Question

My app needs to pull data from MULTIPLE sources. The following needs to be user defined

1. Order of the data sources
2. Within the Order, Conditions (if at all) determine whether to use a data source or skip it

Once the user has defined the above two, the app will assign a name to his “orchestration” that he subsequently will use to pull the data.

I want some guidance on

1. whether I am better of writing my own code to handle the order and conditions OR is this a state machine use case
2. If it is a state machine use case then which is the better open source project providing this support
3. Is there any specification for state machines? SCXML?
4. Which are the best open source implementations of state machine? Criteria (in this order): Based on standard, easy to debug, easy to embed, actively being developed, support for runtime creation of the state machine

Use case: This is a multi-tenant app. Each customer is going to design his own orcehstration and save it to the db. This we call as boarding the customer. Once a customer has been boarded, all his requests will use the orcestration he had saved earlier.

Now let us say there are 3 data sources A, B and C. Few of the possible orchestrations are

• Orchestrion 1: Pull A, Pull B and
  then Pull C in that ORDER
• Orchestrion 2: Pull A, if A returned error then pull B, pull B (so basically (A or C) and B)
• Crchestrion 3: Pull A, B and C. order DOESNT matter. all need to be pulled (may be in parallel)

Answer 1

Rules of thumb for using a state machine:

• you can diagram it as a start chart
• the process is likely the change but the algorithms executed at a given point (i.e. state) are not – still OK if what the algorithm does is impacted by state
• you want to extend the ability to someone who is not a programmer to alter the process

For your purpose I would look at the JBoss Business Process Management suite. You have the ability to define and manage the state machine using an XML representation, including a GUI editor for allowing others to alter it, and then pass it to a “virtual machine” for execution. The definition language provides a lot of different hooks for custom code execution and modelling more complex behaviors such as processes in parallel, iterative activities and scheduled events.

Update in Response to Comment #1

1. Speaking more abstractly, if you can draw it as boxes and lines (UML would qualify), then I think it qualifies.
2. Given the use case you’ve described I would model each data source as a State – you could break this one-to-one mapping for convenience sake (i.e. execute more than one data source per State). Transitions between States determine Order. Restrict Transitions based on Conditions.
3. That another user drives this is the most compelling reason for a state machine solution in my opinion. SCXML or jPDL (XML definition jPBM runs) effectively are serializations of the state transitions as XML. This XML can be manipulated in a GUI – both projects have Eclipse based GUIs. This allows data (i.e. Data Sources) to travel with process (State, Transitions and Conditions) specific to each user.
4. jPBM execution engine is embeddable, so I don’t think it’s as heavy weight as you suspect. jBPM’s language – jPDL – has good tooling support. I reviewed SCXML and it appears to do roughly the same things as jPDL, but documentation and maturity are lower. Also, jPBM allows defining what class to execute as a class path reference inline, whereas SCXML uses the indirection of XML namespaces.

I would study and compare SCXML and the jPDL specs. My impression is that SXCML is superset of jPDL, for instance jPDL has the ability to repeat tasks as part of its definition whereas SCXML does not appear so.