HDFS-15549. Improve DISK/ARCHIVE movement if they are on same filesystem

hadoop-common-project/hadoop-common/src/main/java/org/apache/hadoop/fs/HardLink.java

@@ -153,11 +153,11 @@ String[] linkCount(File file) throws IOException {
    */
 
   /**
-   * Creates a hardlink 
+   * Creates a hardlink.
    * @param file - existing source file
    * @param linkName - desired target link file
    */
-  public static void createHardLink(File file, File linkName) 
+  public static void createHardLink(File file, File linkName)
       throws IOException {
     if (file == null) {
       throw new IOException(

hadoop-common-project/hadoop-common/src/main/java/org/apache/hadoop/fs/StorageType.java

@@ -91,6 +91,11 @@ public static StorageType parseStorageType(String s) {
     return StorageType.valueOf(StringUtils.toUpperCase(s));
   }
 
+  public static boolean allowSameDiskTiering(StorageType storageType) {
+    return storageType == StorageType.DISK
+        || storageType == StorageType.ARCHIVE;
+  }
+
   private static List<StorageType> getNonTransientTypes() {
     List<StorageType> nonTransientTypes = new ArrayList<>();
     for (StorageType t : VALUES) {

hadoop-hdfs-project/hadoop-hdfs/src/main/java/org/apache/hadoop/hdfs/server/datanode/DirectoryScanner.java

@@ -322,7 +322,8 @@ public DirectoryScanner(FsDatasetSpi<?> dataset, Configuration conf) {
    * Start the scanner. The scanner will run every
    * {@link DFSConfigKeys#DFS_DATANODE_DIRECTORYSCAN_INTERVAL_KEY} seconds.
    */
-  void start() {
+  @VisibleForTesting
+  public void start() {
     shouldRun.set(true);
     long firstScanTime = ThreadLocalRandom.current().nextLong(scanPeriodMsecs);
 

hadoop-hdfs-project/hadoop-hdfs/src/main/java/org/apache/hadoop/hdfs/server/datanode/fsdataset/impl/FsDatasetImpl.java

@@ -48,6 +48,7 @@
 import javax.management.ObjectName;
 import javax.management.StandardMBean;
 
+import org.apache.hadoop.fs.HardLink;
 import org.apache.hadoop.thirdparty.com.google.common.annotations.VisibleForTesting;
 
 import org.apache.hadoop.HadoopIllegalArgumentException;
@@ -994,6 +995,20 @@ static File[] copyBlockFiles(long blockId, long genStamp,
         smallBufferSize, conf);
   }
 
+  /**
+   * Link the block and meta files for the given block to the given destination.
+   * @return the new meta and block files.
+   * @throws IOException
+   */
+  static File[] hardLinkBlockFiles(long blockId, long genStamp,
+      ReplicaInfo srcReplica, File destRoot) throws IOException {
+    final File destDir = DatanodeUtil.idToBlockDir(destRoot, blockId);
+    // blockName is same as the filename for the block
+    final File dstFile = new File(destDir, srcReplica.getBlockName());
+    final File dstMeta = FsDatasetUtil.getMetaFile(dstFile, genStamp);
+    return hardLinkBlockFiles(srcReplica, dstMeta, dstFile);
+  }
+
   static File[] copyBlockFiles(ReplicaInfo srcReplica, File dstMeta,
                                File dstFile, boolean calculateChecksum,
                                int smallBufferSize, final Configuration conf)
@@ -1026,6 +1041,34 @@ static File[] copyBlockFiles(ReplicaInfo srcReplica, File dstMeta,
     return new File[] {dstMeta, dstFile};
   }
 
+  static File[] hardLinkBlockFiles(ReplicaInfo srcReplica, File dstMeta,
+      File dstFile)
+      throws IOException {
+    // Create parent folder if not exists.
+    srcReplica.getFileIoProvider()
+        .mkdirs(srcReplica.getVolume(), dstFile.getParentFile());
+    try {
+      HardLink.createHardLink(
+          new File(srcReplica.getBlockURI()), dstFile);
+    } catch (IOException e) {
+      throw new IOException("Failed to hardLink "
+          + srcReplica + " block file to "
+          + dstFile, e);
+    }
+    try {
+      HardLink.createHardLink(
+          new File(srcReplica.getMetadataURI()), dstMeta);
+    } catch (IOException e) {
+      throw new IOException("Failed to hardLink "
+          + srcReplica + " metadata to "
+          + dstMeta, e);
+    }
+    if (LOG.isDebugEnabled()) {
+      LOG.info("Linked " + srcReplica.getBlockURI() + " to " + dstFile);
+    }
+    return new File[]{dstMeta, dstFile};
+  }
+
   /**
    * Move block files from one storage to another storage.
    * @return Returns the Old replicaInfo
@@ -1058,12 +1101,30 @@ public ReplicaInfo moveBlockAcrossStorage(ExtendedBlock block,
     }
 
