SPARK-1305: Support persisting RDD's directly to Tachyon

Move the PR#468 of apache-incubator-spark to the apache-spark
"Adding an option to persist Spark RDD blocks into Tachyon."

Author: Haoyuan Li <[email protected]>
Author: RongGu <[email protected]>

Closes #158 from RongGu/master and squashes the following commits:

72b7768 [Haoyuan Li] merge master
9f7fa1b [Haoyuan Li] fix code style
ae7834b [Haoyuan Li] minor cleanup
a8b3ec6 [Haoyuan Li] merge master branch
e0f4891 [Haoyuan Li] better check offheap.
55b5918 [RongGu] address matei's comment on the replication of offHeap storagelevel
7cd4600 [RongGu] remove some logic code for tachyonstore's replication
51149e7 [RongGu] address aaron's comment on returning value of the remove() function in tachyonstore
8adfcfa [RongGu] address arron's comment on inTachyonSize
120e48a [RongGu] changed the root-level dir name in Tachyon
5cc041c [Haoyuan Li] address aaron's comments
9b97935 [Haoyuan Li] address aaron's comments
d9a6438 [Haoyuan Li] fix for pspark
77d2703 [Haoyuan Li] change python api.git status
3dcace4 [Haoyuan Li] address matei's comments
91fa09d [Haoyuan Li] address patrick's comments
589eafe [Haoyuan Li] use TRY_CACHE instead of MUST_CACHE
64348b2 [Haoyuan Li] update conf docs.
ed73e19 [Haoyuan Li] Merge branch 'master' of github.com:RongGu/spark-1
619a9a8 [RongGu] set number of directories in TachyonStore back to 64; added a TODO tag for duplicated code from the DiskStore
be79d77 [RongGu] find a way to clean up some unnecessay metods and classed to make the code simpler
49cc724 [Haoyuan Li] update docs with off_headp option
4572f9f [RongGu] reserving the old apply function API of StorageLevel
04301d3 [RongGu] rename StorageLevel.TACHYON to Storage.OFF_HEAP
c9aeabf [RongGu] rename the StorgeLevel.TACHYON as StorageLevel.OFF_HEAP
76805aa [RongGu] unifies the config properties name prefix; add the configs into docs/configuration.md
e700d9c [RongGu] add the SparkTachyonHdfsLR example and some comments
fd84156 [RongGu] use randomUUID to generate sparkapp directory name on tachyon;minor code style fix
939e467 [Haoyuan Li] 0.4.1-thrift from maven central
86a2eab [Haoyuan Li] tachyon 0.4.1-thrift is in the staging repo. but jenkins failed to download it. temporarily revert it back to 0.4.1
16c5798 [RongGu] make the dependency on tachyon as tachyon-0.4.1-thrift
eacb2e8 [RongGu] Merge branch 'master' of https://github.com/RongGu/spark-1
bbeb4de [RongGu] fix the JsonProtocolSuite test failure problem
6adb58f [RongGu] Merge branch 'master' of https://github.com/RongGu/spark-1
d827250 [RongGu] fix JsonProtocolSuie test failure
716e93b [Haoyuan Li] revert the version
ca14469 [Haoyuan Li] bump tachyon version to 0.4.1-thrift
2825a13 [RongGu] up-merging to the current master branch of the apache spark
6a22c1a [Haoyuan Li] fix scalastyle
8968b67 [Haoyuan Li] exclude more libraries from tachyon dependency to be the same as referencing tachyon-client.
77be7e8 [RongGu] address mateiz's comment about the temp folder name problem. The implementation followed mateiz's advice.
1dcadf9 [Haoyuan Li] typo
bf278fa [Haoyuan Li] fix python tests
e82909c [Haoyuan Li] minor cleanup
776a56c [Haoyuan Li] address patrick's and ali's comments from the previous PR
8859371 [Haoyuan Li] various minor fixes and clean up
e3ddbba [Haoyuan Li] add doc to use Tachyon cache mode.
fcaeab2 [Haoyuan Li] address Aaron's comment
e554b1e [Haoyuan Li] add python code
47304b3 [Haoyuan Li] make tachyonStore in BlockMananger lazy val; add more comments StorageLevels.
dc8ef24 [Haoyuan Li] add old storelevel constructor
e01a271 [Haoyuan Li] update tachyon 0.4.1
8011a96 [RongGu] fix a brought-in mistake in StorageLevel
70ca182 [RongGu] a bit change in comment
556978b [RongGu] fix the scalastyle errors
791189b [RongGu] "Adding an option to persist Spark RDD blocks into Tachyon." move the PR#468 of apache-incubator-spark to the apache-spark

