Modify to propagete error using ConnectionManager

xiliu82 · Jul 19, 2014 · e2b8c4a · e2b8c4a
1 parent 30b8d36
commit e2b8c4a
Show file tree

Hide file tree

Showing 6 changed files with 46 additions and 19 deletions.
diff --git a/core/src/main/scala/org/apache/spark/network/BufferMessage.scala b/core/src/main/scala/org/apache/spark/network/BufferMessage.scala
@@ -48,7 +48,7 @@ class BufferMessage(id_ : Int, val buffers: ArrayBuffer[ByteBuffer], var ackId:
     val security = if (isSecurityNeg) 1 else 0
     if (size == 0 && !gotChunkForSendingOnce) {
       val newChunk = new MessageChunk(
-        new MessageChunkHeader(typ, id, 0, 0, ackId, security, senderAddress), null)
+        new MessageChunkHeader(typ, id, 0, 0, ackId, hasError, security, senderAddress), null)
       gotChunkForSendingOnce = true
       return Some(newChunk)
     }
@@ -66,7 +66,7 @@ class BufferMessage(id_ : Int, val buffers: ArrayBuffer[ByteBuffer], var ackId:
         }
         buffer.position(buffer.position + newBuffer.remaining)
         val newChunk = new MessageChunk(new MessageChunkHeader(
-            typ, id, size, newBuffer.remaining, ackId, security, senderAddress), newBuffer)
+            typ, id, size, newBuffer.remaining, ackId, hasError, security, senderAddress), newBuffer)
         gotChunkForSendingOnce = true
         return Some(newChunk)
       }
@@ -88,7 +88,7 @@ class BufferMessage(id_ : Int, val buffers: ArrayBuffer[ByteBuffer], var ackId:
       val newBuffer = buffer.slice().limit(chunkSize).asInstanceOf[ByteBuffer]
       buffer.position(buffer.position + newBuffer.remaining)
       val newChunk = new MessageChunk(new MessageChunkHeader(
-          typ, id, size, newBuffer.remaining, ackId, security, senderAddress), newBuffer)
+          typ, id, size, newBuffer.remaining, ackId, hasError, security, senderAddress), newBuffer)
       return Some(newChunk)
     }
     None

diff --git a/core/src/main/scala/org/apache/spark/network/ConnectionManager.scala b/core/src/main/scala/org/apache/spark/network/ConnectionManager.scala
@@ -674,7 +674,6 @@ private[spark] class ConnectionManager(port: Int, conf: SparkConf,
               ackMessage.get.asInstanceOf[BufferMessage].ackId = bufferMessage.id
             }
           }
-
           sendMessage(connectionManagerId, ackMessage.getOrElse {
             Message.createBufferMessage(bufferMessage.id)
           })

diff --git a/core/src/main/scala/org/apache/spark/network/Message.scala b/core/src/main/scala/org/apache/spark/network/Message.scala
@@ -28,6 +28,7 @@ private[spark] abstract class Message(val typ: Long, val id: Int) {
   var startTime = -1L
   var finishTime = -1L
   var isSecurityNeg = false
+  var hasError = false
 
   def size: Int
 
@@ -87,6 +88,7 @@ private[spark] object Message {
       case BUFFER_MESSAGE => new BufferMessage(header.id,
         ArrayBuffer(ByteBuffer.allocate(header.totalSize)), header.other)
     }
+    newMessage.hasError = header.hasError
     newMessage.senderAddress = header.address
     newMessage
   }

diff --git a/core/src/main/scala/org/apache/spark/network/MessageChunkHeader.scala b/core/src/main/scala/org/apache/spark/network/MessageChunkHeader.scala
@@ -27,6 +27,7 @@ private[spark] class MessageChunkHeader(
     val totalSize: Int,
     val chunkSize: Int,
     val other: Int,
+    val hasError: Boolean,
     val securityNeg: Int,
     val address: InetSocketAddress) {
   lazy val buffer = {
@@ -41,6 +42,13 @@ private[spark] class MessageChunkHeader(
       putInt(totalSize).
       putInt(chunkSize).
       putInt(other).
+      put{
+        if (hasError) {
+          1.asInstanceOf[Byte]
+        } else {
+          0.asInstanceOf[Byte]
+        }
+      }.
       putInt(securityNeg).
       putInt(ip.size).
       put(ip).
@@ -56,7 +64,7 @@ private[spark] class MessageChunkHeader(
 
