search

OpenMined / PipelineDP

PipelineDP is a Python framework for applying differentially private aggregations to large datasets using batch processing systems such as Apache Spark, Apache Beam, and more.
https://pipelinedp.io/
Apache License 2.0
270 stars 75 forks source link

Spark Connect Support #513

Open wchau opened 6 months ago

wchau commented 6 months ago

Feature Description

Spark Connect Support

Is your feature request related to a problem?

In Spark Connect, RDD is not supported, so PipelineDP does not work. See https://github.com/apache/spark/blob/master/python/pyspark/sql/connect/dataframe.py

What alternatives have you considered?

N/A

Additional Context

Add any other context or screenshots about the feature request here.

dvadym commented 6 months ago

Thanks for filing the issue!

There is already an experimental Spark DataFrame support:

  1. End2End example, building DP query in this example.
  2. QueryBuilder class API (with an example in the docstring): the main API for supporting DataFrames.

We're planning to make it as official API in the next release in ~2 months (including adding documentation for this API).

Any feedback on the this API is welcome.

wchau commented 6 months ago

Hey Vadym,

Is there a reason why you use RDD instead of DataFrame directly? I think the QueryBuilder tries to convert to RDD, which Spark Connect does not support.

Thanks!

dvadym commented 6 months ago

Ah, I see, Spark Connect doesn't have RDD at all.

The main reason why RDD is used is that the direct handling of Spark DataFrame is not yet impelemented. The PipelineDP main logic is pretty agnostic to the input colllection type. So it should be reasonable simple to extend to DataFrames. I'll check how it can be done.