[jira] [Comment Edited] (SPARK-10731) The head() implementation of dataframe is very slow

["Yin Huai (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/SPARK-10731?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14901092#comment-14901092 ] 

Yin Huai edited comment on SPARK-10731 at 9/21/15 6:05 PM:
-----------------------------------------------------------

Looks like the problem is df.collect does not work well with limit. In Scala, {{df.limit(1).rdd.count()}} will also trigger the problem. When we call {{df.limit(1).rdd}}, we will launch a job to get 1 record for every partition. 


was (Author: yhuai):
Looks like the problem is df.collect does not work well with limit. In Scala, {{df.limit(1).rdd.count()}} will also trigger the problem.

> The head() implementation of dataframe is very slow
> ---------------------------------------------------
>
>                 Key: SPARK-10731
>                 URL: https://issues.apache.org/jira/browse/SPARK-10731
>             Project: Spark
>          Issue Type: Bug
>          Components: PySpark
>    Affects Versions: 1.4.1, 1.5.0
>            Reporter: Jerry Lam
>              Labels: pyspark
>
> {code}
> df=sqlContext.read.parquet("someparquetfiles")
> df.head()
> {code}
> The above lines take over 15 minutes. It seems the dataframe requires 3 stages to return the first row. It reads all data (which is about 1 billion rows) and run Limit twice. The take(1) implementation in the RDD performs much better.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscribe@spark.apache.org
For additional commands, e-mail: issues-help@spark.apache.org