Hyukjin Kwon created SPARK-61755:
------------------------------------
Summary: Remove the protobuf requirement of the Python drop-in
Key: SPARK-61755
URL: https://issues.apache.org/jira/browse/SPARK-61755
Project: Spark
Issue Type: Improvement
Components: Connect
Affects Versions: connect-rust-4.3.0
Reporter: Hyukjin Kwon
The Python drop-in (pyspark-client-rust) needs protobuf at import time today,
though it is not a declared dependency: in a clean environment with only the
declared dependencies, `import pyspark.sql.connect.column` and
`pyspark.sql.connect.dataframe` fail with "No module named 'google'"
(`_column_reduce.py` imports google.protobuf at module level), and so do the
fits of the vendored `pyspark.ml` code, which build Spark Connect messages of
the generated `pyspark.sql.connect.proto`.
This removes the requirement: the protobuf imports that only the server's
Python worker needs become lazy, and `pyspark.sql.connect.proto` gives
Rust-backed Spark Connect message classes when protobuf is missing, so the
vendored `pyspark.ml` and pipelines code runs unchanged without it (no vendored
file is edited). A CI lane runs the SQL, ML, pickling, pandas and pipelines
tests with protobuf uninstalled.
--
This message was sent by Atlassian Jira
(v10.3.26#10030026)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]