Author: haoyuan

core/pom.xml

@@ -200,6 +200,53 @@
       <artifactId>derby</artifactId>
       <scope>test</scope>
     </dependency>
+    <dependency>
+      <groupId>org.tachyonproject</groupId>
+      <artifactId>tachyon</artifactId>
+      <version>0.4.1-thrift</version>
+      <exclusions>
+        <exclusion>
+          <groupId>org.apache.hadoop</groupId>
+          <artifactId>hadoop-client</artifactId>
+        </exclusion>
+        <exclusion>
+          <groupId>org.apache.curator</groupId>
+          <artifactId>curator-recipes</artifactId>
+        </exclusion>
+        <exclusion>
+          <groupId>org.eclipse.jetty</groupId>
+          <artifactId>jetty-jsp</artifactId>
+        </exclusion>
+        <exclusion>
+          <groupId>org.eclipse.jetty</groupId>
+          <artifactId>jetty-webapp</artifactId>
+        </exclusion>
+        <exclusion>
+          <groupId>org.eclipse.jetty</groupId>
+          <artifactId>jetty-server</artifactId>
+        </exclusion>
+        <exclusion>
+          <groupId>org.eclipse.jetty</groupId>
+          <artifactId>jetty-servlet</artifactId>
+        </exclusion>
+        <exclusion>
+          <groupId>junit</groupId>
+          <artifactId>junit</artifactId>
+        </exclusion>
+        <exclusion>
+          <groupId>org.powermock</groupId>
+          <artifactId>powermock-module-junit4</artifactId>
+        </exclusion>
+        <exclusion>
+          <groupId>org.powermock</groupId>
+          <artifactId>powermock-api-mockito</artifactId>
+        </exclusion>
+        <exclusion>
+          <groupId>org.apache.curator</groupId>
+          <artifactId>curator-test</artifactId>
+        </exclusion>
+      </exclusions>
+    </dependency>
     <dependency>
       <groupId>org.scalatest</groupId>
       <artifactId>scalatest_${scala.binary.version}</artifactId>

