[SPARK-25299] Local shuffle implementation of the shuffle writer API (apache-spark-on-k8s#524)

Implements the shuffle writer API by writing shuffle files to local disk and using the index block resolver to commit data and write index files.

The logic in `BypassMergeSortShuffleWriter` has been refactored to use the base implementation of the plugin instead.

APIs have been slightly renamed to clarify semantics after considering nuances in how these are to be implemented by other developers.

Follow-up commits are to come for `SortShuffleWriter` and `UnsafeShuffleWriter`.

Ported from #6, credits to @ifilonenko.

Author: mccheah

core/src/main/java/org/apache/spark/api/shuffle/ShuffleExecutorComponents.java

@@ -27,7 +27,7 @@
  */
 @Experimental
 public interface ShuffleExecutorComponents {
-  void intitializeExecutor(String appId, String execId);
+  void initializeExecutor(String appId, String execId);
 
   ShuffleWriteSupport writes();
 }

core/src/main/java/org/apache/spark/api/shuffle/ShufflePartitionWriter.java

@@ -17,6 +17,7 @@
 
 package org.apache.spark.api.shuffle;
 
+import java.io.Closeable;
 import java.io.IOException;
 import java.io.OutputStream;
 import java.nio.channels.Channels;
@@ -26,17 +27,48 @@
 
 /**
  * :: Experimental ::
- * An interface for giving streams / channels for shuffle writes
+ * An interface for giving streams / channels for shuffle writes.
  *
  * @since 3.0.0
  */
 @Experimental
-public interface ShufflePartitionWriter {
-  OutputStream openStream() throws IOException;
+public interface ShufflePartitionWriter extends Closeable {
 
-  long getLength();
+  /**
+   * Returns an underlying {@link OutputStream} that can write bytes to the underlying data store.
+   * <p>
+   * Note that this stream itself is not closed by the caller; close the stream in the
+   * implementation of this interface's {@link #close()}.
+   */
+  OutputStream toStream() throws IOException;
 
-  default WritableByteChannel openChannel() throws IOException {
-    return Channels.newChannel(openStream());
+  /**
+   * Returns an underlying {@link WritableByteChannel} that can write bytes to the underlying data
+   * store.
+   * <p>
+   * Note that this channel itself is not closed by the caller; close the channel in the
+   * implementation of this interface's {@link #close()}.
+   */
+  default WritableByteChannel toChannel() throws IOException {
+    return Channels.newChannel(toStream());
   }
+
+  /**
+   * Get the number of bytes written by this writer's stream returned by {@link #toStream()} or
+   * the channel returned by {@link #toChannel()}.
+   */
+  long getNumBytesWritten();
+
+  /**
+   * Close all resources created by this ShufflePartitionWriter, via calls to {@link #toStream()}
+   * or {@link #toChannel()}.
+   * <p>
+   * This must always close any stream returned by {@link #toStream()}.
+   * <p>
+   * Note that the default version of {@link #toChannel()} returns a {@link WritableByteChannel}
+   * that does not itself need to be closed up front; only the underlying output stream given by
+   * {@link #toStream()} must be closed.
+   */
+  @Override
+  void close() throws IOException;
 }

core/src/main/java/org/apache/spark/api/shuffle/ShuffleWriteSupport.java

@@ -30,7 +30,6 @@
 @Experimental
 public interface ShuffleWriteSupport {
   ShuffleMapOutputWriter createMapOutputWriter(
-    String appId,
     int shuffleId,
     int mapId,
     int numPartitions) throws IOException;

core/src/main/java/org/apache/spark/shuffle/sort/BypassMergeSortShuffleWriter.java

@@ -19,8 +19,10 @@
 
 import java.io.File;
 import java.io.FileInputStream;
-import java.io.FileOutputStream;
 import java.io.IOException;
+import java.io.OutputStream;
+import java.nio.channels.FileChannel;
+import java.nio.channels.WritableByteChannel;
 import javax.annotation.Nullable;
 
 import scala.None$;
@@ -34,6 +36,9 @@
 import org.slf4j.Logger;
 import org.slf4j.LoggerFactory;
 
