How to create a Dataflow pipeline from Pub/Sub to GCS in Python

I ran into this same error, and found a workaround, but not a fix:

TypeError: Cannot convert GlobalWindow to apache_beam.utils.windowed_value._IntervalWindowBase [while running 'test-file-out/Write/WriteImpl/WriteBundles']

running locally with DirectRunner and on dataflow with DataflowRunner.

Reverting to apache-beam[gcp]==2.9.0 allows my pipeline to run as expected.

I have had so much trouble trying to figure out the

TypeError: Cannot convert GlobalWindow to apache_beam.utils.windowed_value._IntervalWindowBase [while running 'generatedPtransform-1090']

There seems to be something with the WriteToText after beam 2.9.0 (I am using beam 2.14.0, python 3.7)

| "Output" >> beam.io.WriteToText("<GCS path or local path>"))

What made it work for me was removing the pipeline part and radding a custom DoFn:

class WriteToGCS(beam.DoFn):
    def __init__(self):
        self.outdir = "gs://<project>/<folder>/<file>"

    def process(self, element):
        from apache_beam.io.filesystems import FileSystems # needed here
        import json
        writer = FileSystems.create(self.outdir + '.csv', 'text/plain')
        writer.write(element)
        writer.close()

and in the pipeline add:

| 'Save file' >> beam.ParDo(WriteToGCS())