2011-02-27 17:27:12 -05:00
|
|
|
package spark
|
|
|
|
|
2012-09-12 17:54:40 -04:00
|
|
|
import java.io.{DataInputStream, DataOutputStream, ByteArrayOutputStream, ByteArrayInputStream}
|
2011-02-27 17:27:12 -05:00
|
|
|
import java.util.concurrent.ConcurrentHashMap
|
|
|
|
|
2012-06-07 03:25:47 -04:00
|
|
|
import akka.actor._
|
2012-06-29 02:51:28 -04:00
|
|
|
import akka.dispatch._
|
|
|
|
import akka.pattern.ask
|
|
|
|
import akka.remote._
|
|
|
|
import akka.util.Duration
|
|
|
|
import akka.util.Timeout
|
2012-06-07 03:25:47 -04:00
|
|
|
import akka.util.duration._
|
|
|
|
|
2012-09-12 17:54:40 -04:00
|
|
|
import scala.collection.mutable.HashMap
|
2011-03-06 19:16:38 -05:00
|
|
|
import scala.collection.mutable.HashSet
|
2011-02-27 22:15:52 -05:00
|
|
|
|
2012-06-07 03:25:47 -04:00
|
|
|
import spark.storage.BlockManagerId
|
|
|
|
|
2012-10-02 22:00:19 -04:00
|
|
|
private[spark] sealed trait MapOutputTrackerMessage
|
|
|
|
private[spark] case class GetMapOutputLocations(shuffleId: Int) extends MapOutputTrackerMessage
|
|
|
|
private[spark] case object StopMapOutputTracker extends MapOutputTrackerMessage
|
2011-03-06 19:16:38 -05:00
|
|
|
|
2012-10-02 22:00:19 -04:00
|
|
|
private[spark] class MapOutputTrackerActor(tracker: MapOutputTracker) extends Actor with Logging {
|
2012-06-07 03:25:47 -04:00
|
|
|
def receive = {
|
|
|
|
case GetMapOutputLocations(shuffleId: Int) =>
|
|
|
|
logInfo("Asked to get map output locations for shuffle " + shuffleId)
|
2012-09-12 17:54:40 -04:00
|
|
|
sender ! tracker.getSerializedLocations(shuffleId)
|
2012-06-07 03:25:47 -04:00
|
|
|
|
|
|
|
case StopMapOutputTracker =>
|
|
|
|
logInfo("MapOutputTrackerActor stopped!")
|
2012-06-29 02:51:28 -04:00
|
|
|
sender ! true
|
|
|
|
context.stop(self)
|
2011-02-27 22:15:52 -05:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2012-10-02 22:00:19 -04:00
|
|
|
private[spark] class MapOutputTracker(actorSystem: ActorSystem, isMaster: Boolean) extends Logging {
|
2012-06-07 03:25:47 -04:00
|
|
|
val ip: String = System.getProperty("spark.master.host", "localhost")
|
|
|
|
val port: Int = System.getProperty("spark.master.port", "7077").toInt
|
2012-06-29 02:51:28 -04:00
|
|
|
val actorName: String = "MapOutputTracker"
|
|
|
|
|
|
|
|
val timeout = 10.seconds
|
2011-05-19 14:19:25 -04:00
|
|
|
|
2012-09-12 17:54:40 -04:00
|
|
|
var bmAddresses = new ConcurrentHashMap[Int, Array[BlockManagerId]]
|
2011-05-20 03:19:53 -04:00
|
|
|
|
|
|
|
// Incremented every time a fetch fails so that client nodes know to clear
|
|
|
|
// their cache of map output locations if this happens.
