How to properly load non-PyPi dependencies in GCP Dataflow using Beam Python SDK?

Question

My Dataflow job is dependent on a private Python library written by our Data Science team. DirectRunner works fine (per usual), but it hangs when I run the job on DataflowRu