core/src/main/java/org/apache/spark/api/java/StorageLevels.java

@@ -23,17 +23,18 @@
  * Expose some commonly useful storage level constants.
  */
 public class StorageLevels {
-  public static final StorageLevel NONE = create(false, false, false, 1);
-  public static final StorageLevel DISK_ONLY = create(true, false, false, 1);
-  public static final StorageLevel DISK_ONLY_2 = create(true, false, false, 2);
-  public static final StorageLevel MEMORY_ONLY = create(false, true, true, 1);
-  public static final StorageLevel MEMORY_ONLY_2 = create(false, true, true, 2);
-  public static final StorageLevel MEMORY_ONLY_SER = create(false, true, false, 1);
-  public static final StorageLevel MEMORY_ONLY_SER_2 = create(false, true, false, 2);
-  public static final StorageLevel MEMORY_AND_DISK = create(true, true, true, 1);
-  public static final StorageLevel MEMORY_AND_DISK_2 = create(true, true, true, 2);
-  public static final StorageLevel MEMORY_AND_DISK_SER = create(true, true, false, 1);
-  public static final StorageLevel MEMORY_AND_DISK_SER_2 = create(true, true, false, 2);
+  public static final StorageLevel NONE = create(false, false, false, false, 1);
+  public static final StorageLevel DISK_ONLY = create(true, false, false, false, 1);
+  public static final StorageLevel DISK_ONLY_2 = create(true, false, false, false, 2);
+  public static final StorageLevel MEMORY_ONLY = create(false, true, false, true, 1);
+  public static final StorageLevel MEMORY_ONLY_2 = create(false, true, false, true, 2);
+  public static final StorageLevel MEMORY_ONLY_SER = create(false, true, false, false, 1);
+  public static final StorageLevel MEMORY_ONLY_SER_2 = create(false, true, false, false, 2);
+  public static final StorageLevel MEMORY_AND_DISK = create(true, true, false, true, 1);
+  public static final StorageLevel MEMORY_AND_DISK_2 = create(true, true, false, true, 2);
+  public static final StorageLevel MEMORY_AND_DISK_SER = create(true, true, false, false, 1);
+  public static final StorageLevel MEMORY_AND_DISK_SER_2 = create(true, true, false, false, 2);
+  public static final StorageLevel OFF_HEAP = create(false, false, true, false, 1);
 
   /**
    * Create a new StorageLevel object.
@@ -42,7 +43,26 @@ public class StorageLevels {
    * @param deserialized saved as deserialized objects, if true
    * @param replication replication factor
    */
-  public static StorageLevel create(boolean useDisk, boolean useMemory, boolean deserialized, int replication) {
-    return StorageLevel.apply(useDisk, useMemory, deserialized, replication);
+  @Deprecated
+  public static StorageLevel create(boolean useDisk, boolean useMemory, boolean deserialized,
+      int replication) {
+    return StorageLevel.apply(useDisk, useMemory, false, deserialized, replication);
+  }
+
+  /**
+   * Create a new StorageLevel object.
+   * @param useDisk saved to disk, if true
+   * @param useMemory saved to memory, if true
+   * @param useOffHeap saved to Tachyon, if true
+   * @param deserialized saved as deserialized objects, if true
+   * @param replication replication factor
+   */
+  public static StorageLevel create(
+    boolean useDisk,
+    boolean useMemory,
+    boolean useOffHeap,
+    boolean deserialized,
+    int replication) {
+    return StorageLevel.apply(useDisk, useMemory, useOffHeap, deserialized, replication);
   }
 }

core/src/main/scala/org/apache/spark/SparkContext.scala

@@ -19,14 +19,13 @@ package org.apache.spark
 
 import java.io._
 import java.net.URI
-import java.util.{Properties, UUID}
 import java.util.concurrent.atomic.AtomicInteger
-
+import java.util.{Properties, UUID}
+import java.util.UUID.randomUUID
 import scala.collection.{Map, Set}
 import scala.collection.generic.Growable
 import scala.collection.mutable.{ArrayBuffer, HashMap}
 import scala.reflect.{ClassTag, classTag}
-
 import org.apache.hadoop.conf.Configuration
 import org.apache.hadoop.fs.Path
 import org.apache.hadoop.io.{ArrayWritable, BooleanWritable, BytesWritable, DoubleWritable, FloatWritable, IntWritable, LongWritable, NullWritable, Text, Writable}
@@ -130,6 +129,11 @@ class SparkContext(
   val master = conf.get("spark.master")
   val appName = conf.get("spark.app.name")
 
+  // Generate the random name for a temp folder in Tachyon
+  // Add a timestamp as the suffix here to make it more safe
+  val tachyonFolderName = "spark-" + randomUUID.toString()
+  conf.set("spark.tachyonStore.folderName", tachyonFolderName)
+
   val isLocal = (master == "local" || master.startsWith("local["))
 
   if (master == "yarn-client") System.setProperty("SPARK_YARN_MODE", "true")

