FR: Accept a file-like object in addition to a path in `fastparquet.write`

pyarrow accepts either a filename or a file-like object when writing a DataFrame. It would be very helpful if fastparquet also does this. I found https://github.com/dask/fastparquet/issues/215 which adds a special in-memory path name, but that seems specific to the dask-ecosystem.

Currently google-cloud-bigquery requires pyarrow rather than fastparquet to upload a DataFrame to BigQuery due to this issue.

[martindurant]

Basically, I think its better to deal with file-systems than file-like objects or path things.

[ospikovets]

@martindurant a code snippet in this stackoverflow comment is a good example of the use case for supporting a file-like object when writing a DataFrame.

def dataframe_to_s3(s3_client, input_datafame, bucket_name, filepath):
    out_buffer = BytesIO()
    input_datafame.to_parquet(out_buffer, index=False)
    s3_client.put_object(Bucket=bucket_name, Key=filepath, Body=out_buffer.getvalue())

It is definitely possible to do what you suggest here, it just doesn’t look as “elegant” as simply passing the file-like.

Any chance fastparquet will support that in the future?