[
https://issues.apache.org/jira/browse/SPARK-16182?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15347869#comment-15347869
]
Sean Owen commented on SPARK-16182:
-----------------------------------
OK that sounds more reasonable, go ahead.
> Utils.scala -- terminateProcess() should call Process.destroyForcibly() if
> and only if Process.destroy() fails
> --------------------------------------------------------------------------------------------------------------
>
> Key: SPARK-16182
> URL: https://issues.apache.org/jira/browse/SPARK-16182
> Project: Spark
> Issue Type: Bug
> Components: Spark Core
> Affects Versions: 1.6.1
> Environment: OSX El Capitan (java "1.8.0_65"), Oracle Linux 6 (java
> 1.8.0_92-b14)
> Reporter: Christian Chua
> Priority: Critical
>
> Spark streaming documentation recommends application developers create static
> connection pools. To clean up this pool, we add a shutdown hook.
> The problem is that in spark 1.6.1, the shutdown hook for an executor will be
> called only for the first submitted job. (on the second and subsequent job
> submissions, the shutdown hook for the executor will NOT be invoked)
> problem not seen when using java 1.7
> problem not seen when using spark 1.6.0
> looks like this bug is caused by this modification from 1.6.0 to 1.6.1:
> https://issues.apache.org/jira/browse/SPARK-12486
> steps to reproduce the problem :
> 1.) install spark 1.6.1
> 2.) submit this basic spark application
> import org.apache.spark.{ SparkContext, SparkConf }
> object MyPool {
> def printToFile( f : java.io.File )( op : java.io.PrintWriter => Unit ) {
> val p = new java.io.PrintWriter(f)
> try {
> op(p)
> }
> finally {
> p.close()
> }
> }
> def myfunc( ) = {
> "a"
> }
> def createEvidence( ) = {
> printToFile(new java.io.File("/var/tmp/evidence.txt")) { p =>
> p.println("the evidence")
> }
> }
> sys.addShutdownHook {
> createEvidence()
> }
> }
> object BasicSpark {
> def main( args : Array[String] ) = {
> val sparkConf = new SparkConf().setAppName("BasicPi")
> val sc = new SparkContext(sparkConf)
> sc.parallelize(1 to 2).foreach { i => println("f : " +
> MyPool.myfunc())
> }
> sc.stop()
> }
> }
> 3.) you will see that /var/tmp/evidence.txt is created
> 4.) now delete this file
> 5.) submit a second job
> 6.) you will see that /var/tmp/evidence.txt is no longer created on the
> second submission
> 7.) if you use java 7 or spark 1.6.0, the evidence file will be created on
> the second and subsequent submits
--
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