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

@@ -53,7 +53,8 @@ private[spark] class CoarseGrainedExecutorBackend(
     case RegisteredExecutor(sparkProperties) =>
       logInfo("Successfully registered with driver")
       // Make this host instead of hostPort ?
-      executor = new Executor(executorId, Utils.parseHostPort(hostPort)._1, sparkProperties)
+      executor = new Executor(executorId, Utils.parseHostPort(hostPort)._1, sparkProperties, 
+        false)
 
     case RegisterExecutorFailed(message) =>
       logError("Slave registration failed: " + message)
@@ -105,7 +106,8 @@ private[spark] object CoarseGrainedExecutorBackend {
     // set it
     val sparkHostPort = hostname + ":" + boundPort
     actorSystem.actorOf(
-      Props(classOf[CoarseGrainedExecutorBackend], driverUrl, executorId, sparkHostPort, cores),
+      Props(classOf[CoarseGrainedExecutorBackend], driverUrl, executorId,
+        sparkHostPort, cores),
       name = "Executor")
     workerUrl.foreach{ url =>
       actorSystem.actorOf(Props(classOf[WorkerWatcher], url), name = "WorkerWatcher")

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

@@ -41,13 +41,22 @@ object ExecutorExitCode {
   /** DiskStore failed to create a local temporary directory after many attempts. */
   val DISK_STORE_FAILED_TO_CREATE_DIR = 53
 
+  /** TachyonStore failed to initialize after many attempts. */
+  val TACHYON_STORE_FAILED_TO_INITIALIZE = 54
+  
+  /** TachyonStore failed to create a local temporary directory after many attempts. */
+  val TACHYON_STORE_FAILED_TO_CREATE_DIR = 55
+  
   def explainExitCode(exitCode: Int): String = {
     exitCode match {
       case UNCAUGHT_EXCEPTION => "Uncaught exception"
       case UNCAUGHT_EXCEPTION_TWICE => "Uncaught exception, and logging the exception failed"
       case OOM => "OutOfMemoryError"
       case DISK_STORE_FAILED_TO_CREATE_DIR =>
         "Failed to create local directory (bad spark.local.dir?)"
+      case TACHYON_STORE_FAILED_TO_INITIALIZE => "TachyonStore failed to initialize."
+      case TACHYON_STORE_FAILED_TO_CREATE_DIR =>
+        "TachyonStore failed to create a local temporary directory."
       case _ => 
         "Unknown executor exit code (" + exitCode + ")" + (
           if (exitCode > 128) {

core/src/main/scala/org/apache/spark/storage/BlockManager.scala

@@ -19,22 +19,20 @@ package org.apache.spark.storage
 
 import java.io.{File, InputStream, OutputStream}
 import java.nio.{ByteBuffer, MappedByteBuffer}
-
 import scala.collection.mutable.{ArrayBuffer, HashMap}
 import scala.concurrent.{Await, Future}
 import scala.concurrent.duration._
 import scala.util.Random
-
 import akka.actor.{ActorSystem, Cancellable, Props}
 import it.unimi.dsi.fastutil.io.{FastBufferedOutputStream, FastByteArrayOutputStream}
 import sun.nio.ch.DirectBuffer
-
 import org.apache.spark.{Logging, SecurityManager, SparkConf, SparkEnv, SparkException}
 import org.apache.spark.io.CompressionCodec
 import org.apache.spark.network._
 import org.apache.spark.serializer.Serializer
 import org.apache.spark.util._
 
+
 sealed trait Values
 
 case class ByteBufferValues(buffer: ByteBuffer) extends Values
@@ -59,6 +57,17 @@ private[spark] class BlockManager(
 
   private[storage] val memoryStore: BlockStore = new MemoryStore(this, maxMemory)
   private[storage] val diskStore = new DiskStore(this, diskBlockManager)
+  var tachyonInitialized = false
+  private[storage] lazy val tachyonStore: TachyonStore = {
+    val storeDir = conf.get("spark.tachyonStore.baseDir", "/tmp_spark_tachyon")
+    val appFolderName = conf.get("spark.tachyonStore.folderName")
+    val tachyonStorePath = s"${storeDir}/${appFolderName}/${this.executorId}"
+    val tachyonMaster = conf.get("spark.tachyonStore.url",  "tachyon://localhost:19998")
+    val tachyonBlockManager = new TachyonBlockManager(
+      shuffleBlockManager, tachyonStorePath, tachyonMaster)
+    tachyonInitialized = true
+    new TachyonStore(this, tachyonBlockManager)
+  }
 
