[
https://issues.apache.org/jira/browse/SPARK-61395?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Hyukjin Kwon resolved SPARK-61395.
----------------------------------
Resolution: Duplicate
Duplicate of SPARK-61402 (apache/spark-connect-rust#476), which pickles
VariantVal in PySpark's plain-class format and loads PySpark's pickles.
> Pickle VariantVal as PySpark does, and let its value and metadata be set, in
> the Python client
> ----------------------------------------------------------------------------------------------
>
> Key: SPARK-61395
> URL: https://issues.apache.org/jira/browse/SPARK-61395
> Project: Spark
> Issue Type: Bug
> Components: Connect
> Affects Versions: connect-rust-4.3.0
> Reporter: Hyukjin Kwon
> Priority: Major
> Fix For: connect-rust-4.3.0
>
>
> A follow-up of SPARK-61291, found in a post-merge review.
> VariantVal.__reduce__ now returns (copyreg.__newobj__, (cls,), (value,
> metadata, __dict__)), with a tuple state that PySpark's VariantVal cannot
> load (UnpicklingError: state is not a dictionary), so a VariantVal pickled by
> the client, e.g. in a UDF closure, does not load in the server's PySpark
> worker; a PySpark pickle does not load in the client either. value and
> metadata are also read-only, so a subclass that assigns self.value in
> __init__, as PySpark's allows, fails, and the constructor rejects values that
> PySpark stores as given.
--
This message was sent by Atlassian Jira
(v10.3.26#10030026)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]