spark-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Stephen Carman <>
Subject Re: Force inner join to shuffle the smallest table
Date Wed, 24 Jun 2015 19:32:58 GMT
Have you tried shuffle compression?

spark.shuffle.compress (true|false)

if you have a filesystem capable also I’ve noticed file consolidation helps disk usage a

spark.shuffle.consolidateFiles (true|false)


On Jun 24, 2015, at 3:27 PM, Ulanov, Alexander <<>>

It also fails, as I mentioned in the original question.

From: CC GP []
Sent: Wednesday, June 24, 2015 12:08 PM
To: Ulanov, Alexander
Subject: Re: Force inner join to shuffle the smallest table

Try below and see if it makes a difference:

val result = sqlContext.sql(“select big.f1, big.f2 from small inner join big on big.s=small.s
and big.d=small.d”)

On Wed, Jun 24, 2015 at 11:35 AM, Ulanov, Alexander <<>>

I try to inner join of two tables on two fields(string and double). One table is 2B rows,
the second is 500K. They are stored in HDFS in Parquet. Spark v 1.4.
val big = sqlContext.paquetFile(“hdfs://big”)
val small = sqlContext.paquetFile(“hdfs://small”)
val result = sqlContext.sql(“select big.f1, big.f2 from big inner join small on big.s=small.s
and big.d=small.d”)

This query fails in the middle due to one of the workers “disk out of space” with shuffle
reported 1.8TB which is the maximum size of my spark working dirs (on total 7 worker nodes).
This is surprising, because the “big” table takes 2TB disk space (unreplicated) and “small”
about 5GB and I would expect that optimizer will shuffle the small table. How to force Spark
to shuffle the small table? I tried to write “small inner join big” however it also fails
with 1.8TB of shuffle.

Best regards, Alexander

This e-mail is intended solely for the above-mentioned recipient and it may contain confidential
or privileged information. If you have received it in error, please notify us immediately
and delete the e-mail. You must not copy, distribute, disclose or take any action in reliance
on it. In addition, the contents of an attachment to this e-mail may contain software viruses
which could damage your own computer system. While ColdLight Solutions, LLC has taken every
reasonable precaution to minimize this risk, we cannot accept liability for any damage which
you sustain as a result of software viruses. You should perform your own virus checks before
opening the attachment.
View raw message