10.1.3.3 fault management to simplify our complex error hospital pattern

Question

We have designed an error hospital pattern on 10.1.3.1, this consists of :
1. A loop around key components of each process which may fail and need retry (e.g. invokes, database operations)
2. Within each error loop a fault handler, this catches faults as they are thrown and passes them onto another BPEL error handling process (with a nested fault handler in case this invoke fails which spits error out to a file), the error handler can reply with retry (loop around again) or cancel (exit loop), if no response is received an automated retry occurs
3. An error service process which creates a human task for an error to be dealt with
4. A web app which allows users to process the tasks created by the error service, user can inspect data, see fault detail and mainly just select to cancel or retry the failed operation (including multi select if for example hundreds of processes had hit the same error)
This has been refined and is now operating nicely in production, but the development (items 1 and 2 above) and testing effort involved is high - frequently more than the actual business logic. In a 3GL this error handling would be simplified by having a nice reusable piece of code (proceudre or method call) that could go in the error handling loop, rather than putting a whole set of logic (invoke, pick, handling of faults in the error handling code, etc).
I'm aware of the fault management framework in 10.1.3.3 which does a subset of our functionality, we are now looking back at the error hospital we have created and wondering if we could simplify. The main stumbing block I see with the fault management framework is that we need to decide whether to do an automated retry after some interval or whether to await human action, we want to do both, await human activity but if they don't action it retry the operation. I'm thinking that we could achieve everything that we have got by a slight manipulation of the error management framework, as follows :
A. Define faults such that all faults cause human intervention.
B. Create a daemon process which scans for activities requiring human intervention and automatrically triggers a rety after a configurable period (so that things like database errors will get auto retried)
C. Adapt our web app to look for and operate on activities outstanding rather than workflow tasks.
Questions:
i) To support B/C above is there a java api this daemon can use or is it a question of updating the rdbms
ii) Am I underestimating what can be done with the java action fault policy ?
iii) To make our existing approach easier, is there a way of doing an invoke and pick operation actually within an embedded java routine to minimise the amount of code we have in each error handling block
Hope this makes sense, thanks in advance - there seem to be a lot of knowledgeable people out there in this forum.

Answer 1

Hi,
The next blogs might be interesting for you. It addresses your questions.
http://technology.amis.nl/blog/?p=2485
http://www.it-eye.nl/weblog/2007/09/10/oracle-bpel-10133-fault-policy-management/
Kind regards,
Harm