   // If we use Netty for shuffle, start a new Netty-based shuffle sender service.
   private val nettyPort: Int = {
@@ -248,8 +257,10 @@ private[spark] class BlockManager(
     if (info.tellMaster) {
       val storageLevel = status.storageLevel
       val inMemSize = Math.max(status.memSize, droppedMemorySize)
+      val inTachyonSize = status.tachyonSize
       val onDiskSize = status.diskSize
-      master.updateBlockInfo(blockManagerId, blockId, storageLevel, inMemSize, onDiskSize)
+      master.updateBlockInfo(
+        blockManagerId, blockId, storageLevel, inMemSize, onDiskSize, inTachyonSize)
     } else true
   }
 
@@ -259,22 +270,24 @@ private[spark] class BlockManager(
    * and the updated in-memory and on-disk sizes.
    */
   private def getCurrentBlockStatus(blockId: BlockId, info: BlockInfo): BlockStatus = {
-    val (newLevel, inMemSize, onDiskSize) = info.synchronized {
+    val (newLevel, inMemSize, onDiskSize, inTachyonSize) = info.synchronized {
       info.level match {
         case null =>
-          (StorageLevel.NONE, 0L, 0L)
+          (StorageLevel.NONE, 0L, 0L, 0L)
         case level =>
           val inMem = level.useMemory && memoryStore.contains(blockId)
+          val inTachyon = level.useOffHeap && tachyonStore.contains(blockId)
           val onDisk = level.useDisk && diskStore.contains(blockId)
           val deserialized = if (inMem) level.deserialized else false
-          val replication = if (inMem || onDisk) level.replication else 1
-          val storageLevel = StorageLevel(onDisk, inMem, deserialized, replication)
+          val replication = if (inMem || inTachyon || onDisk) level.replication else 1
+          val storageLevel = StorageLevel(onDisk, inMem, inTachyon, deserialized, replication)
           val memSize = if (inMem) memoryStore.getSize(blockId) else 0L
+          val tachyonSize = if (inTachyon) tachyonStore.getSize(blockId) else 0L
           val diskSize = if (onDisk) diskStore.getSize(blockId) else 0L
-          (storageLevel, memSize, diskSize)
+          (storageLevel, memSize, diskSize, tachyonSize)
       }
     }
-    BlockStatus(newLevel, inMemSize, onDiskSize)
+    BlockStatus(newLevel, inMemSize, onDiskSize, inTachyonSize)
   }
 
   /**
@@ -354,6 +367,24 @@ private[spark] class BlockManager(
               logDebug("Block " + blockId + " not found in memory")
           }
         }
+        
+        // Look for the block in Tachyon
+        if (level.useOffHeap) {
+          logDebug("Getting block " + blockId + " from tachyon")
+          if (tachyonStore.contains(blockId)) {
+            tachyonStore.getBytes(blockId) match {
+              case Some(bytes) => {
+                if (!asValues) {
+                  return Some(bytes)
+                } else {
+                  return Some(dataDeserialize(blockId, bytes))
+                }
+              }
+              case None =>
+                logDebug("Block " + blockId + " not found in tachyon")
+            }
+          }
+        }
 