+import org.apache.spark.api.shuffle.ShuffleMapOutputWriter;
+import org.apache.spark.api.shuffle.ShufflePartitionWriter;
+import org.apache.spark.api.shuffle.ShuffleWriteSupport;
 import org.apache.spark.internal.config.package$;
 import org.apache.spark.Partitioner;
 import org.apache.spark.ShuffleDependency;
@@ -82,6 +87,7 @@ final class BypassMergeSortShuffleWriter<K, V> extends ShuffleWriter<K, V> {
   private final int shuffleId;
   private final int mapId;
   private final Serializer serializer;
+  private final ShuffleWriteSupport shuffleWriteSupport;
   private final IndexShuffleBlockResolver shuffleBlockResolver;
 
   /** Array of file writers, one for each partition */
@@ -103,7 +109,8 @@ final class BypassMergeSortShuffleWriter<K, V> extends ShuffleWriter<K, V> {
       BypassMergeSortShuffleHandle<K, V> handle,
       int mapId,
       SparkConf conf,
-      ShuffleWriteMetricsReporter writeMetrics) {
+      ShuffleWriteMetricsReporter writeMetrics,
+      ShuffleWriteSupport shuffleWriteSupport) {
     // Use getSizeAsKb (not bytes) to maintain backwards compatibility if no units are provided
     this.fileBufferSize = (int) (long) conf.get(package$.MODULE$.SHUFFLE_FILE_BUFFER_SIZE()) * 1024;
     this.transferToEnabled = conf.getBoolean("spark.file.transferTo", true);
@@ -116,57 +123,61 @@ final class BypassMergeSortShuffleWriter<K, V> extends ShuffleWriter<K, V> {
     this.writeMetrics = writeMetrics;
     this.serializer = dep.serializer();
     this.shuffleBlockResolver = shuffleBlockResolver;
+    this.shuffleWriteSupport = shuffleWriteSupport;
   }
 
   @Override
   public void write(Iterator<Product2<K, V>> records) throws IOException {
     assert (partitionWriters == null);
-    if (!records.hasNext()) {
-      partitionLengths = new long[numPartitions];
-      shuffleBlockResolver.writeIndexFileAndCommit(shuffleId, mapId, partitionLengths, null);
-      mapStatus = MapStatus$.MODULE$.apply(blockManager.shuffleServerId(), partitionLengths);
-      return;
-    }
-    final SerializerInstance serInstance = serializer.newInstance();
-    final long openStartTime = System.nanoTime();
-    partitionWriters = new DiskBlockObjectWriter[numPartitions];
-    partitionWriterSegments = new FileSegment[numPartitions];
-    for (int i = 0; i < numPartitions; i++) {
-      final Tuple2<TempShuffleBlockId, File> tempShuffleBlockIdPlusFile =
-        blockManager.diskBlockManager().createTempShuffleBlock();
-      final File file = tempShuffleBlockIdPlusFile._2();
-      final BlockId blockId = tempShuffleBlockIdPlusFile._1();
-      partitionWriters[i] =
-        blockManager.getDiskWriter(blockId, file, serInstance, fileBufferSize, writeMetrics);
-    }
-    // Creating the file to write to and creating a disk writer both involve interacting with
-    // the disk, and can take a long time in aggregate when we open many files, so should be
-    // included in the shuffle write time.
-    writeMetrics.incWriteTime(System.nanoTime() - openStartTime);
-
-    while (records.hasNext()) {
-      final Product2<K, V> record = records.next();
-      final K key = record._1();
-      partitionWriters[partitioner.getPartition(key)].write(key, record._2());
-    }
+    ShuffleMapOutputWriter mapOutputWriter = shuffleWriteSupport
+        .createMapOutputWriter(shuffleId, mapId, numPartitions);
+    try {
+      if (!records.hasNext()) {
+        partitionLengths = new long[numPartitions];
+        mapOutputWriter.commitAllPartitions();
+        mapStatus = MapStatus$.MODULE$.apply(blockManager.shuffleServerId(), partitionLengths);
+        return;
+      }
+      final SerializerInstance serInstance = serializer.newInstance();
+      final long openStartTime = System.nanoTime();
+      partitionWriters = new DiskBlockObjectWriter[numPartitions];
+      partitionWriterSegments = new FileSegment[numPartitions];
+      for (int i = 0; i < numPartitions; i++) {
+        final Tuple2<TempShuffleBlockId, File> tempShuffleBlockIdPlusFile =
+            blockManager.diskBlockManager().createTempShuffleBlock();
+        final File file = tempShuffleBlockIdPlusFile._2();
+        final BlockId blockId = tempShuffleBlockIdPlusFile._1();
+        partitionWriters[i] =
+            blockManager.getDiskWriter(blockId, file, serInstance, fileBufferSize, writeMetrics);
+      }
+      // Creating the file to write to and creating a disk writer both involve interacting with
+      // the disk, and can take a long time in aggregate when we open many files, so should be
+      // included in the shuffle write time.
+      writeMetrics.incWriteTime(System.nanoTime() - openStartTime);
 
