[
https://issues.apache.org/jira/browse/SPARK-61912?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Hyukjin Kwon reassigned SPARK-61912:
------------------------------------
Assignee: Hyukjin Kwon
> Copy the cached schema of to() where PySpark copies it, request the schema of
> an unpickled DataFrame, compare metadata as Python does, and keep the
> isStreaming of observe
> --------------------------------------------------------------------------------------------------------------------------------------------------------------------------
>
> Key: SPARK-61912
> URL: https://issues.apache.org/jira/browse/SPARK-61912
> Project: Spark
> Issue Type: Bug
> Components: Connect
> Affects Versions: connect-rust-4.4.0
> Reporter: Hyukjin Kwon
> Assignee: Hyukjin Kwon
> Priority: Major
> Labels: pull-request-available
> Fix For: connect-rust-4.4.0
>
>
> Follow-ups of SPARK-61875 from a round-2 review against PySpark 4.3.0:
> - The schema that to() of a value that is not a DataType sets survives only
> the set operations, where PySpark copies _cached_schema in alias, filter,
> limit, sort, hint, the repartitions, coalesce, distinct, dropDuplicates,
> sample, randomSplit and observe; to(LongType()) and other non-struct types
> cache nothing, where PySpark caches them.
> - An unpickled DataFrame sends no request and has no cached schema, where
> PySpark's sends one schema request at the unpickle and restores its cached
> schema; a DataFrame whose plan fails raises at the pickle, where PySpark's
> raises at the unpickle.
> - The set operations compare field metadata as JSON values, so 1 != 1.0, and
> StructType/StructField == differ the same way, where Python compares the
> dicts.
> - df.observe(Observation, ...) sends AnalyzePlan(is_streaming) at each call,
> where PySpark reads the DataFrame's cached isStreaming.
--
This message was sent by Atlassian Jira
(v10.3.26#10030026)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]