[WIP] SPARK-1676: Cache Hadoop UGIs by default to prevent FileSystem leak

Move the doAs in Executor higher up so that we only have 1 ugi and aren't leaking filesystems.
Fix spark on yarn to work when the cluster is running as user "yarn" but the clients are launched as the user and want to read/write to hdfs as the user.

Note this hasn't been fully tested yet.  Need to test in standalone mode.

Putting this up for people to look at and possibly test.  I don't have access to a mesos cluster.

This is alternative to #607

Author: Thomas Graves <[email protected]>

Closes #621 from tgravescs/SPARK-1676 and squashes the following commits:

244d55a [Thomas Graves] fix line length
44163d4 [Thomas Graves] Rework
9398853 [Thomas Graves] change to have doAs in executor higher up.

Author: tgravescs

core/src/main/scala/org/apache/spark/deploy/SparkHadoopUtil.scala

@@ -24,25 +24,36 @@ import org.apache.hadoop.mapred.JobConf
 import org.apache.hadoop.security.Credentials
 import org.apache.hadoop.security.UserGroupInformation
 
-import org.apache.spark.{SparkContext, SparkException}
+import org.apache.spark.{Logging, SparkContext, SparkException}
 
 import scala.collection.JavaConversions._
 
 /**
  * Contains util methods to interact with Hadoop from Spark.
  */
-class SparkHadoopUtil {
+class SparkHadoopUtil extends Logging {
   val conf: Configuration = newConfiguration()
   UserGroupInformation.setConfiguration(conf)
 
-  def runAsUser(user: String)(func: () => Unit) {
+  /**
+   * Runs the given function with a Hadoop UserGroupInformation as a thread local variable
+   * (distributed to child threads), used for authenticating HDFS and YARN calls.
+   *
+   * IMPORTANT NOTE: If this function is going to be called repeated in the same process
+   * you need to look https://issues.apache.org/jira/browse/HDFS-3545 and possibly
+   * do a FileSystem.closeAllForUGI in order to avoid leaking Filesystems
+   */
+  def runAsSparkUser(func: () => Unit) {
+    val user = Option(System.getenv("SPARK_USER")).getOrElse(SparkContext.SPARK_UNKNOWN_USER)
     if (user != SparkContext.SPARK_UNKNOWN_USER) {
+      logDebug("running as user: " + user)
       val ugi = UserGroupInformation.createRemoteUser(user)
       transferCredentials(UserGroupInformation.getCurrentUser(), ugi)
       ugi.doAs(new PrivilegedExceptionAction[Unit] {
         def run: Unit = func()
       })
     } else {
+      logDebug("running as SPARK_UNKNOWN_USER")
       func()
     }
   }

core/src/main/scala/org/apache/spark/executor/CoarseGrainedExecutorBackend.scala

@@ -22,8 +22,9 @@ import java.nio.ByteBuffer
 import akka.actor._
 import akka.remote._
 
-import org.apache.spark.{SecurityManager, SparkConf, Logging}
+import org.apache.spark.{Logging, SecurityManager, SparkConf}
 import org.apache.spark.TaskState.TaskState
+import org.apache.spark.deploy.SparkHadoopUtil
 import org.apache.spark.deploy.worker.WorkerWatcher
 import org.apache.spark.scheduler.cluster.CoarseGrainedClusterMessages._
 import org.apache.spark.util.{AkkaUtils, Utils}
@@ -94,25 +95,30 @@ private[spark] class CoarseGrainedExecutorBackend(
 
