[
https://issues.apache.org/jira/browse/SPARK-39931?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Apache Spark reassigned SPARK-39931:
------------------------------------
Assignee: (was: Apache Spark)
> Improve performance of applyInPandas for very small groups
> ----------------------------------------------------------
>
> Key: SPARK-39931
> URL: https://issues.apache.org/jira/browse/SPARK-39931
> Project: Spark
> Issue Type: Improvement
> Components: PySpark
> Affects Versions: 3.4.0
> Reporter: Enrico Minack
> Priority: Major
>
> Calling `DataFrame.groupby(...).applyInPandas(...)` for very small groups in
> PySpark is very slow. The reason is that for each group, PySpark creates a
> Pandas DataFrame and calls into the Python code. For very small groups, the
> overhead is huge, for large groups, it is smaller.
> Here is a benchmarks (seconds to groupBy(...).applyInPandas(...) 10m rows):
> ||groupSize||Scala||pyspark.sql||pyspark.pandas||
> |1024|8.9|20.9|7.8|
> |512|9.4|31.8|9.8|
> |256|9.3|47.0|20.2|
> |128|9.5|83.3|48.8|
> |64|9.5|137.8|91.9|
> |32|9.6|263.6|207.3|
> |16|9.6|525.9|261.5|
> |8|9.5|1,043|663.0|
> |4|9.8|2,073|1,168|
> |2|10.4|4,132|2,456|
> |1|11.3|8,162|4,642|
> *Idea to overcome this* is to call into Python side with a Pandas DataFrame
> that contains potentially multiple groups, then perform a Pandas
> DataFrame.groupBy(...).apply(...). With large groups, that Panadas DataFrame
> has all rows of single group, with small groups it contains many groups. This
> should improve efficiency.
> I have prepared a PoC to benchmark that idea but am struggling to massage the
> internal rows before sending them to Python. I have the key and the grouped
> values of each row as {{InternalRow}}s and want to turn them into a single
> {{InternalRow}} in such a way that in Python I can easily group by the key
> and get the same grouped values. For this, I think adding the key as a single
> column (struct) while leaving the grouped values as is should work best. But
> I have not found a way to get this working.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org
For additional commands, e-mail: issues-h...@spark.apache.org
