spark-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Sean Owen (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (SPARK-3875) Add TEMP DIRECTORY configuration
Date Sat, 24 Jan 2015 14:08:34 GMT

    [ https://issues.apache.org/jira/browse/SPARK-3875?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14290602#comment-14290602
] 

Sean Owen commented on SPARK-3875:
----------------------------------

You can already set {{java.io.tmpdir}}, without making a new property, and that will control
where all Java code puts its temp files. There is already {{spark.local.dir}} which sounds
like exactly what you're suggesting. This gets set to a big fast disk because it's where things
like shuffle files go.

Is the question here perhaps whether a few bits of code that don't use {{spark.local.dir}}
should use it? Yes, it looks like it's used by {{HttpBroadcast.scala}} and downloading dependencies.

> Add TEMP DIRECTORY configuration
> --------------------------------
>
>                 Key: SPARK-3875
>                 URL: https://issues.apache.org/jira/browse/SPARK-3875
>             Project: Spark
>          Issue Type: Improvement
>          Components: Spark Core
>    Affects Versions: 1.1.0
>            Reporter: Patrick Liu
>
> Currently, the Spark uses "java.io.tmpdir" to find the /tmp/ directory.
> Then, the /tmp/ directory is used to 
> 1. Setup the HTTP File Server
> 2. Broadcast directory
> 3. Fetch Dependency files or jars by Executors
> The size of the /tmp/ directory will keep growing. The free space of the system disk
will be less.
> I think we could add a configuration "spark.tmp.dir" in conf/spark-env.sh or conf/spark-defaults.conf
to set this particular directory. Let's say, set the directory to a data disk.
> If "spark.tmp.dir" is not set, use the default "java.io.tmpdir"



--
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


Mime
View raw message