|
|
|
|
private var generation: Long = 0
|
|
|
|
private var generationLock = new java.lang.Object
|
2012-06-07 03:25:47 -04:00
|
|
|
|
2012-09-12 17:54:40 -04:00
|
|
|
// Cache a serialized version of the output locations for each shuffle to send them out faster
|
|
|
|
var cacheGeneration = generation
|
|
|
|
val cachedSerializedLocs = new HashMap[Int, Array[Byte]]
|
|
|
|
|
2012-06-07 03:25:47 -04:00
|
|
|
var trackerActor: ActorRef = if (isMaster) {
|
2012-09-12 17:54:40 -04:00
|
|
|
val actor = actorSystem.actorOf(Props(new MapOutputTrackerActor(this)), name = actorName)
|
2012-06-29 02:51:28 -04:00
|
|
|
logInfo("Registered MapOutputTrackerActor actor")
|
2012-06-07 03:25:47 -04:00
|
|
|
actor
|
2011-05-17 15:41:13 -04:00
|
|
|
} else {
|
2012-06-29 19:01:36 -04:00
|
|
|
val url = "akka://spark@%s:%s/user/%s".format(ip, port, actorName)
|
2012-06-29 02:51:28 -04:00
|
|
|
actorSystem.actorFor(url)
|
|
|
|
}
|
|
|
|
|
|
|
|
// Send a message to the trackerActor and get its result within a default timeout, or
|
|
|
|
// throw a SparkException if this fails.
|
|
|
|
def askTracker(message: Any): Any = {
|
|
|
|
try {
|
|
|
|
val future = trackerActor.ask(message)(timeout)
|
|
|
|
return Await.result(future, timeout)
|
|
|
|
} catch {
|
|
|
|
case e: Exception =>
|
|
|
|
throw new SparkException("Error communicating with MapOutputTracker", e)
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
// Send a one-way message to the trackerActor, to which we expect it to reply with true.
|
|
|
|
def communicate(message: Any) {
|
|
|
|
if (askTracker(message) != true) {
|
|
|
|
throw new SparkException("Error reply received from MapOutputTracker")
|
|
|
|
}
|
2011-02-27 22:15:52 -05:00
|
|
|
}
|
2012-01-05 15:59:20 -05:00
|
|
|
|
|
|
|
def registerShuffle(shuffleId: Int, numMaps: Int) {
|
2012-06-07 03:25:47 -04:00
|
|
|
if (bmAddresses.get(shuffleId) != null) {
|
2012-01-05 15:59:20 -05:00
|
|
|
throw new IllegalArgumentException("Shuffle ID " + shuffleId + " registered twice")
|
|
|
|
}
|
2012-06-07 03:25:47 -04:00
|
|
|
bmAddresses.put(shuffleId, new Array[BlockManagerId](numMaps))
|
2012-01-05 15:59:20 -05:00
|
|
|
}
|
2011-02-27 22:15:52 -05:00
|
|
|
|
2012-06-07 03:25:47 -04:00
|
|
|
def registerMapOutput(shuffleId: Int, mapId: Int, bmAddress: BlockManagerId) {
|
|
|
|
var array = bmAddresses.get(shuffleId)
|
2011-05-20 03:19:53 -04:00
|
|
|
array.synchronized {
|
2012-06-07 03:25:47 -04:00
|
|
|
array(mapId) = bmAddress
|
2011-05-20 03:19:53 -04:00
|
|
|
}
|
2011-02-27 17:27:12 -05:00
|
|
|
}
|
|
|
|
|
2012-06-07 03:25:47 -04:00
|
|
|
def registerMapOutputs(shuffleId: Int, locs: Array[BlockManagerId], changeGeneration: Boolean = false) {
|
|
|
|
bmAddresses.put(shuffleId, Array[BlockManagerId]() ++ locs)
|
|
|
|
if (changeGeneration) {
|
|
|
|
incrementGeneration()
|
|
|
|
}
|
2011-02-27 17:27:12 -05:00
|
|
|
}
|
2011-05-20 03:19:53 -04:00
|
|
|
|
2012-06-07 03:25:47 -04:00
|
|
|
def unregisterMapOutput(shuffleId: Int, mapId: Int, bmAddress: BlockManagerId) {
|
|
|
|
var array = bmAddresses.get(shuffleId)
|
2011-05-20 03:19:53 -04:00
|
|
|
if (array != null) {
|
|
|
|
array.synchronized {
|
2012-06-07 03:25:47 -04:00
|
|
|
if (array(mapId) == bmAddress) {
|
2011-05-20 03:19:53 -04:00
|
|
|
