OK, I want to go back to basics for this post. What is the point of Windows Workflow? Here's the first paragraph of the preface to "Essential Windows Workflow Foundation"
"Windows Workflow Foundation is a general-purpose programming framework for creating reactive programs that act in response to stimulus from external entities. The basic characteristics of reactive programs is that they pause during their execution, for unknown amounts of time, awaiting input."
I've highlighted what I consider to be the important point here. A WF workflow may lay dormant for minutes, days or weeks. In the normal world of .NET code, implementing this kind of application can be troublesome. For a start, the application may shutdown and restart before the workflow resumes, so the application needs to cope with this possibility. This means the objects must be persisted and restarted with the correct state. A simple solution like making the executing thread sleep for the required amount of time isn't going to work, in terms of robustness or scalability (imagine 1000s of these workflows being executed at the same time).
So WF helps solves this problem. Without this feature, WF is really just a different approach to producing standard .NET code, either through the designer or generating declarative XML documents to explain how the code should execute. It might well be useful in some scenarios but it isn't really a killer reason to use the technology.
So with that in mind I created a very simple workflow to execute from Metastorm BPM, which you can see on the left. All it does is write to a text file, wait for ten minutes, then write to the text file again.
The first thing to ensure is to set the "Microsoft Workflow timeout (secs)" registry setting to something sensible. Unfortunately setting this to zero doesn't mean no timeout, it just uses the default 120 seconds timeout. So this needs setting to something very large. This is a DWORD value with a maximum value of 4294967295, which I calculate is approximately 49710 days, that hopefully is large enough for most workflows.
Executing the workflow worked fine in this scenario. The workflow idled for 10 minutes then resumed and completed.
So the next experiment was to start the workflow, shut down the engine and then restart the engine and see what happened. I waited... and waited... Nothing. I started another workflow just in case the workflow runtime hadn't been kicked off but still nothing. So there's a bit of a problem there, particularly if you want to run long running workflows.
As an aside, if you're running your workflow asynchronously then you won't be able to use most of the activities provided for interaction with Metastorm BPM. This is a shame. In particular the RaiseFlag activity would be exceedingly useful since it could be used to signal to the parent process that the workflow had completed. As it is, you'll need to do this with the ECL activities provided, which are more complicated to configure or write your own code.