[
https://issues.apache.org/jira/browse/SPARK-13140?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
spencerlee updated SPARK-13140:
-------------------------------
Remaining Estimate: 10h (was: 168h)
Original Estimate: 10h (was: 168h)
> spark sql aggregate performance decrease
> -------------------------------------------
>
> Key: SPARK-13140
> URL: https://issues.apache.org/jira/browse/SPARK-13140
> Project: Spark
> Issue Type: Question
> Affects Versions: 1.6.0
> Reporter: spencerlee
> Original Estimate: 10h
> Remaining Estimate: 10h
>
> In our scenario, their are 30 + key columns with 60+ metric columns.
> our typical query is: select key1, key2, key3, key4, key5, sum(metric1),
> sum(metric2), sum(metric3).... sum(metric30) from table_name group by key1,
> key2, key3, key4, key5.
> I import a single parquet file(60M, about 250w+ records) into sparksql , and
> do the typical query with local mode. I found that, when I only aggregate 24
> metrics, the response time is about 4.81s, when I aggregate 25+ metrics, the
> response time is 45.9s, which is almost 10 times slower. that's obviously
> unreasonable.
> Is this a bug or need modify some configuration to tune the query?
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org
For additional commands, e-mail: issues-h...@spark.apache.org
