[
https://issues.apache.org/jira/browse/SPARK-58923?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Yicong Huang updated SPARK-58923:
---------------------------------
Affects Version/s: 4.4.0
(was: 5.0.0)
> Bound the output Arrow batch size of Python UDF workers
> -------------------------------------------------------
>
> Key: SPARK-58923
> URL: https://issues.apache.org/jira/browse/SPARK-58923
> Project: Spark
> Issue Type: Improvement
> Components: PySpark
> Affects Versions: 4.4.0
> Reporter: Yicong Huang
> Priority: Major
>
> An Arrow-based Python UDF worker can emit a single oversized output
> RecordBatch (for example, applyInPandas produces one batch per group, so a
> large group yields one large batch), which strains the JVM and downstream
> operators. We need a Python-side (worker-side) mechanism to resize such
> output batches down toward a configurable byte limit before they are sent to
> the JVM.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
