[
https://issues.apache.org/jira/browse/SPARK-56929?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Ruifeng Zheng reassigned SPARK-56929:
-------------------------------------
Assignee: Yicong Huang
> Pass prefers_large_types when building expected schema for Arrow
> grouped/cogrouped map UDFs
> -------------------------------------------------------------------------------------------
>
> Key: SPARK-56929
> URL: https://issues.apache.org/jira/browse/SPARK-56929
> Project: Spark
> Issue Type: Sub-task
> Components: PySpark
> Affects Versions: 5.0.0
> Reporter: Yicong Huang
> Assignee: Yicong Huang
> Priority: Major
> Labels: pull-request-available
>
> *Bug*: In {{python/pyspark/worker.py}}, the per-field
> {{expected_cols_and_types}} schema built for the three Arrow map eval types
> * {{SQL_GROUPED_MAP_ARROW_UDF}}
> * {{SQL_GROUPED_MAP_ARROW_ITER_UDF}}
> * {{SQL_COGROUPED_MAP_ARROW_UDF}}
> calls {{to_arrow_type(col.dataType, timezone="UTC")}} per field without
> forwarding {{prefers_large_types=runner_conf.use_large_var_types}}, while the
> corresponding {{arrow_return_type}} (used to derive the actual result schema)
> *is* built with the flag.
> *Effect*: When {{spark.sql.execution.arrow.useLargeVarTypes=true}}, fields of
> type {{StringType}}/{{BinaryType}} are produced as Arrow
> {{large_string}}/{{large_binary}} in the result table but expected as regular
> {{string}}/{{binary}}. {{verify_arrow_result}} then raises a spurious
> {{RESULT_COLUMN_TYPES_MISMATCH}}.
> *Fix*: Pass {{prefers_large_types=runner_conf.use_large_var_types}} when
> constructing {{expected_cols_and_types}} in all three sites, matching the
> arrow_return_type construction immediately above.
> This is a pre-requisite for SPARK-56608 (Migrate verify_arrow_result checks
> into enforce_schema).
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
