The BPEL engine should be able to handle transactional synchronous
request / response invocations, both as a consumer and a provider.
For example if you invoke an EJB, the whole invocation can be
transactional, not only the request operation.
From the sequence diagram, it seems that such cases can not be handled.
Cheers,
Guillaume Nodet
Maciej Szefler wrote:
The attachment mentioned can be found at
http://pxe.intalio.org/public/iapi-20060414.tgz
Apparently, the list does not bigger attachments.
On Fri, 2006-04-14 at 12:47 -0400, Maciej Szefler wrote:
As to the last points, attached is a proposed "Integration API". The
purpose of this API is to facilitate integration of the BPEL engine with
service technologies / messaging frameworks (e.g. JBI, Axis, Mule,
etc.). This API is consistent with the current working of the PXE BPEL
engine (that is to say only trivial modifications to the bpel-runtime
will be required). This API is also consistent with JBI. Otherwise,
every effort has been made to keep it as simple as possible. Features
are:
* abstraction of thread management -- integration layer has full
responsibility for thread management, BPEL engine does not start
threads.
* abstraction of transaction management -- integration layer has full
responsibility for transaction management, JTA can be used, but not
necessary
* abstraction of scheduling -- integration layer provides the
transactional scheduler: permits leveraging of Quartz, J2EE scheduling
mechanisms
* abstraction of content -- content is opaque, content handling
mechanism TDB: permits specialized (non-XML) implementations
* abstraction of communication -- all communication with partners is
handled by the integration layer
* asynchronous communication model -- permits
request-only/request-response message exchange pattern with unreliable
(e.g. HTTP), reliable (e.g. WS-RM, JMS), and transacted (e.g. WS-TX
(shudder), JCA, JDBC) partners.
I will send sequence diagrams showing how a correct integration
implementation would handle invocations of different types of
partners.
We have not defined the content handling mechanism except to say that
content is opaque to the engine for the most part.
-- attachment left out --
-mbs
On Fri, 2006-04-14 at 08:49 -0600, Bill Flood wrote:
I missed two paragraphs on the front side of this commentrary...
- We need to determine the most capable and useful entry point API for
any common Ode implementation. It's probably productive to avoid
introducing any external dependencies to any bus architecture or
external protocols so that the core engine can be used by as many
external projects as possible while avoiding unnecessary dependencies.
This implies that the proprietary JBI-like bus currently used by the
PXE contribution would need to be abandoned.
- The implementation should support both a transactionally contiguous
invoke-to-endpoint transaction model as well as an asynchronous
coupling (callback-enable implementation) to the endpoint from the
engine (the latter requirement is specific to current PXE usage
scenarios). The transactional requirements imply some isolation of
the thread semantics, currently present in the PXE contribution, would
be necessary.
On 4/14/06, Bill Flood <[EMAIL PROTECTED]> wrote:
I'd like to try to recap some of the ongoing efforts and assumptions
in order to get some level of consensus for planning purposes...
In theory, we would like to set on a common implementation simply as a
matter of economy of scale. The devil is in the details so let's get
to some of those now.
Several of us, unfamiliar with the PXE implementation, are spinning up
on the approach taken in that implementation to determine what
components might be leveraged for a common Ode implementation. There
is a lot of productive dialog and education about both the motivation
and make up of the PXE contribution. We are not quite there yet but
there are several statements that we can make at this point in our
analysis.
- There appears to be a desire to remove the current PXE dependency
on Hibernate, which is licensed under the LGPL. The abstraction layer
developed for BPE could be useful for a common Ode implementation..
The BPE implementation relies on J2EE CMP for stateful persistence.
Stateless usage scenarios do not require persistence. EJB3 might be a
suitable candidate for state persistence where needed.
- A data abstraction layer developed for the BPE could be useful for
isolating the BPEL engine from a specific content model. This
probably has implications to the entry point API mentioned above.
There is an existing ODE thread of discussion on the topic.
- Two deployment models will be supported, one through pre-compilation
at design time (model currently supported by PXE), and the other
through pre-compilation at deployment time (model currently supported
by BPE). The prior scenario is useful for tools that rely on early
error detection specific to BPEL and the latter is useful when BPEL is
not the originating language.
- Neither implementation (BPE or PXE) supports BPEL 2.0 today so this
needs to be remedied in any common Ode implementation. within the
constraints of the WS-BPEL 2.0 specification timeframe.
- We will support optimizations for stateless business processes
within the implementation
- We desire to supply run-time engine debugging support that is
capable of referring back to the originating BPEL markup for purposes
of tooling support
- The management interfaces represented in the PXE implementation are
compelling
- The Jacob engine is interesting in itself because it appears to be
analogous to the focused approach taken in BPE of walking the object
model. We need to examine this further and consider the thread and
container isolation issues.
- Testing
a) Unit test suites can be contributed for BPEL 1.1 and BPEL
2.0 initially based on the PXE and BPE contributions and the final set
will be derived from the intersection of the two test suite
contributions.
b) A benchmarking framework should be created to assess
typical messaging scenarios and stress tests
I may have missed some points but I hope this is a useful starter.