-    for (int i = 0; i < numPartitions; i++) {
-      try (DiskBlockObjectWriter writer = partitionWriters[i]) {
-        partitionWriterSegments[i] = writer.commitAndGet();
+      while (records.hasNext()) {
+        final Product2<K, V> record = records.next();
+        final K key = record._1();
+        partitionWriters[partitioner.getPartition(key)].write(key, record._2());
       }
-    }
 
-    File output = shuffleBlockResolver.getDataFile(shuffleId, mapId);
-    File tmp = Utils.tempFileWith(output);
-    try {
-      partitionLengths = writePartitionedFile(tmp);
-      shuffleBlockResolver.writeIndexFileAndCommit(shuffleId, mapId, partitionLengths, tmp);
-    } finally {
-      if (tmp.exists() && !tmp.delete()) {
-        logger.error("Error while deleting temp file {}", tmp.getAbsolutePath());
+      for (int i = 0; i < numPartitions; i++) {
+        try (DiskBlockObjectWriter writer = partitionWriters[i]) {
+          partitionWriterSegments[i] = writer.commitAndGet();
+        }
       }
+
+      partitionLengths = writePartitionedData(mapOutputWriter);
+      mapOutputWriter.commitAllPartitions();
+      mapStatus = MapStatus$.MODULE$.apply(blockManager.shuffleServerId(), partitionLengths);
+    } catch (Exception e) {
+      try {
+        mapOutputWriter.abort(e);
+      } catch (Exception e2) {
+        logger.error("Failed to abort the writer after failing to write map output.", e2);
+      }
+      throw e;
     }
-    mapStatus = MapStatus$.MODULE$.apply(blockManager.shuffleServerId(), partitionLengths);
   }
 
   @VisibleForTesting
@@ -179,37 +190,54 @@ long[] getPartitionLengths() {
    *
    * @return array of lengths, in bytes, of each partition of the file (used by map output tracker).
    */
-  private long[] writePartitionedFile(File outputFile) throws IOException {
+  private long[] writePartitionedData(ShuffleMapOutputWriter mapOutputWriter) throws IOException {
     // Track location of the partition starts in the output file
     final long[] lengths = new long[numPartitions];
     if (partitionWriters == null) {
       // We were passed an empty iterator
       return lengths;
     }
-
-    final FileOutputStream out = new FileOutputStream(outputFile, true);
     final long writeStartTime = System.nanoTime();
-    boolean threwException = true;
     try {
       for (int i = 0; i < numPartitions; i++) {
         final File file = partitionWriterSegments[i].file();
-        if (file.exists()) {
-          final FileInputStream in = new FileInputStream(file);
-          boolean copyThrewException = true;
-          try {
-            lengths[i] = Utils.copyStream(in, out, false, transferToEnabled);
+        boolean copyThrewException = true;
+        ShufflePartitionWriter writer = null;
+        try {
+          writer = mapOutputWriter.getNextPartitionWriter();
+          if (!file.exists()) {
             copyThrewException = false;
-          } finally {
-            Closeables.close(in, copyThrewException);
-          }
-          if (!file.delete()) {
-            logger.error("Unable to delete file for partition {}", i);
+          } else {
+            if (transferToEnabled) {
+              WritableByteChannel outputChannel = writer.toChannel();
+              FileInputStream in = new FileInputStream(file);
+              try (FileChannel inputChannel = in.getChannel()) {
+                Utils.copyFileStreamNIO(inputChannel, outputChannel, 0, inputChannel.size());
+                copyThrewException = false;
+              } finally {
+                Closeables.close(in, copyThrewException);
+              }
+            } else {
+              OutputStream tempOutputStream = writer.toStream();
+              FileInputStream in = new FileInputStream(file);
+              try {
+                Utils.copyStream(in, tempOutputStream, false, false);
+                copyThrewException = false;
+              } finally {
+                Closeables.close(in, copyThrewException);
+              }
+            }
+            if (!file.delete()) {
+              logger.error("Unable to delete file for partition {}", i);
+            }
           }
+        } finally {
+          Closeables.close(writer, copyThrewException);
         }
+
+        lengths[i] = writer.getNumBytesWritten();
       }
-      threwException = false;
     } finally {
-      Closeables.close(out, threwException);
       writeMetrics.incWriteTime(System.nanoTime() - writeStartTime);
     }
     partitionWriters = null;

