Hyukjin Kwon created SPARK-61035:
------------------------------------
Summary: Send the resource profile of mapInPandas and mapInArrow
in the Python client as PySpark does
Key: SPARK-61035
URL: https://issues.apache.org/jira/browse/SPARK-61035
Project: Spark
Issue Type: Bug
Components: Connect
Affects Versions: connect-rust-4.3.0
Reporter: Hyukjin Kwon
In PySpark Connect, {{df.mapInPandas(func, schema, barrier, profile)}} and
{{df.mapInArrow(func, schema, barrier, profile)}} send the id of the
{{ResourceProfile}} {{profile}}, which they create on the server, as the
{{profile_id}} of their {{MapPartitions}}, so that the stage runs with the
resources of the profile (stage-level scheduling). Against a server that does
not support the profile, as one in local mode, an action of the DataFrame fails
with {{SparkException: TaskResourceProfiles are only supported for Standalone,
Yarn and Kubernetes cluster for now when dynamic allocation is disabled.}}
The Rust-backed Python client takes {{profile}} and ignores it: the stage runs
without the resources of the profile, and the action succeeds against a server
that does not support it. Send the profile 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]
