HDFS-7208. NN doesn't schedule replication when a DN storage fails. Contributed by Ming Ma

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

@@ -959,6 +959,9 @@ Release 2.6.0 - UNRELEASED
     HDFS-7185. The active NameNode will not accept an fsimage sent from the
     standby during rolling upgrade. (jing9)
 
+    HDFS-7208. NN doesn't schedule replication when a DN storage fails.
+    (Ming Ma via szetszwo)
+
     BREAKDOWN OF HDFS-6134 AND HADOOP-10150 SUBTASKS AND RELATED JIRAS
   
       HDFS-6387. HDFS CLI admin tool for creating & deleting an

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

@@ -1062,6 +1062,19 @@ public class BlockManager {
     }
   }
 
+  /** Remove the blocks associated to the given DatanodeStorageInfo. */
+  void removeBlocksAssociatedTo(final DatanodeStorageInfo storageInfo) {
+    assert namesystem.hasWriteLock();
+    final Iterator<? extends Block> it = storageInfo.getBlockIterator();
+    DatanodeDescriptor node = storageInfo.getDatanodeDescriptor();
+    while(it.hasNext()) {
+      Block block = it.next();
+      removeStoredBlock(block, node);
+      invalidateBlocks.remove(node, block);
+    }
+    namesystem.checkSafeMode();
+  }
+
   /**
    * Adds block to list of blocks which will be invalidated on specified
    * datanode and log the operation

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

@@ -722,6 +722,12 @@ public class BlockPlacementPolicyDefault extends BlockPlacementPolicy {
       logNodeIsNotChosen(storage, "storage is read-only");
       return false;
     }
+
+    if (storage.getState() == State.FAILED) {
+      logNodeIsNotChosen(storage, "storage has failed");
+      return false;
+    }
+
     DatanodeDescriptor node = storage.getDatanodeDescriptor();
     // check if the node is (being) decommissioned
     if (node.isDecommissionInProgress() || node.isDecommissioned()) {

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

@@ -21,11 +21,13 @@ import java.util.ArrayList;
 import java.util.Collection;
 import java.util.Collections;
 import java.util.HashMap;
+import java.util.HashSet;
 import java.util.Iterator;
 import java.util.LinkedList;
 import java.util.List;
 import java.util.Map;
 import java.util.Queue;
+import java.util.Set;
 
 import com.google.common.annotations.VisibleForTesting;
 
@@ -222,13 +224,16 @@ public class DatanodeDescriptor extends DatanodeInfo {
   // The number of replication work pending before targets are determined
   private int PendingReplicationWithoutTargets = 0;
 
+  // HB processing can use it to tell if it is the first HB since DN restarted
+  private boolean heartbeatedSinceRegistration = false;
+
   /**
    * DatanodeDescriptor constructor
    * @param nodeID id of the data node
    */
   public DatanodeDescriptor(DatanodeID nodeID) {
     super(nodeID);
-    updateHeartbeat(StorageReport.EMPTY_ARRAY, 0L, 0L, 0, 0);
+    updateHeartbeatState(StorageReport.EMPTY_ARRAY, 0L, 0L, 0, 0);
   }
 
   /**
@@ -239,7 +244,7 @@ public class DatanodeDescriptor extends DatanodeInfo {
   public DatanodeDescriptor(DatanodeID nodeID, 
                             String networkLocation) {
     super(nodeID, networkLocation);
-    updateHeartbeat(StorageReport.EMPTY_ARRAY, 0L, 0L, 0, 0);
+    updateHeartbeatState(StorageReport.EMPTY_ARRAY, 0L, 0L, 0, 0);
   }
 
