[
https://issues.apache.org/jira/browse/SPARK-3095?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Josh Rosen resolved SPARK-3095.
-------------------------------
Resolution: Won't Fix
I'm going to close this as "Won't Fix", since changing the count() to
implicitly cache will break user programs that rely on the existing behavior.
> [PySpark] Speed up RDD.count()
> ------------------------------
>
> Key: SPARK-3095
> URL: https://issues.apache.org/jira/browse/SPARK-3095
> Project: Spark
> Issue Type: Improvement
> Components: PySpark
> Reporter: Davies Liu
> Assignee: Davies Liu
> Priority: Minor
>
> RDD.count() can fall back to RDD._jrdd.count(), when the RDD is not
> PipelineRDD.
> If the JavaRDD is serialized in batch mode, it's possible to skip the
> deserialization of chunks (except the last one), because they all have the
> same number of elements in them. There are some special cases that the chunks
> are re-ordered, so this will not work.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
