I have a google cloud run service which is triggered via pubsub and grabs data from an azure api and places it in a gcp bucket using smart_open.

          with smart_open.open(destination, "wb") as fout:
            schema = pyarrow_adapters.json_schema_to_pyarrow_schema(self.desired_schema)
            data_frame.to_parquet(
                path=fout,
                engine="pyarrow",
                compression="gzip",
                schema=schema,
                use_compliant_nested_type=False,
            )

This saves 100s of thousads of files in a few hours time, each day there’s a few failures with this error which I cannot figure out how to fix.

Connection broken: IncompleteRead(15773 bytes read, 125469 more expected)’, IncompleteRead(15773 bytes read, 125469 more expected))

It’s always 15773 but with differing numbers of expected.

The metrics for cloudrun look good, and I have no idea how to debug this further.