[jira] [Commented] (SPARK-33143) Make SocketAuthServer socket timeout configurable

["Gabor Somogyi (Jira)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/SPARK-33143?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17231244#comment-17231244 ] 

Gabor Somogyi commented on SPARK-33143:
---------------------------------------

I've had a look at this issue and under some circumstances at heavy users one of the following calls took more than 15 seconds ending up in timeout:
 * getaddrinfo
 * socket
 * settimeout

It needs more investigation what's the rootcause of this issue (I'm on it).
 There are 2 suspects:
 * DNS is involved in getaddrinfo which is not responding or super slow
 * The OS is super slow somehow

Either way is super hard to provide workaround with hardcoded timeout. I tend to believe it would be good to make it configurable otherwise such intermittent issues could make temporary workaround extremely hard.

[~hyukjin.kwon] WDYT?

> Make SocketAuthServer socket timeout configurable
> -------------------------------------------------
>
>                 Key: SPARK-33143
>                 URL: https://issues.apache.org/jira/browse/SPARK-33143
>             Project: Spark
>          Issue Type: Improvement
>          Components: PySpark, Spark Core
>    Affects Versions: 2.4.7, 3.0.1
>            Reporter: Miklos Szurap
>            Priority: Major
>
> In SPARK-21551 the socket timeout for the Pyspark applications has been increased from 3 to 15 seconds. However it is still hardcoded.
> In certain situations even the 15 seconds is not enough, so it should be made configurable. 
> This is requested after seeing it in real-life workload failures.
> Also it has been suggested and requested in an earlier comment in [SPARK-18649|https://issues.apache.org/jira/browse/SPARK-18649?focusedCommentId=16493498&page=com.atlassian.jira.plugin.system.issuetabpanels%3Acomment-tabpanel#comment-16493498]
> In 
> Spark 2.4 it is under
> [PythonRDD.scala|https://github.com/apache/spark/blob/branch-2.4/core/src/main/scala/org/apache/spark/api/python/PythonRDD.scala#L899]
> in Spark 3.x the code has been moved to
> [SocketAuthServer.scala|https://github.com/apache/spark/blob/master/core/src/main/scala/org/apache/spark/security/SocketAuthServer.scala#L51]
> {code}
> serverSocket.setSoTimeout(15000)
> {code}
> Please include this in both 2.4 and 3.x branches.



--
This message was sent by Atlassian Jira
(v8.3.4#803005)

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