 
 private[spark] object MessageChunkHeader {
-  val HEADER_SIZE = 44
+  val HEADER_SIZE = 45
 
   def create(buffer: ByteBuffer): MessageChunkHeader = {
     if (buffer.remaining != HEADER_SIZE) {
@@ -67,13 +75,20 @@ private[spark] object MessageChunkHeader {
     val totalSize = buffer.getInt()
     val chunkSize = buffer.getInt()
     val other = buffer.getInt()
+    val hasError = {
+      if (buffer.get() == 0) {
+        false
+      } else {
+        true
+      }
+    }
     val securityNeg = buffer.getInt()
     val ipSize = buffer.getInt()
     val ipBytes = new Array[Byte](ipSize)
     buffer.get(ipBytes)
     val ip = InetAddress.getByAddress(ipBytes)
     val port = buffer.getInt()
-    new MessageChunkHeader(typ, id, totalSize, chunkSize, other, securityNeg,
+    new MessageChunkHeader(typ, id, totalSize, chunkSize, other, hasError, securityNeg,
       new InetSocketAddress(ip, port))
   }
 }
diff --git a/core/src/main/scala/org/apache/spark/storage/BlockFetcherIterator.scala b/core/src/main/scala/org/apache/spark/storage/BlockFetcherIterator.scala
@@ -122,18 +122,25 @@ object BlockFetcherIterator {
       future.onSuccess {
         case Some(message) => {
           val bufferMessage = message.asInstanceOf[BufferMessage]
-          val blockMessageArray = BlockMessageArray.fromBufferMessage(bufferMessage)
-          for (blockMessage <- blockMessageArray) {
-            if (blockMessage.getType != BlockMessage.TYPE_GOT_BLOCK) {
-              throw new SparkException(
-                "Unexpected message " + blockMessage.getType + " received from " + cmId)
+          if (bufferMessage.hasError) {
+            logError("Could not get block(s) from " + cmId)
+            for ((blockId, size) <- req.blocks) {
+              results.put(new FetchResult(blockId, -1, null))
+            }
+          } else {
+            val blockMessageArray = BlockMessageArray.fromBufferMessage(bufferMessage)
+            for (blockMessage <- blockMessageArray) {
+              if (blockMessage.getType != BlockMessage.TYPE_GOT_BLOCK) {
+                throw new SparkException(
+                  "Unexpected message " + blockMessage.getType + " received from " + cmId)
+              }
+              val blockId = blockMessage.getId
+              val networkSize = blockMessage.getData.limit()
+              results.put(new FetchResult(blockId, sizeMap(blockId),
+                () => dataDeserialize(blockId, blockMessage.getData, serializer)))
+              _remoteBytesRead += networkSize
+              logDebug("Got remote block " + blockId + " after " + Utils.getUsedTimeMs(startTime))
             }
-            val blockId = blockMessage.getId
-            val networkSize = blockMessage.getData.limit()
-            results.put(new FetchResult(blockId, sizeMap(blockId),
-              () => dataDeserialize(blockId, blockMessage.getData, serializer)))
-            _remoteBytesRead += networkSize
-            logDebug("Got remote block " + blockId + " after " + Utils.getUsedTimeMs(startTime))
           }
         }
         case None => {

diff --git a/core/src/main/scala/org/apache/spark/storage/BlockManagerWorker.scala b/core/src/main/scala/org/apache/spark/storage/BlockManagerWorker.scala
@@ -44,8 +44,12 @@ private[spark] class BlockManagerWorker(val blockManager: BlockManager) extends
           val responseMessages = blockMessages.map(processBlockMessage).filter(_ != None).map(_.get)
           Some(new BlockMessageArray(responseMessages).toBufferMessage)
         } catch {
-          case e: Exception => logError("Exception handling buffer message", e)
-          None
+          case e: Exception => {
+            logError("Exception handling buffer message", e)
+            val errorMessage = Message.createBufferMessage(msg.id)
+            errorMessage.hasError = true
+            Some(errorMessage)
+          }
         }
       }
       case otherMessage: Any => {
-Original file line number
+Diff line change
@@ Expand Up @@
                   ackMessage.get.asInstanceOf[BufferMessage].ackId = bufferMessage.id
                 }
               }
               sendMessage(connectionManagerId, ackMessage.getOrElse {
                 Message.createBufferMessage(bufferMessage.id)
               })
@@ Expand Down @@