 private[spark] object CoarseGrainedExecutorBackend {
   def run(driverUrl: String, executorId: String, hostname: String, cores: Int,
-          workerUrl: Option[String]) {
-    // Debug code
-    Utils.checkHost(hostname)
-
-    val conf = new SparkConf
-    // Create a new ActorSystem to run the backend, because we can't create a SparkEnv / Executor
-    // before getting started with all our system properties, etc
-    val (actorSystem, boundPort) = AkkaUtils.createActorSystem("sparkExecutor", hostname, 0,
-      indestructible = true, conf = conf, new SecurityManager(conf))
-    // set it
-    val sparkHostPort = hostname + ":" + boundPort
-    actorSystem.actorOf(
-      Props(classOf[CoarseGrainedExecutorBackend], driverUrl, executorId,
-        sparkHostPort, cores),
-      name = "Executor")
-    workerUrl.foreach{ url =>
-      actorSystem.actorOf(Props(classOf[WorkerWatcher], url), name = "WorkerWatcher")
+    workerUrl: Option[String]) {
+
+    SparkHadoopUtil.get.runAsSparkUser { () =>
+        // Debug code
+        Utils.checkHost(hostname)
+
+        val conf = new SparkConf
+        // Create a new ActorSystem to run the backend, because we can't create a
+        // SparkEnv / Executor before getting started with all our system properties, etc
+        val (actorSystem, boundPort) = AkkaUtils.createActorSystem("sparkExecutor", hostname, 0,
+          indestructible = true, conf = conf, new SecurityManager(conf))
+        // set it
+        val sparkHostPort = hostname + ":" + boundPort
+        actorSystem.actorOf(
+          Props(classOf[CoarseGrainedExecutorBackend], driverUrl, executorId,
+            sparkHostPort, cores),
+          name = "Executor")
+        workerUrl.foreach {
+          url =>
+            actorSystem.actorOf(Props(classOf[WorkerWatcher], url), name = "WorkerWatcher")
+        }
+        actorSystem.awaitTermination()
+
     }
-    actorSystem.awaitTermination()
   }
 
   def main(args: Array[String]) {

core/src/main/scala/org/apache/spark/executor/Executor.scala

@@ -128,8 +128,6 @@ private[spark] class Executor(
   // Maintains the list of running tasks.
   private val runningTasks = new ConcurrentHashMap[Long, TaskRunner]
 
-  val sparkUser = Option(System.getenv("SPARK_USER")).getOrElse(SparkContext.SPARK_UNKNOWN_USER)
-
   def launchTask(context: ExecutorBackend, taskId: Long, serializedTask: ByteBuffer) {
     val tr = new TaskRunner(context, taskId, serializedTask)
     runningTasks.put(taskId, tr)
@@ -172,7 +170,7 @@ private[spark] class Executor(
       }
     }
 
-    override def run(): Unit = SparkHadoopUtil.get.runAsUser(sparkUser) { () =>
+    override def run() {
       val startTime = System.currentTimeMillis()
       SparkEnv.set(env)
       Thread.currentThread.setContextClassLoader(replClassLoader)

core/src/main/scala/org/apache/spark/executor/MesosExecutorBackend.scala

@@ -23,10 +23,10 @@ import com.google.protobuf.ByteString
 import org.apache.mesos.{Executor => MesosExecutor, ExecutorDriver, MesosExecutorDriver, MesosNativeLibrary}
 import org.apache.mesos.Protos.{TaskStatus => MesosTaskStatus, _}
 
-import org.apache.spark.Logging
-import org.apache.spark.TaskState
+import org.apache.spark.{Logging, TaskState}
 import org.apache.spark.TaskState.TaskState
 import org.apache.spark.util.Utils
+import org.apache.spark.deploy.SparkHadoopUtil
 
 private[spark] class MesosExecutorBackend
   extends MesosExecutor
@@ -95,9 +95,11 @@ private[spark] class MesosExecutorBackend
  */
 private[spark] object MesosExecutorBackend {
   def main(args: Array[String]) {
-    MesosNativeLibrary.load()
-    // Create a new Executor and start it running
-    val runner = new MesosExecutorBackend()
-    new MesosExecutorDriver(runner).run()
+    SparkHadoopUtil.get.runAsSparkUser { () =>
+        MesosNativeLibrary.load()
+        // Create a new Executor and start it running
+        val runner = new MesosExecutorBackend()
+        new MesosExecutorDriver(runner).run()
+    }
   }
 }

yarn/alpha/src/main/scala/org/apache/spark/deploy/yarn/ApplicationMaster.scala

@@ -70,9 +70,6 @@ class ApplicationMaster(args: ApplicationMasterArguments, conf: Configuration,
 
   private var registered = false
 
-  private val sparkUser = Option(System.getenv("SPARK_USER")).getOrElse(
-    SparkContext.SPARK_UNKNOWN_USER)
-
   def run() {
     // Setup the directories so things go to yarn approved directories rather
     // then user specified and /tmp.
@@ -192,7 +189,8 @@ class ApplicationMaster(args: ApplicationMasterArguments, conf: Configuration,
       false /* initialize */ ,
       Thread.currentThread.getContextClassLoader).getMethod("main", classOf[Array[String]])
     val t = new Thread {
-      override def run(): Unit = SparkHadoopUtil.get.runAsUser(sparkUser) { () =>
+      override def run() {
+
         var successed = false
         try {
           // Copy
@@ -480,6 +478,8 @@ object ApplicationMaster {
 