array(mapId) = null
|
2012-02-10 11:19:53 -05:00
|
|
|
}
|
2011-05-20 03:19:53 -04:00
|
|
|
}
|
|
|
|
incrementGeneration()
|
|
|
|
} else {
|
|
|
|
throw new SparkException("unregisterMapOutput called for nonexistent shuffle ID")
|
|
|
|
}
|
|
|
|
}
|
2011-02-27 17:27:12 -05:00
|
|
|
|
2011-05-20 03:19:53 -04:00
|
|
|
// Remembers which map output locations are currently being fetched on a worker
|
2011-03-06 19:16:38 -05:00
|
|
|
val fetching = new HashSet[Int]
|
|
|
|
|
2011-05-20 03:19:53 -04:00
|
|
|
// Called on possibly remote nodes to get the server URIs for a given shuffle
|
2012-06-07 03:25:47 -04:00
|
|
|
def getServerAddresses(shuffleId: Int): Array[BlockManagerId] = {
|
|
|
|
val locs = bmAddresses.get(shuffleId)
|
2011-03-06 19:16:38 -05:00
|
|
|
if (locs == null) {
|
2012-08-28 02:07:50 -04:00
|
|
|
logInfo("Don't have map outputs for shuffle " + shuffleId + ", fetching them")
|
2011-03-06 19:16:38 -05:00
|
|
|
fetching.synchronized {
|
|
|
|
if (fetching.contains(shuffleId)) {
|
|
|
|
// Someone else is fetching it; wait for them to be done
|
|
|
|
while (fetching.contains(shuffleId)) {
|
2012-02-10 11:19:53 -05:00
|
|
|
try {
|
|
|
|
fetching.wait()
|
|
|
|
} catch {
|
|
|
|
case _ =>
|
|
|
|
}
|
2011-03-06 19:16:38 -05:00
|
|
|
}
|
2012-06-07 03:25:47 -04:00
|
|
|
return bmAddresses.get(shuffleId)
|
2011-03-06 19:16:38 -05:00
|
|
|
} else {
|
|
|
|
fetching += shuffleId
|
|
|
|
}
|
|
|
|
}
|
|
|
|
// We won the race to fetch the output locs; do so
|
|
|
|
logInfo("Doing the fetch; tracker actor = " + trackerActor)
|
2012-09-12 17:54:40 -04:00
|
|
|
val fetchedBytes = askTracker(GetMapOutputLocations(shuffleId)).asInstanceOf[Array[Byte]]
|
|
|
|
val fetchedLocs = deserializeLocations(fetchedBytes)
|
2012-06-07 03:25:47 -04:00
|
|
|
|
|
|
|
logInfo("Got the output locations")
|
2012-09-12 17:54:40 -04:00
|
|
|
bmAddresses.put(shuffleId, fetchedLocs)
|
2011-03-06 19:16:38 -05:00
|
|
|
fetching.synchronized {
|
|
|
|
fetching -= shuffleId
|
|
|
|
fetching.notifyAll()
|
|
|
|
}
|
2012-09-12 17:54:40 -04:00
|
|
|
return fetchedLocs
|
2011-03-06 19:16:38 -05:00
|
|
|
} else {
|
|
|
|
return locs
|
|
|
|
}
|
2011-02-27 17:27:12 -05:00
|
|
|
}
|
2011-05-13 15:03:58 -04:00
|
|
|
|
|
|
|
def stop() {
|
2012-06-29 02:51:28 -04:00
|
|
|
communicate(StopMapOutputTracker)
|
2012-06-07 03:25:47 -04:00
|
|
|
bmAddresses.clear()
|
2011-05-13 15:03:58 -04:00
|
|
|
trackerActor = null
|
|
|
|
}
|
2011-05-20 03:19:53 -04:00
|
|
|
|
|
|
|
// Called on master to increment the generation number
|
|
|
|
def incrementGeneration() {
|
|
|
|
generationLock.synchronized {
|
|
|
|
generation += 1
|
2012-08-28 02:07:50 -04:00
|
|
|
logDebug("Increasing generation to " + generation)
|
2011-05-20 03:19:53 -04:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
// Called on master or workers to get current generation number
|
|
|
|
def getGeneration: Long = {
|
|
|
|
generationLock.synchronized {
|
|
|
|
return generation
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
// Called on workers to update the generation number, potentially clearing old outputs
|
|
|
|
// because of a fetch failure. (Each Mesos task calls this with the latest generation
|
|
|
|
// number on the master at the time it was created.)
