HDFS-7128. Decommission slows way down when it gets towards the end. Contributed by Ming Ma.

hadoop-hdfs-project/hadoop-hdfs/CHANGES.txt

@@ -1066,6 +1066,9 @@ Release 2.6.0 - UNRELEASED
     HDFS-6995. Block should be placed in the client's 'rack-local' node
     if 'client-local' node is not available (vinayakumarb)
 
+    HDFS-7128. Decommission slows way down when it gets towards the end.
+    (Ming Ma via cnauroth)
+
   BREAKDOWN OF HDFS-6584 ARCHIVAL STORAGE
 
     HDFS-6677. Change INodeFile and FSImage to support storage policy ID.

hadoop-hdfs-project/hadoop-hdfs/src/main/java/org/apache/hadoop/hdfs/server/blockmanagement/BlockManager.java

@@ -3599,6 +3599,7 @@ public ReplicationWork(Block block,
       this.block = block;
       this.bc = bc;
       this.srcNode = srcNode;
+      this.srcNode.incrementPendingReplicationWithoutTargets();
       this.containingNodes = containingNodes;
       this.liveReplicaStorages = liveReplicaStorages;
       this.additionalReplRequired = additionalReplRequired;
@@ -3609,10 +3610,14 @@ public ReplicationWork(Block block,
     private void chooseTargets(BlockPlacementPolicy blockplacement,
         BlockStoragePolicySuite storagePolicySuite,
         Set<Node> excludedNodes) {
+      try {
         targets = blockplacement.chooseTarget(bc.getName(),
             additionalReplRequired, srcNode, liveReplicaStorages, false,
             excludedNodes, block.getNumBytes(),
             storagePolicySuite.getPolicy(bc.getStoragePolicyID()));
+      } finally {
+        srcNode.decrementPendingReplicationWithoutTargets();
+      }
     }
   }
 

hadoop-hdfs-project/hadoop-hdfs/src/main/java/org/apache/hadoop/hdfs/server/blockmanagement/DatanodeDescriptor.java

@@ -219,6 +219,9 @@ public CachedBlocksList getPendingUncached() {
    */
   private boolean disallowed = false;
 
+  // The number of replication work pending before targets are determined
+  private int PendingReplicationWithoutTargets = 0;
+
   /**
    * DatanodeDescriptor constructor
    * @param nodeID id of the data node
@@ -408,6 +411,14 @@ Iterator<BlockInfo> getBlockIterator(final String storageID) {
     return new BlockIterator(getStorageInfo(storageID));
   }
 
+  void incrementPendingReplicationWithoutTargets() {
+    PendingReplicationWithoutTargets++;
+  }
+
+  void decrementPendingReplicationWithoutTargets() {
+    PendingReplicationWithoutTargets--;
+  }
+
   /**
    * Store block replication work.
    */
@@ -444,7 +455,7 @@ void addBlocksToBeInvalidated(List<Block> blocklist) {
    * The number of work items that are pending to be replicated
    */
   int getNumberOfBlocksToBeReplicated() {
-    return replicateBlocks.size();
+    return PendingReplicationWithoutTargets + replicateBlocks.size();
   }
 
   /**

hadoop-hdfs-project/hadoop-hdfs/src/test/java/org/apache/hadoop/hdfs/server/blockmanagement/TestUnderReplicatedBlocks.java

@@ -18,11 +18,13 @@
 package org.apache.hadoop.hdfs.server.blockmanagement;
 
 import static org.junit.Assert.assertEquals;
+import static org.junit.Assert.assertTrue;
 
 import org.apache.hadoop.conf.Configuration;
 import org.apache.hadoop.fs.FileSystem;
 import org.apache.hadoop.fs.FsShell;
 import org.apache.hadoop.fs.Path;
+import org.apache.hadoop.hdfs.DFSConfigKeys;
 import org.apache.hadoop.hdfs.DFSTestUtil;
 import org.apache.hadoop.hdfs.HdfsConfiguration;
 import org.apache.hadoop.hdfs.MiniDFSCluster;
@@ -30,6 +32,9 @@
 import org.apache.hadoop.hdfs.server.datanode.DataNodeTestUtils;
 import org.junit.Test;
 