   def main(argStrings: Array[String]) {
     val args = new ApplicationMasterArguments(argStrings)
-    new ApplicationMaster(args).run()
+    SparkHadoopUtil.get.runAsSparkUser { () =>
+      new ApplicationMaster(args).run()
+    }
   }
 }

yarn/alpha/src/main/scala/org/apache/spark/deploy/yarn/ExecutorLauncher.scala

@@ -29,10 +29,11 @@ import org.apache.hadoop.yarn.util.{ConverterUtils, Records}
 import akka.actor._
 import akka.remote._
 import akka.actor.Terminated
-import org.apache.spark.{Logging, SecurityManager, SparkConf, SparkContext}
+import org.apache.spark.{Logging, SecurityManager, SparkConf}
 import org.apache.spark.util.{Utils, AkkaUtils}
 import org.apache.spark.scheduler.cluster.CoarseGrainedSchedulerBackend
 import org.apache.spark.scheduler.SplitInfo
+import org.apache.spark.deploy.SparkHadoopUtil
 
 /**
  * An application master that allocates executors on behalf of a driver that is running outside
@@ -279,6 +280,8 @@ class ExecutorLauncher(args: ApplicationMasterArguments, conf: Configuration, sp
 object ExecutorLauncher {
   def main(argStrings: Array[String]) {
     val args = new ApplicationMasterArguments(argStrings)
-    new ExecutorLauncher(args).run()
+    SparkHadoopUtil.get.runAsSparkUser { () =>
+      new ExecutorLauncher(args).run()
+    }
   }
 }

yarn/stable/src/main/scala/org/apache/spark/deploy/yarn/ApplicationMaster.scala

@@ -71,9 +71,6 @@ class ApplicationMaster(args: ApplicationMasterArguments, conf: Configuration,
     sparkConf.getInt("spark.yarn.max.worker.failures", math.max(args.numExecutors * 2, 3)))
 
   private var registered = false
-  
-  private val sparkUser = Option(System.getenv("SPARK_USER")).getOrElse(
-    SparkContext.SPARK_UNKNOWN_USER)
 
   def run() {
     // Setup the directories so things go to YARN approved directories rather
@@ -179,8 +176,9 @@ class ApplicationMaster(args: ApplicationMasterArguments, conf: Configuration,
       false /* initialize */ ,
       Thread.currentThread.getContextClassLoader).getMethod("main", classOf[Array[String]])
     val t = new Thread {
-      override def run(): Unit = SparkHadoopUtil.get.runAsUser(sparkUser) { () =>
-        var successed = false
+      override def run() {
+
+      var successed = false
         try {
           // Copy
           var mainArgs: Array[String] = new Array[String](args.userArgs.size)
@@ -462,6 +460,8 @@ object ApplicationMaster {
 
   def main(argStrings: Array[String]) {
     val args = new ApplicationMasterArguments(argStrings)
-    new ApplicationMaster(args).run()
+    SparkHadoopUtil.get.runAsSparkUser { () =>
+      new ApplicationMaster(args).run()
+    }
   }
 }

yarn/stable/src/main/scala/org/apache/spark/deploy/yarn/ExecutorLauncher.scala

@@ -28,12 +28,13 @@ import org.apache.hadoop.yarn.util.{ConverterUtils, Records}
 import akka.actor._
 import akka.remote._
 import akka.actor.Terminated
-import org.apache.spark.{Logging, SecurityManager, SparkConf, SparkContext}
+import org.apache.spark.{Logging, SecurityManager, SparkConf}
 import org.apache.spark.util.{Utils, AkkaUtils}
 import org.apache.spark.scheduler.cluster.CoarseGrainedSchedulerBackend
 import org.apache.spark.scheduler.SplitInfo
 import org.apache.hadoop.yarn.client.api.AMRMClient
 import org.apache.hadoop.yarn.client.api.AMRMClient.ContainerRequest
+import org.apache.spark.deploy.SparkHadoopUtil
 
 /**
  * An application master that allocates executors on behalf of a driver that is running outside
@@ -255,6 +256,8 @@ class ExecutorLauncher(args: ApplicationMasterArguments, conf: Configuration, sp
 object ExecutorLauncher {
   def main(argStrings: Array[String]) {
     val args = new ApplicationMasterArguments(argStrings)
-    new ExecutorLauncher(args).run()
+    SparkHadoopUtil.get.runAsSparkUser { () =>
+      new ExecutorLauncher(args).run()
+    }
   }
 }