and I have an EJB
Control for the session bean.
I can invoke this method only with a Control Send node, since a Control
Send
and Return node does not see
this method, and when I drag the method from the EJB Control and drop it
onto a Control Send and Return
node, it gets converted to a Control Send node.
My questions is:
1. Is the method invoked synchronouly, meaning, does it wait until the
actual EJB method returns, or is
it just put into a message buffer?
2. Is this method participating in the implicit transaction the previous
(Control Send and Return) node is part of?
3. If the answer to either of the following two questions is no, whatcan I
do, to ensure the synchronous invocation and the participation in the
implicit transaction?
A fourth question:
When I start a workflow with a Client Request and Return node, and Ihave a
couple of
synchronous invocation nodes after the Client Request node, to aWorklist
Control and a
Session EJB, they form an implicit transaction before the Client Return
node.
If I throw an exception in any of these nodes, and don't catch the
exception, what happens?
1. The workflow aborts, all method calls on the previous nodes arerolled
back.
2. The method calls on the previous nodes are rolled back, the workflowdoes
not abort.
Which one? If the second, what can I do with that workflow?
A fifth question:
What exactly happens in the following case:
My workflow processing is blocking (waiting for, does not matter that
actively or passively) at a Control Receive node waiting for a callback.
What happens if an invocation of a Client Request related method or a
callback on another Control or another callback no the same Controlcomes
in?
1. Will it be buffered in a Message Buffer and be retrievable when the
processing leads to a place
where it will be read from the buffer (appropriate node for the methodcall,
or event choice containing the appropriate call) and calling process
returns?
2. Calling process will block until the method call can be processed ata
similar node?
3. Will it be silently dropped?
4. Will an exception be thrown in the calling process? If yes, what
Exception?
5. Will something weird happen, and the receiving process jump to some
totally improper place in the execution
of the workflow?
Thanks for the (hopefully fast) answers in advance,
Regards,
Robert Varga