Merge remote-tracking branch 'upstream/master'

Author: mubarak

README.md

@@ -4,8 +4,8 @@ Spark is a fast and general cluster computing system for Big Data. It provides
 high-level APIs in Scala, Java, and Python, and an optimized engine that
 supports general computation graphs for data analysis. It also supports a
 rich set of higher-level tools including Spark SQL for SQL and structured
-data processing, MLLib for machine learning, GraphX for graph processing,
-and Spark Streaming.
+data processing, MLlib for machine learning, GraphX for graph processing,
+and Spark Streaming for stream processing.
 
 <http://spark.apache.org/>
 

bin/pyspark

@@ -85,6 +85,8 @@ export PYSPARK_SUBMIT_ARGS
 
 # For pyspark tests
 if [[ -n "$SPARK_TESTING" ]]; then
+  unset YARN_CONF_DIR
+  unset HADOOP_CONF_DIR
   if [[ -n "$PYSPARK_DOC_TEST" ]]; then
     exec "$PYSPARK_PYTHON" -m doctest $1
   else

core/src/main/scala/org/apache/spark/rdd/RDD.scala

@@ -1064,11 +1064,10 @@ abstract class RDD[T: ClassTag](
       // greater than totalParts because we actually cap it at totalParts in runJob.
       var numPartsToTry = 1
       if (partsScanned > 0) {
-        // If we didn't find any rows after the first iteration, just try all partitions next.
-        // Otherwise, interpolate the number of partitions we need to try, but overestimate it
-        // by 50%.
+        // If we didn't find any rows after the previous iteration, quadruple and retry.  Otherwise,
+        // interpolate the number of partitions we need to try, but overestimate it by 50%.
         if (buf.size == 0) {
-          numPartsToTry = totalParts - 1
+          numPartsToTry = partsScanned * 4
         } else {
           numPartsToTry = (1.5 * num * partsScanned / buf.size).toInt
         }

ec2/spark_ec2.py

@@ -102,9 +102,17 @@ def parse_args():
              "(for debugging)")
     parser.add_option(
         "--ebs-vol-size", metavar="SIZE", type="int", default=0,
-        help="Attach a new EBS volume of size SIZE (in GB) to each node as " +
-             "/vol. The volumes will be deleted when the instances terminate. " +
-             "Only possible on EBS-backed AMIs.")
+        help="Size (in GB) of each EBS volume.")
+    parser.add_option(
+        "--ebs-vol-type", default="standard",
+        help="EBS volume type (e.g. 'gp2', 'standard').")
+    parser.add_option(
+        "--ebs-vol-num", type="int", default=1,
+        help="Number of EBS volumes to attach to each node as /vol[x]. " +
+             "The volumes will be deleted when the instances terminate. " +
+             "Only possible on EBS-backed AMIs. " +
+             "EBS volumes are only attached if --ebs-vol-size > 0." +
+             "Only support up to 8 EBS volumes.")
     parser.add_option(
         "--swap", metavar="SWAP", type="int", default=1024,
         help="Swap space to set up per node, in MB (default: 1024)")
@@ -348,13 +356,16 @@ def launch_cluster(conn, opts, cluster_name):
         print >> stderr, "Could not find AMI " + opts.ami
         sys.exit(1)
 
-    # Create block device mapping so that we can add an EBS volume if asked to
+    # Create block device mapping so that we can add EBS volumes if asked to.
+    # The first drive is attached as /dev/sds, 2nd as /dev/sdt, ... /dev/sdz
     block_map = BlockDeviceMapping()
     if opts.ebs_vol_size > 0:
-        device = EBSBlockDeviceType()
-        device.size = opts.ebs_vol_size
-        device.delete_on_termination = True
-        block_map["/dev/sdv"] = device
+        for i in range(opts.ebs_vol_num):
+            device = EBSBlockDeviceType()
+            device.size = opts.ebs_vol_size
+            device.volume_type=opts.ebs_vol_type
+            device.delete_on_termination = True
+            block_map["/dev/sd" + chr(ord('s') + i)] = device
 