core/src/main/java/org/apache/spark/shuffle/sort/io/DefaultShuffleDataIO.java

@@ -0,0 +1,36 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one or more
+ * contributor license agreements.  See the NOTICE file distributed with
+ * this work for additional information regarding copyright ownership.
+ * The ASF licenses this file to You under the Apache License, Version 2.0
+ * (the "License"); you may not use this file except in compliance with
+ * the License.  You may obtain a copy of the License at
+ *
+ *    http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package org.apache.spark.shuffle.sort.io;
+
+import org.apache.spark.SparkConf;
+import org.apache.spark.api.shuffle.ShuffleExecutorComponents;
+import org.apache.spark.api.shuffle.ShuffleDataIO;
+
+public class DefaultShuffleDataIO implements ShuffleDataIO {
+
+  private final SparkConf sparkConf;
+
+  public DefaultShuffleDataIO(SparkConf sparkConf) {
+    this.sparkConf = sparkConf;
+  }
+
+  @Override
+  public ShuffleExecutorComponents executor() {
+    return new DefaultShuffleExecutorComponents(sparkConf);
+  }
+}

core/src/main/java/org/apache/spark/shuffle/sort/io/DefaultShuffleExecutorComponents.java

@@ -0,0 +1,51 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one or more
+ * contributor license agreements.  See the NOTICE file distributed with
+ * this work for additional information regarding copyright ownership.
+ * The ASF licenses this file to You under the Apache License, Version 2.0
+ * (the "License"); you may not use this file except in compliance with
+ * the License.  You may obtain a copy of the License at
+ *
+ *    http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package org.apache.spark.shuffle.sort.io;
+
+import org.apache.spark.SparkConf;
+import org.apache.spark.SparkEnv;
+import org.apache.spark.api.shuffle.ShuffleExecutorComponents;
+import org.apache.spark.api.shuffle.ShuffleWriteSupport;
+import org.apache.spark.shuffle.IndexShuffleBlockResolver;
+import org.apache.spark.storage.BlockManager;
+
+public class DefaultShuffleExecutorComponents implements ShuffleExecutorComponents {
+
+  private final SparkConf sparkConf;
+  private BlockManager blockManager;
+  private IndexShuffleBlockResolver blockResolver;
+
+  public DefaultShuffleExecutorComponents(SparkConf sparkConf) {
+    this.sparkConf = sparkConf;
+  }
+
+  @Override
+  public void initializeExecutor(String appId, String execId) {
+    blockManager = SparkEnv.get().blockManager();
+    blockResolver = new IndexShuffleBlockResolver(sparkConf, blockManager);
+  }
+
+  @Override
+  public ShuffleWriteSupport writes() {
+    if (blockResolver == null) {
+      throw new IllegalStateException(
+        "Executor components must be initialized before getting writers.");
+    }
+    return new DefaultShuffleWriteSupport(sparkConf, blockResolver);
+  }
+}