     FsVolumeReference volumeRef = null;
+    boolean shouldConsiderSameMountVolume =
+        shouldConsiderSameMountVolume(replicaInfo.getVolume(),
+            targetStorageType, targetStorageId);
+    boolean useVolumeOnSameMount = false;
+
     try (AutoCloseableLock lock = datasetReadLock.acquire()) {
-      volumeRef = volumes.getNextVolume(targetStorageType, targetStorageId,
-          block.getNumBytes());
+      if (shouldConsiderSameMountVolume) {
+        volumeRef = volumes.getVolumeByMount(targetStorageType,
+            ((FsVolumeImpl) replicaInfo.getVolume()).getMount(),
+            block.getNumBytes());
+        if (volumeRef != null) {
+          useVolumeOnSameMount = true;
+        }
+      }
+      if (!useVolumeOnSameMount) {
+        volumeRef = volumes.getNextVolume(
+            targetStorageType,
+            targetStorageId,
+            block.getNumBytes()
+        );
+      }
     }
     try {
-      moveBlock(block, replicaInfo, volumeRef);
+      moveBlock(block, replicaInfo, volumeRef, useVolumeOnSameMount);
     } finally {
       if (volumeRef != null) {
         volumeRef.close();
@@ -1074,20 +1135,54 @@ public ReplicaInfo moveBlockAcrossStorage(ExtendedBlock block,
     return replicaInfo;
   }
 
+  /**
+   * When configuring DISK/ARCHIVE on same volume,
+   * check if we should find the counterpart on the same disk mount.
+   */
+  @VisibleForTesting
+  boolean shouldConsiderSameMountVolume(FsVolumeSpi fsVolume,
+      StorageType targetStorageType, String targetStorageID) {
+    if (targetStorageID != null && !targetStorageID.isEmpty()) {
+      return false;
+    }
+    if (!(fsVolume instanceof FsVolumeImpl)
+        || ((FsVolumeImpl) fsVolume).getMount().isEmpty()) {
+      return false;
+    }
+    StorageType sourceStorageType = fsVolume.getStorageType();
+    // Source/dest storage types are different
+    if (sourceStorageType == targetStorageType) {
+      return false;
+    }
+    // Source/dest storage types are either DISK or ARCHIVE.
+    return StorageType.allowSameDiskTiering(sourceStorageType)
+        && StorageType.allowSameDiskTiering(targetStorageType);
+  }
+
   /**
    * Moves a block from a given volume to another.
    *
    * @param block       - Extended Block
    * @param replicaInfo - ReplicaInfo
    * @param volumeRef   - Volume Ref - Closed by caller.
+   * @param moveBlockToLocalMount   - Whether we use shortcut
+   *                               to move block on same mount.
    * @return newReplicaInfo
    * @throws IOException
    */
   @VisibleForTesting
   ReplicaInfo moveBlock(ExtendedBlock block, ReplicaInfo replicaInfo,
-      FsVolumeReference volumeRef) throws IOException {
-    ReplicaInfo newReplicaInfo = copyReplicaToVolume(block, replicaInfo,
-        volumeRef);
+      FsVolumeReference volumeRef, boolean moveBlockToLocalMount)
+      throws IOException {
+    ReplicaInfo newReplicaInfo;
+    if (moveBlockToLocalMount) {
+      newReplicaInfo = moveReplicaToVolumeOnSameMount(block, replicaInfo,
+          volumeRef);
+    } else {
+      newReplicaInfo = copyReplicaToVolume(block, replicaInfo,
+          volumeRef);
+    }
+
     finalizeNewReplica(newReplicaInfo, block);
     removeOldReplica(replicaInfo, newReplicaInfo, block.getBlockPoolId());
     return newReplicaInfo;
@@ -1128,6 +1223,33 @@ ReplicaInfo copyReplicaToVolume(ExtendedBlock block, ReplicaInfo replicaInfo,
     return newReplicaInfo;
   }
 
