I experienced this frustrating and very difficult to diagnose exception in a very large workflow, and for a long time, but I was eventually able to find a way to detect what node caused it, identify the cause, and eventually solve it permamently. I see that many cases seem to be - like mine - related specifically to workflows with database connections. So I hope this specific case may be able to help at least some others out, although the error seem to cover many issues. Here is my 2 cents:
How to detect the problematic node?
The exception doesnt offer any information about where in the workflow it was encountered. The way I found where the problem was specifically, was using the Browse Everywhere functionality introduced in v 10. I simply begain in the start and ended up where the data stopped being passed through, evident in the results output of the nodes.
What caused it?
My specific scenario related to reading and writing to the same database, specifically I used dynamic input some places (which has been mentioned by others), but I am not sure if that makes a difference. Obviously it would be a problem to write to the same tables due to table/record locking, violate constraints, etc. but all that was thought through and handled before I got this error. Yet it caused problems sending data both ways simultaneously, and I suspect it may be related to the way Alteryx handles alias connections, and specifically using the same alias connection for multiple purposes simultaneously. So in short the conclusion so far (in my case, can't generalize): don't send data both ways simultaneously using the same connection.
My solution (hopefully it applies to others)
Avoiding simultaneous two way data flow is simple. Simply strategically add block until done nodes to avoid it, like you would do in more obvious cases that includes file and record locks.