     # AWS ignores the AMI-specified block device mapping for M3 (see SPARK-3342).
     if opts.instance_type.startswith('m3.'):
@@ -828,6 +839,12 @@ def get_partition(total, num_partitions, current_partitions):
 
 def real_main():
     (opts, action, cluster_name) = parse_args()
+
+    # Input parameter validation
+    if opts.ebs_vol_num > 8:
+        print >> stderr, "ebs-vol-num cannot be greater than 8"
+        sys.exit(1)
+
     try:
         conn = ec2.connect_to_region(opts.region)
     except Exception as e:

python/pyspark/rdd.py

@@ -1089,11 +1089,11 @@ def take(self, num):
             # we actually cap it at totalParts in runJob.
             numPartsToTry = 1
             if partsScanned > 0:
-                # If we didn't find any rows after the first iteration, just
-                # try all partitions next. Otherwise, interpolate the number
-                # of partitions we need to try, but overestimate it by 50%.
+                # If we didn't find any rows after the previous iteration,
+                # quadruple and retry.  Otherwise, interpolate the number of
+                # partitions we need to try, but overestimate it by 50%.
                 if len(items) == 0:
-                    numPartsToTry = totalParts - 1
+                    numPartsToTry = partsScanned * 4
                 else:
                     numPartsToTry = int(1.5 * num * partsScanned / len(items))
 

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

@@ -35,7 +35,7 @@ import org.apache.hadoop.yarn.conf.YarnConfiguration
 import org.apache.hadoop.yarn.ipc.YarnRPC
 import org.apache.hadoop.yarn.util.{Apps, ConverterUtils, Records, ProtoUtils}
 
-import org.apache.spark.{SparkConf, Logging}
+import org.apache.spark.{SecurityManager, SparkConf, Logging}
 
 
 class ExecutorRunnable(
@@ -46,7 +46,8 @@ class ExecutorRunnable(
     slaveId: String,
     hostname: String,
     executorMemory: Int,
-    executorCores: Int)
+    executorCores: Int,
+    securityMgr: SecurityManager)
   extends Runnable with ExecutorRunnableUtil with Logging {
 
   var rpc: YarnRPC = YarnRPC.create(conf)
@@ -86,6 +87,8 @@ class ExecutorRunnable(
     logInfo("Setting up executor with commands: " + commands)
     ctx.setCommands(commands)
 
+    ctx.setApplicationACLs(YarnSparkHadoopUtil.getApplicationAclsForYarn(securityMgr))
+
     // Send the start request to the ContainerManager
     val startReq = Records.newRecord(classOf[StartContainerRequest])
     .asInstanceOf[StartContainerRequest]

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

@@ -23,7 +23,7 @@ import java.util.concurrent.atomic.AtomicInteger
 import scala.collection.JavaConversions._
 import scala.collection.mutable.{ArrayBuffer, HashMap}
 
-import org.apache.spark.SparkConf
+import org.apache.spark.{SecurityManager, SparkConf}
 import org.apache.spark.scheduler.SplitInfo
 
 import org.apache.hadoop.conf.Configuration
@@ -41,21 +41,23 @@ private[yarn] class YarnAllocationHandler(
     resourceManager: AMRMProtocol,
     appAttemptId: ApplicationAttemptId,
     args: ApplicationMasterArguments,
-    preferredNodes: collection.Map[String, collection.Set[SplitInfo]])
-  extends YarnAllocator(conf, sparkConf, args, preferredNodes) {
+    preferredNodes: collection.Map[String, collection.Set[SplitInfo]],
+    securityMgr: SecurityManager)
+  extends YarnAllocator(conf, sparkConf, args, preferredNodes, securityMgr) {
 
   private val lastResponseId = new AtomicInteger()
   private val releaseList: CopyOnWriteArrayList[ContainerId] = new CopyOnWriteArrayList()
 
   override protected def allocateContainers(count: Int): YarnAllocateResponse = {
     var resourceRequests: List[ResourceRequest] = null
 
-    // default.
-    if (count <= 0 || preferredHostToCount.isEmpty) {
-      logDebug("numExecutors: " + count + ", host preferences: " +
-        preferredHostToCount.isEmpty)
-      resourceRequests = List(createResourceRequest(
-        AllocationType.ANY, null, count, YarnSparkHadoopUtil.RM_REQUEST_PRIORITY))
+    logDebug("numExecutors: " + count)
+    if (count <= 0) {
+      resourceRequests = List()
+    } else if (preferredHostToCount.isEmpty) {
+        logDebug("host preferences is empty")
+        resourceRequests = List(createResourceRequest(
+          AllocationType.ANY, null, count, YarnSparkHadoopUtil.RM_REQUEST_PRIORITY))
     } else {
       // request for all hosts in preferred nodes and for numExecutors -
       // candidates.size, request by default allocation policy.