+  /**
+   * Shortcut to use hardlink to move blocks on same mount.
+   * This is useful when moving blocks between storage types on same disk mount.
+   * Two cases need to be considered carefully:
+   * 1) Datanode restart in the middle should not cause data loss.
+   *    We use hardlink to avoid this.
+   * 2) Finalized blocks can be reopened to append.
+   *    This is already handled by dataset lock and gen stamp.
+   *    See HDFS-12942
+   *
+   * @param block       - Extended Block
+   * @param replicaInfo - ReplicaInfo
+   * @param volumeRef   - Volume Ref - Closed by caller.
+   * @return newReplicaInfo new replica object created in specified volume.
+   * @throws IOException
+   */
+  @VisibleForTesting
+  ReplicaInfo moveReplicaToVolumeOnSameMount(ExtendedBlock block,
+      ReplicaInfo replicaInfo,
+      FsVolumeReference volumeRef) throws IOException {
+    FsVolumeImpl targetVolume = (FsVolumeImpl) volumeRef.getVolume();
+    // Move files to temp dir first
+    ReplicaInfo newReplicaInfo = targetVolume.hardLinkBlockToTmpLocation(block,
+        replicaInfo);
+    return newReplicaInfo;
+  }
+
   /**
    * Finalizes newReplica by calling finalizeReplica internally.
    *
@@ -1177,7 +1299,7 @@ public ReplicaInfo moveBlockAcrossVolumes(ExtendedBlock block, FsVolumeSpi
     }
 
     try {
-      moveBlock(block, replicaInfo, volumeRef);
+      moveBlock(block, replicaInfo, volumeRef, false);
     } finally {
       if (volumeRef != null) {
         volumeRef.close();

hadoop-hdfs-project/hadoop-hdfs/src/main/java/org/apache/hadoop/hdfs/server/datanode/fsdataset/impl/FsVolumeImpl.java

@@ -484,9 +484,8 @@ long getActualNonDfsUsed() throws IOException {
     // should share the same amount of reserved capacity.
     // When calculating actual non dfs used,
     // exclude DFS used capacity by another volume.
-    if (enableSameDiskTiering &&
-        (storageType == StorageType.DISK
-            || storageType == StorageType.ARCHIVE)) {
+    if (enableSameDiskTiering
+        && StorageType.allowSameDiskTiering(storageType)) {
       StorageType counterpartStorageType = storageType == StorageType.DISK
           ? StorageType.ARCHIVE : StorageType.DISK;
       FsVolumeReference counterpartRef = dataset
@@ -1529,6 +1528,24 @@ public ReplicaInfo moveBlockToTmpLocation(ExtendedBlock block,
     return newReplicaInfo;
   }
 
+  public ReplicaInfo hardLinkBlockToTmpLocation(ExtendedBlock block,
+      ReplicaInfo replicaInfo) throws IOException {
+
+    File[] blockFiles = FsDatasetImpl.hardLinkBlockFiles(block.getBlockId(),
+        block.getGenerationStamp(), replicaInfo,
+        getTmpDir(block.getBlockPoolId()));
+
+    ReplicaInfo newReplicaInfo = new ReplicaBuilder(ReplicaState.TEMPORARY)
+        .setBlockId(replicaInfo.getBlockId())
+        .setGenerationStamp(replicaInfo.getGenerationStamp())
+        .setFsVolume(this)
+        .setDirectoryToUse(blockFiles[0].getParentFile())
+        .setBytesToReserve(0)
+        .build();
+    newReplicaInfo.setNumBytes(blockFiles[1].length());
+    return newReplicaInfo;
+  }
+
   public File[] copyBlockToLazyPersistLocation(String bpId, long blockId,
       long genStamp,
       ReplicaInfo replicaInfo,

hadoop-hdfs-project/hadoop-hdfs/src/main/java/org/apache/hadoop/hdfs/server/datanode/fsdataset/impl/FsVolumeList.java

@@ -111,6 +111,30 @@ private FsVolumeReference chooseVolume(List<FsVolumeImpl> list,
     }
   }
 
+  /**
+   * Get volume by disk mount to place a block.
+   * This is useful for same disk tiering.
+   *
+   * @param storageType The desired {@link StorageType}
+   * @param mount Disk mount of the volume
+   * @param blockSize Free space needed on the volume
+   * @return
+   * @throws IOException
+   */
+  FsVolumeReference getVolumeByMount(StorageType storageType,
+      String mount, long blockSize) throws IOException {
+    if (!enableSameDiskTiering) {
+      return null;
+    }
+    FsVolumeReference volume = mountVolumeMap
+        .getVolumeRefByMountAndStorageType(mount, storageType);
+    // Check if volume has enough capacity
+    if (volume != null && volume.getVolume().getAvailable() > blockSize) {
+      return volume;
+    }
+    return null;
+  }
+
   /** 
    * Get next volume.
    *
@@ -354,9 +378,8 @@ private void removeVolume(FsVolumeImpl target) {
    * Check if same disk tiering is applied to the volume.
    */
   private boolean isSameDiskTieringApplied(FsVolumeImpl target) {
-    return enableSameDiskTiering &&
-        (target.getStorageType() == StorageType.DISK
-            || target.getStorageType() == StorageType.ARCHIVE);
+    return enableSameDiskTiering
+        && StorageType.allowSameDiskTiering(target.getStorageType());
   }
 
   /**