Nice writeup! How are you finding the Apache Beam python bindings for Dataflow? We are a Python shop, but have resorted to writing our production pipelines in Java since Apache Beam doesn't yet support realtime jobs.
Thanks! As you can see this was an incredibly simple use case of Beam, and my first try :). Given that all I wanted to do was map a custom function and then reduce with CountPerElement, Python did me just fine. I would imagine that resorting to Java land is in my future as well.