Hyukjin Kwon created SPARK-61140:
------------------------------------
Summary: Send the renames of DataFrame.withColumnsRenamed in their
order, in the Rust core and the Python client, as PySpark does
Key: SPARK-61140
URL: https://issues.apache.org/jira/browse/SPARK-61140
Project: Spark
Issue Type: Bug
Components: Connect
Affects Versions: connect-rust-4.3.0
Reporter: Hyukjin Kwon
PySpark's DataFrame.withColumnsRenamed sends its renames in the order of their
dict (the repeated renames of WithColumnsRenamed), and the server applies them
in that order. So df.withColumnsRenamed({"id": "a", "a": "b"}) has the columns
["b"], and df.withColumnsRenamed({"a": "b", "id": "a"}) has ["a"].
The Rust core of spark-connect-rust keeps the renames of a WithColumnsRenamed
plan in a HashMap. So it sends them in the order of their hashes, not in the
order given, and the columns of such renames depend on that order. Apache
Spark's test_ordering_of_with_columns_renamed fails on the client: ['a'] !=
['b'].
The core should keep the renames in their order, and send them in it, 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]