|
|
|
|
def updateGeneration(newGen: Long) {
|
|
|
|
generationLock.synchronized {
|
|
|
|
if (newGen > generation) {
|
|
|
|
logInfo("Updating generation to " + newGen + " and clearing cache")
|
2012-06-07 03:25:47 -04:00
|
|
|
bmAddresses = new ConcurrentHashMap[Int, Array[BlockManagerId]]
|
2011-05-20 03:19:53 -04:00
|
|
|
generation = newGen
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
2012-09-12 17:54:40 -04:00
|
|
|
|
|
|
|
def getSerializedLocations(shuffleId: Int): Array[Byte] = {
|
|
|
|
var locs: Array[BlockManagerId] = null
|
|
|
|
var generationGotten: Long = -1
|
|
|
|
generationLock.synchronized {
|
|
|
|
if (generation > cacheGeneration) {
|
|
|
|
cachedSerializedLocs.clear()
|
|
|
|
cacheGeneration = generation
|
|
|
|
}
|
|
|
|
cachedSerializedLocs.get(shuffleId) match {
|
|
|
|
case Some(bytes) =>
|
|
|
|
return bytes
|
|
|
|
case None =>
|
|
|
|
locs = bmAddresses.get(shuffleId)
|
|
|
|
generationGotten = generation
|
|
|
|
}
|
|
|
|
}
|
|
|
|
// If we got here, we failed to find the serialized locations in the cache, so we pulled
|
|
|
|
// out a snapshot of the locations as "locs"; let's serialize and return that
|
|
|
|
val bytes = serializeLocations(locs)
|
|
|
|
// Add them into the table only if the generation hasn't changed while we were working
|
|
|
|
generationLock.synchronized {
|
|
|
|
if (generation == generationGotten) {
|
|
|
|
cachedSerializedLocs(shuffleId) = bytes
|
|
|
|
}
|
|
|
|
}
|
|
|
|
return bytes
|
|
|
|
}
|
|
|
|
|
|
|
|
// Serialize an array of map output locations into an efficient byte format so that we can send
|
|
|
|
// it to reduce tasks. We do this by grouping together the locations by block manager ID.
|
|
|
|
def serializeLocations(locs: Array[BlockManagerId]): Array[Byte] = {
|
|
|
|
val out = new ByteArrayOutputStream
|
|
|
|
val dataOut = new DataOutputStream(out)
|
|
|
|
dataOut.writeInt(locs.length)
|
|
|
|
val grouped = locs.zipWithIndex.groupBy(_._1)
|
|
|
|
dataOut.writeInt(grouped.size)
|
|
|
|
for ((id, pairs) <- grouped if id != null) {
|
|
|
|
dataOut.writeUTF(id.ip)
|
|
|
|
dataOut.writeInt(id.port)
|
|
|
|
dataOut.writeInt(pairs.length)
|
|
|
|
for ((_, blockIndex) <- pairs) {
|
|
|
|
dataOut.writeInt(blockIndex)
|
|
|
|
}
|
|
|
|
}
|
|
|
|
dataOut.close()
|
|
|
|
out.toByteArray
|
|
|
|
}
|
|
|
|
|
|
|
|
// Opposite of serializeLocations.
|
|
|
|
def deserializeLocations(bytes: Array[Byte]): Array[BlockManagerId] = {
|
|
|
|
val dataIn = new DataInputStream(new ByteArrayInputStream(bytes))
|
|
|
|
val length = dataIn.readInt()
|
|
|
|
val array = new Array[BlockManagerId](length)
|
|
|
|
val numGroups = dataIn.readInt()
|
|
|
|
for (i <- 0 until numGroups) {
|
|
|
|
val ip = dataIn.readUTF()
|
|
|
|
val port = dataIn.readInt()
|
|
|
|
val id = new BlockManagerId(ip, port)
|
|
|
|
val numBlocks = dataIn.readInt()
|
|
|
|
for (j <- 0 until numBlocks) {
|
|
|
|
array(dataIn.readInt()) = id
|
|
|
|
}
|
|
|
|
}
|
|
|
|
array
|
|
|
|
}
|
2011-05-13 15:03:58 -04:00
|
|
|
}
|