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

@@ -27,7 +27,7 @@ import org.apache.hadoop.yarn.conf.YarnConfiguration
 import org.apache.hadoop.yarn.ipc.YarnRPC
 import org.apache.hadoop.yarn.util.{ConverterUtils, Records}
 
-import org.apache.spark.{Logging, SparkConf}
+import org.apache.spark.{Logging, SecurityManager, SparkConf}
 import org.apache.spark.scheduler.SplitInfo
 import org.apache.spark.util.Utils
 
@@ -45,15 +45,16 @@ private class YarnRMClientImpl(args: ApplicationMasterArguments) extends YarnRMC
       sparkConf: SparkConf,
       preferredNodeLocations: Map[String, Set[SplitInfo]],
       uiAddress: String,
-      uiHistoryAddress: String) = {
+      uiHistoryAddress: String,
+      securityMgr: SecurityManager) = {
     this.rpc = YarnRPC.create(conf)
     this.uiHistoryAddress = uiHistoryAddress
 
     resourceManager = registerWithResourceManager(conf)
     registerApplicationMaster(uiAddress)
 
     new YarnAllocationHandler(conf, sparkConf, resourceManager, getAttemptId(), args,
-      preferredNodeLocations)
+      preferredNodeLocations, securityMgr)
   }
 
   override def getAttemptId() = {

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

@@ -116,7 +116,7 @@ private[spark] class ApplicationMaster(args: ApplicationMasterArguments,
     val securityMgr = new SecurityManager(sparkConf)
 
     if (isDriver) {
-      runDriver()
+      runDriver(securityMgr)
     } else {
       runExecutorLauncher(securityMgr)
     }
@@ -157,7 +157,7 @@ private[spark] class ApplicationMaster(args: ApplicationMasterArguments,
     sparkContextRef.compareAndSet(sc, null)
   }
 
-  private def registerAM(uiAddress: String) = {
+  private def registerAM(uiAddress: String, securityMgr: SecurityManager) = {
     val sc = sparkContextRef.get()
 
     val appId = client.getAttemptId().getApplicationId().toString()
@@ -170,13 +170,14 @@ private[spark] class ApplicationMaster(args: ApplicationMasterArguments,
       if (sc != null) sc.getConf else sparkConf,
       if (sc != null) sc.preferredNodeLocationData else Map(),
       uiAddress,
-      historyAddress)
+      historyAddress,
+      securityMgr)
 
     allocator.allocateResources()
     reporterThread = launchReporterThread()
   }
 
-  private def runDriver(): Unit = {
+  private def runDriver(securityMgr: SecurityManager): Unit = {
     addAmIpFilter()
     val userThread = startUserClass()
 
@@ -188,7 +189,7 @@ private[spark] class ApplicationMaster(args: ApplicationMasterArguments,
     if (sc == null) {
       finish(FinalApplicationStatus.FAILED, "Timed out waiting for SparkContext.")
     } else {
-      registerAM(sc.ui.appUIHostPort)
+      registerAM(sc.ui.appUIHostPort, securityMgr)
       try {
         userThread.join()
       } finally {
@@ -203,7 +204,7 @@ private[spark] class ApplicationMaster(args: ApplicationMasterArguments,
       conf = sparkConf, securityManager = securityMgr)._1
     actor = waitForSparkDriver()
     addAmIpFilter()
-    registerAM(sparkConf.get("spark.driver.appUIAddress", ""))
+    registerAM(sparkConf.get("spark.driver.appUIAddress", ""), securityMgr)
 
     // In client mode the actor will stop the reporter thread.
     reporterThread.join()

yarn/common/src/main/scala/org/apache/spark/deploy/yarn/ClientBase.scala

@@ -430,10 +430,8 @@ trait ClientBase extends Logging {
 
     // send the acl settings into YARN to control who has access via YARN interfaces
     val securityManager = new SecurityManager(sparkConf)
-    val acls = Map[ApplicationAccessType, String] (
-      ApplicationAccessType.VIEW_APP -> securityManager.getViewAcls,
-      ApplicationAccessType.MODIFY_APP -> securityManager.getModifyAcls)
-    amContainer.setApplicationACLs(acls)
+    amContainer.setApplicationACLs(YarnSparkHadoopUtil.getApplicationAclsForYarn(securityManager))
+
     amContainer
   }
 }