Hyukjin Kwon created SPARK-60976:
------------------------------------
Summary: Build applyInPandasWithState as an ApplyInPandasWithState
relation in Spark Connect Rust as PySpark does
Key: SPARK-60976
URL: https://issues.apache.org/jira/browse/SPARK-60976
Project: Spark
Issue Type: Bug
Components: Connect
Affects Versions: connect-rust-4.3.0
Reporter: Hyukjin Kwon
{{GroupedData.applyInPandasWithState}} in the Rust-backed Python client, and
{{GroupedData::apply_in_pandas_with_state}} in the Rust client, build a
{{GroupMap}} relation, which the server rejects for a pandas UDF with state:
{noformat}
[CONNECT_INVALID_PLAN.FUNCTION_EVAL_TYPE_NOT_SUPPORTED] The Spark Connect plan
is invalid. Function with EvalType: 208 is not supported
{noformat}
PySpark builds an {{ApplyInPandasWithState}} relation, with the output and
state schemas as JSON. So every streaming query of {{applyInPandasWithState}}
fails in the client, while it works in PySpark 4.3.0. Build the relation 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]
