Attempting to create a DataFrame using a BinaryType field fails under Python 3 because the underlying Pyrolite library is out of date. Spark appears to be using Pyrolite 4.9; this issue was fixed in Pyrolite 4.12. See original bug report and patch
Test case & output attached. I'm just a Python guy, not really sure how to build Spark / do classpath magic to test if this works correctly with updated Pyrolite.