+import java.util.Iterator;
+
+
 public class TestUnderReplicatedBlocks {
   @Test(timeout=60000) // 1 min timeout
   public void testSetrepIncWithUnderReplicatedBlocks() throws Exception {
@@ -69,4 +74,76 @@ public void testSetrepIncWithUnderReplicatedBlocks() throws Exception {
     
   }
 
+  /**
+   * The test verifies the number of outstanding replication requests for a
+   * given DN shouldn't exceed the limit set by configuration property
+   * dfs.namenode.replication.max-streams-hard-limit.
+   * The test does the followings:
+   * 1. Create a mini cluster with 2 DNs. Set large heartbeat interval so that
+   *    replication requests won't be picked by any DN right away.
+   * 2. Create a file with 10 blocks and replication factor 2. Thus each
+   *    of the 2 DNs have one replica of each block.
+   * 3. Add a DN to the cluster for later replication.
+   * 4. Remove a DN that has data.
+   * 5. Ask BlockManager to compute the replication work. This will assign
+   *    replication requests to the only DN that has data.
+   * 6. Make sure the number of pending replication requests of that DN don't
+   *    exceed the limit.
+   * @throws Exception
+   */
+  @Test(timeout=60000) // 1 min timeout
+  public void testNumberOfBlocksToBeReplicated() throws Exception {
+    Configuration conf = new HdfsConfiguration();
+
+    conf.setLong(DFSConfigKeys.DFS_NAMENODE_MIN_BLOCK_SIZE_KEY, 0);
+    conf.setLong(DFSConfigKeys.DFS_BLOCK_SIZE_KEY, 1);
+    conf.setInt(DFSConfigKeys.DFS_BYTES_PER_CHECKSUM_KEY, 1);
+
+    // Large value to make sure the pending replication request can stay in
+    // DatanodeDescriptor.replicateBlocks before test timeout.
+    conf.setInt(DFSConfigKeys.DFS_HEARTBEAT_INTERVAL_KEY, 100);
+
+    // Make sure BlockManager can pull all blocks from UnderReplicatedBlocks via
+    // chooseUnderReplicatedBlocks at once.
+     conf.setInt(
+        DFSConfigKeys.DFS_NAMENODE_REPLICATION_WORK_MULTIPLIER_PER_ITERATION, 5);
+
+    int NUM_OF_BLOCKS = 10;
+    final short REP_FACTOR = 2;
+    final String FILE_NAME = "/testFile";
+    final Path FILE_PATH = new Path(FILE_NAME);
+    MiniDFSCluster cluster = new MiniDFSCluster.Builder(conf).numDataNodes(
+            REP_FACTOR).build();
+    try {
+      // create a file with 10 blocks with a replication factor of 2
+      final FileSystem fs = cluster.getFileSystem();
+      DFSTestUtil.createFile(fs, FILE_PATH, NUM_OF_BLOCKS, REP_FACTOR, 1L);
+      DFSTestUtil.waitReplication(fs, FILE_PATH, REP_FACTOR);
+
+      cluster.startDataNodes(conf, 1, true, null, null, null, null);
+
+      final BlockManager bm = cluster.getNamesystem().getBlockManager();
+      ExtendedBlock b = DFSTestUtil.getFirstBlock(fs, FILE_PATH);
+      Iterator<DatanodeStorageInfo> storageInfos =
+          bm.blocksMap.getStorages(b.getLocalBlock())
+          .iterator();
+      DatanodeDescriptor firstDn = storageInfos.next().getDatanodeDescriptor();
+      DatanodeDescriptor secondDn = storageInfos.next().getDatanodeDescriptor();
+
+      bm.getDatanodeManager().removeDatanode(firstDn);
+
+      assertEquals(NUM_OF_BLOCKS, bm.getUnderReplicatedNotMissingBlocks());
+      bm.computeDatanodeWork();
+
+
+      assertTrue("The number of blocks to be replicated should be less than "
+          + "or equal to " + bm.replicationStreamsHardLimit,
+          secondDn.getNumberOfBlocksToBeReplicated()
+          <= bm.replicationStreamsHardLimit);
+    } finally {
+      cluster.shutdown();
+    }
+
+  }
+
 }