So, my understanding of the way PP currently works is that every node must not only have the required dependencies of the code that you're trying to distribute, but also the source of the code itself.  We currently manage this with version control software (SVN) so that all nodes can be updated to the latest versions easily.

Is there currently a method for actually dynamically distributing the required functions (or full files) to PP nodes when submitting a job?  I think this could prove useful, as it would support a generic cluster of PP nodes that (assuming the required dependencies already existed on the target node) could handle incoming jobs much more flexibly.  I imagine it's even theoretically possible to only distribute work out to nodes that are known to support the required dependencies and ignore nodes that do not.

Has anyone thought about or tried this?  I would certainly be willing to help out, however my skills with python are not currently up to par with the other languages I'm much more familiar with.