         // Look for block on disk, potentially storing it back into memory if required:
         if (level.useDisk) {
@@ -620,6 +651,23 @@ private[spark] class BlockManager(
           }
           // Keep track of which blocks are dropped from memory
           res.droppedBlocks.foreach { block => updatedBlocks += block }
+        } else if (level.useOffHeap) {
+          // Save to Tachyon.
+          val res = data match {
+            case IteratorValues(iterator) =>
+              tachyonStore.putValues(blockId, iterator, level, false)
+            case ArrayBufferValues(array) =>
+              tachyonStore.putValues(blockId, array, level, false)
+            case ByteBufferValues(bytes) => {
+              bytes.rewind();
+              tachyonStore.putBytes(blockId, bytes, level)
+            }
+          }
+          size = res.size
+          res.data match {
+            case Right(newBytes) => bytesAfterPut = newBytes
+            case _ =>
+          }
         } else {
           // Save directly to disk.
           // Don't get back the bytes unless we replicate them.
@@ -644,8 +692,8 @@ private[spark] class BlockManager(
 
         val putBlockStatus = getCurrentBlockStatus(blockId, putBlockInfo)
         if (putBlockStatus.storageLevel != StorageLevel.NONE) {
-          // Now that the block is in either the memory or disk store, let other threads read it,
-          // and tell the master about it.
+          // Now that the block is in either the memory, tachyon, or disk store,
+          // let other threads read it, and tell the master about it.
           marked = true
           putBlockInfo.markReady(size)
           if (tellMaster) {
@@ -707,7 +755,8 @@ private[spark] class BlockManager(
    */
   var cachedPeers: Seq[BlockManagerId] = null
   private def replicate(blockId: BlockId, data: ByteBuffer, level: StorageLevel) {
-    val tLevel = StorageLevel(level.useDisk, level.useMemory, level.deserialized, 1)
+    val tLevel = StorageLevel(
+      level.useDisk, level.useMemory, level.useOffHeap, level.deserialized, 1)
     if (cachedPeers == null) {
       cachedPeers = master.getPeers(blockManagerId, level.replication - 1)
     }
@@ -832,9 +881,10 @@ private[spark] class BlockManager(
       // Removals are idempotent in disk store and memory store. At worst, we get a warning.
       val removedFromMemory = memoryStore.remove(blockId)
       val removedFromDisk = diskStore.remove(blockId)
-      if (!removedFromMemory && !removedFromDisk) {
+      val removedFromTachyon = if (tachyonInitialized) tachyonStore.remove(blockId) else false
+      if (!removedFromMemory && !removedFromDisk && !removedFromTachyon) {
         logWarning("Block " + blockId + " could not be removed as it was not found in either " +
-          "the disk or memory store")
+          "the disk, memory, or tachyon store")
       }
       blockInfo.remove(blockId)
       if (tellMaster && info.tellMaster) {
@@ -871,6 +921,9 @@ private[spark] class BlockManager(
           if (level.useDisk) {
             diskStore.remove(id)
           }
+          if (level.useOffHeap) {
+            tachyonStore.remove(id)
+          }
           iterator.remove()
           logInfo("Dropped block " + id)
         }
@@ -946,6 +999,9 @@ private[spark] class BlockManager(
     blockInfo.clear()
     memoryStore.clear()
     diskStore.clear()
+    if (tachyonInitialized) {
+      tachyonStore.clear() 
+    }
     metadataCleaner.cancel()
     broadcastCleaner.cancel()
     logInfo("BlockManager stopped")

core/src/main/scala/org/apache/spark/storage/BlockManagerMaster.scala

@@ -63,9 +63,10 @@ class BlockManagerMaster(var driverActor: ActorRef, conf: SparkConf) extends Log
       blockId: BlockId,
       storageLevel: StorageLevel,
       memSize: Long,
-      diskSize: Long): Boolean = {
+      diskSize: Long,
+      tachyonSize: Long): Boolean = {
     val res = askDriverWithReply[Boolean](
-      UpdateBlockInfo(blockManagerId, blockId, storageLevel, memSize, diskSize))
+      UpdateBlockInfo(blockManagerId, blockId, storageLevel, memSize, diskSize, tachyonSize))
     logInfo("Updated info of block " + blockId)
     res
   }