[
https://issues.apache.org/jira/browse/SPARK-60962?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Hyukjin Kwon reassigned SPARK-60962:
------------------------------------
Assignee: Hyukjin Kwon
> Pickle user-defined types as Python objects, and make them of arguments, in
> the Python client as PySpark does
> -------------------------------------------------------------------------------------------------------------
>
> Key: SPARK-60962
> URL: https://issues.apache.org/jira/browse/SPARK-60962
> Project: Spark
> Issue Type: Bug
> Components: Connect
> Affects Versions: connect-rust-4.3.0
> Reporter: Hyukjin Kwon
> Assignee: Hyukjin Kwon
> Priority: Major
> Labels: pull-request-available
> Fix For: connect-rust-4.3.0
>
>
> In PySpark, a user-defined type is a Python object: it pickles as its class
> and its state, so cloudpickle sends a class that a notebook or script
> defines, in {{__main__}}, by value, and a UDT of parameters keeps them; and
> as {{DataType.__new__}} is {{object.__new__}}, a UDT whose {{__init__}} takes
> arguments is made with them. So a UDT of {{__main__}}, of parameters too, can
> be the return type of a UDF.
> The Rust-backed Python client pickles a UDT as its JSON, which the worker
> rebuilds with {{UserDefinedType.fromJson}}: that imports the class by its
> module, which {{__main__}} on the worker does not define, so {{udf(f,
> returnType=PointUDT())}} fails on the worker with {{UNSUPPORTED_OPERATION}},
> and a UDT whose {{__init__}} takes arguments or keeps state is rebuilt
> without them. A UDT whose {{__init__}} takes arguments cannot even be made:
> {{PointUDT(10.0)}} raises {{TypeError: DataType.__new__() takes 0 positional
> arguments}}. Pickle and make UDTs as PySpark does.
--
This message was sent by Atlassian Jira
(v10.3.26#10030026)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
