[
https://issues.apache.org/jira/browse/SPARK-56477?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Yicong Huang updated SPARK-56477:
---------------------------------
Description:
As part of the PySpark eval type refactoring effort (SPARK-55388), the
SQL_GROUPED_MAP_PANDAS_UDF processing logic should be made self-contained
within read_udfs() in worker.py.
Currently the logic is spread across the wrapper function
wrap_grouped_map_pandas_udf(), the GroupPandasUDFSerializer, and the mapper
block in read_udfs(). The refactoring should consolidate all UDF dispatch,
verification, and conversion logic into an inline handler in read_udfs(), using
a pure I/O serializer.
was:
As part of the PySpark eval type refactoring effort (SPARK-55388), the
SQL_COGROUPED_MAP_ARROW_UDF processing logic should be made self-contained
within read_udfs() in worker.py.
Currently the logic is spread across three locations:
- wrap_cogrouped_map_arrow_udf() for verification and UDF dispatch
- CogroupArrowUDFSerializer.dump_stream() (inherited) for column reordering and
struct wrapping
- The mapper block in read_udfs() for key/value extraction from batches
The refactoring should:
- Use ArrowStreamCoGroupSerializer as the pure I/O serializer (dual-stream
cogroup reading)
- Move all UDF dispatch, verification, column reordering, schema enforcement,
and struct wrapping into an inline handler in read_udfs()
- Delete wrap_cogrouped_map_arrow_udf and the mapper block
- Move extract_key_value_indexes earlier in read_udfs() so the new handler can
use it
Summary: Refactor SQL_GROUPED_MAP_PANDAS_UDF to be self-contained in
read_udfs (was: Refactor SQL_COGROUPED_MAP_ARROW_UDF to be self-contained in
read_udfs)
> Refactor SQL_GROUPED_MAP_PANDAS_UDF to be self-contained in read_udfs
> ---------------------------------------------------------------------
>
> Key: SPARK-56477
> URL: https://issues.apache.org/jira/browse/SPARK-56477
> Project: Spark
> Issue Type: Sub-task
> Components: PySpark
> Affects Versions: 4.2.0
> Reporter: Yicong Huang
> Priority: Major
>
> As part of the PySpark eval type refactoring effort (SPARK-55388), the
> SQL_GROUPED_MAP_PANDAS_UDF processing logic should be made self-contained
> within read_udfs() in worker.py.
> Currently the logic is spread across the wrapper function
> wrap_grouped_map_pandas_udf(), the GroupPandasUDFSerializer, and the mapper
> block in read_udfs(). The refactoring should consolidate all UDF dispatch,
> verification, and conversion logic into an inline handler in read_udfs(),
> using a pure I/O serializer.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
