I am attempting to use dataflow to read a pubsub message and write it to big query. I was given alpha access by the Google team and have gotten the provided examples working
Data written to Python SDK's BigQuery sink should be in the form of a dictionary where each key of the dictionary gives a field of the BigQuery table and corresponding value gives the value to be written to that field. For a BigQuery RECORD type, value itself should be a dictionary with corresponding key,value pairs.
I filed a JIRA to improve documentation of corresponding python module in Beam: https://issues.apache.org/jira/browse/BEAM-3090