Merge pull request #865 from tgravescs/fixtmpdir

Spark on Yarn should use yarn approved directories for spark.local.dir and tmp

3 files changed, 26 insertions, 0 deletions

diff --git a/yarn/src/main/scala/spark/deploy/yarn/ApplicationMaster.scala b/yarn/src/main/scala/spark/deploy/yarn/ApplicationMaster.scala
index 15dbd1c0fb..0f3b6bc1a6 100644
--- a/yarn/src/main/scala/spark/deploy/yarn/ApplicationMaster.scala
+++ b/yarn/src/main/scala/spark/deploy/yarn/ApplicationMaster.scala
@@ -47,6 +47,9 @@ class ApplicationMaster(args: ApplicationMasterArguments, conf: Configuration) e
   private var isFinished:Boolean = false
 
   def run() {
+    // setup the directories so things go to yarn approved directories rather
+    // then user specified and /tmp
+    System.setProperty("spark.local.dir", getLocalDirs())
     
     appAttemptId = getApplicationAttemptId()
     resourceManager = registerWithResourceManager()
@@ -89,6 +92,21 @@ class ApplicationMaster(args: ApplicationMasterArguments, conf: Configuration) e
 
     System.exit(0)
   }
+
+  /** Get the Yarn approved local directories. */
+  private def getLocalDirs(): String = {
+    // Hadoop 0.23 and 2.x have different Environment variable names for the
+    // local dirs, so lets check both. We assume one of the 2 is set.
+    // LOCAL_DIRS => 2.X, YARN_LOCAL_DIRS => 0.23.X
+    val localDirs = Option(System.getenv("YARN_LOCAL_DIRS"))
+      .getOrElse(Option(System.getenv("LOCAL_DIRS"))
+      .getOrElse(""))
+
+    if (localDirs.isEmpty()) {
+      throw new Exception("Yarn Local dirs can't be empty")
+    }
+    return localDirs
+  }
   
   private def getApplicationAttemptId(): ApplicationAttemptId = {
     val envs = System.getenv()
diff --git a/yarn/src/main/scala/spark/deploy/yarn/Client.scala b/yarn/src/main/scala/spark/deploy/yarn/Client.scala
index e84fb7c985..eb2a8cc642 100644
--- a/yarn/src/main/scala/spark/deploy/yarn/Client.scala
+++ b/yarn/src/main/scala/spark/deploy/yarn/Client.scala
@@ -223,6 +223,10 @@ class Client(conf: Configuration, args: ClientArguments) extends YarnClientImpl
     // Add Xmx for am memory
     JAVA_OPTS += "-Xmx" + amMemory + "m "
 
+    JAVA_OPTS += " -Djava.io.tmpdir=" + new Path(Environment.PWD.$(),
+                                                 YarnConfiguration.DEFAULT_CONTAINER_TEMP_DIR)
+
+
     // Commenting it out for now - so that people can refer to the properties if required. Remove it once cpuset version is pushed out.
     // The context is, default gc for server class machines end up using all cores to do gc - hence if there are multiple containers in same
     // node, spark gc effects all other containers performance (which can also be other spark containers)
diff --git a/yarn/src/main/scala/spark/deploy/yarn/WorkerRunnable.scala b/yarn/src/main/scala/spark/deploy/yarn/WorkerRunnable.scala
index 3e007509e6..0e1fd9b680 100644
--- a/yarn/src/main/scala/spark/deploy/yarn/WorkerRunnable.scala
+++ b/yarn/src/main/scala/spark/deploy/yarn/WorkerRunnable.scala
@@ -75,6 +75,10 @@ class WorkerRunnable(container: Container, conf: Configuration, masterAddress: S
     if (env.isDefinedAt("SPARK_JAVA_OPTS")) {
       JAVA_OPTS += env("SPARK_JAVA_OPTS") + " "
     }
+
+    JAVA_OPTS += " -Djava.io.tmpdir=" + new Path(Environment.PWD.$(),
+                                                 YarnConfiguration.DEFAULT_CONTAINER_TEMP_DIR)
+
     // Commenting it out for now - so that people can refer to the properties if required. Remove it once cpuset version is pushed out.
     // The context is, default gc for server class machines end up using all cores to do gc - hence if there are multiple containers in same
     // node, spark gc effects all other containers performance (which can also be other spark containers)