   @VisibleForTesting
@@ -341,10 +346,48 @@ public class DatanodeDescriptor extends DatanodeInfo {
    */
   public void updateHeartbeat(StorageReport[] reports, long cacheCapacity,
       long cacheUsed, int xceiverCount, int volFailures) {
+    updateHeartbeatState(reports, cacheCapacity, cacheUsed, xceiverCount,
+        volFailures);
+    heartbeatedSinceRegistration = true;
+  }
+
+  /**
+   * process datanode heartbeat or stats initialization.
+   */
+  public void updateHeartbeatState(StorageReport[] reports, long cacheCapacity,
+      long cacheUsed, int xceiverCount, int volFailures) {
     long totalCapacity = 0;
     long totalRemaining = 0;
     long totalBlockPoolUsed = 0;
     long totalDfsUsed = 0;
+    Set<DatanodeStorageInfo> failedStorageInfos = null;
+
+    // Decide if we should check for any missing StorageReport and mark it as
+    // failed. There are different scenarios.
+    // 1. When DN is running, a storage failed. Given the current DN
+    //    implementation doesn't add recovered storage back to its storage list
+    //    until DN restart, we can assume volFailures won't decrease
+    //    during the current DN registration session.
+    //    When volumeFailures == this.volumeFailures, it implies there is no
+    //    state change. No need to check for failed storage. This is an
+    //    optimization.
+    // 2. After DN restarts, volFailures might not increase and it is possible
+    //    we still have new failed storage. For example, admins reduce
+    //    available storages in configuration. Another corner case
+    //    is the failed volumes might change after restart; a) there
+    //    is one good storage A, one restored good storage B, so there is
+    //    one element in storageReports and that is A. b) A failed. c) Before
+    //    DN sends HB to NN to indicate A has failed, DN restarts. d) After DN
+    //    restarts, storageReports has one element which is B.
+    boolean checkFailedStorages = (volFailures > this.volumeFailures) ||
+        !heartbeatedSinceRegistration;
+
+    if (checkFailedStorages) {
+      LOG.info("Number of failed storage changes from "
+          + this.volumeFailures + " to " + volFailures);
+      failedStorageInfos = new HashSet<DatanodeStorageInfo>(
+          storageMap.values());
+    }
 
     setCacheCapacity(cacheCapacity);
     setCacheUsed(cacheUsed);
@@ -353,6 +396,10 @@ public class DatanodeDescriptor extends DatanodeInfo {
     this.volumeFailures = volFailures;
     for (StorageReport report : reports) {
       DatanodeStorageInfo storage = updateStorage(report.getStorage());
+      if (checkFailedStorages) {
+        failedStorageInfos.remove(storage);
+      }
+
       storage.receivedHeartbeat(report);
       totalCapacity += report.getCapacity();
       totalRemaining += report.getRemaining();
@@ -366,6 +413,19 @@ public class DatanodeDescriptor extends DatanodeInfo {
     setRemaining(totalRemaining);
     setBlockPoolUsed(totalBlockPoolUsed);
     setDfsUsed(totalDfsUsed);
+    if (checkFailedStorages) {
+      updateFailedStorage(failedStorageInfos);
+    }
+  }
+
+  private void updateFailedStorage(
+      Set<DatanodeStorageInfo> failedStorageInfos) {
+    for (DatanodeStorageInfo storageInfo : failedStorageInfos) {
+      if (storageInfo.getState() != DatanodeStorage.State.FAILED) {
+        LOG.info(storageInfo + " failed.");
+        storageInfo.setState(DatanodeStorage.State.FAILED);
+      }
+    }
   }
 
   private static class BlockIterator implements Iterator<BlockInfo> {
@@ -639,6 +699,7 @@ public class DatanodeDescriptor extends DatanodeInfo {
     for(DatanodeStorageInfo storage : getStorageInfos()) {
       storage.setBlockReportCount(0);
     }
+    heartbeatedSinceRegistration = false;
   }
 
   /**

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

@@ -183,6 +183,14 @@ public class DatanodeStorageInfo {
     return this.state;
   }
 
+  void setState(State state) {
+    this.state = state;
+  }
+
+  boolean areBlocksOnFailedStorage() {
+    return getState() == State.FAILED && numBlocks != 0;
+  }
+
   String getStorageID() {
     return storageID;
   }

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

@@ -192,7 +192,7 @@ class HeartbeatManager implements DatanodeStatistics {
       addDatanode(d);
 
       //update its timestamp
-      d.updateHeartbeat(StorageReport.EMPTY_ARRAY, 0L, 0L, 0, 0);
+      d.updateHeartbeatState(StorageReport.EMPTY_ARRAY, 0L, 0L, 0, 0);
     }
   }
 
@@ -242,6 +242,25 @@ class HeartbeatManager implements DatanodeStatistics {
    * While removing dead datanodes, make sure that only one datanode is marked
    * dead at a time within the synchronized section. Otherwise, a cascading
    * effect causes more datanodes to be declared dead.
+   * Check if there are any failed storage and if so,
+   * Remove all the blocks on the storage. It also covers the following less
+   * common scenarios. After DatanodeStorage is marked FAILED, it is still
+   * possible to receive IBR for this storage.
+   * 1) DN could deliver IBR for failed storage due to its implementation.
+   *    a) DN queues a pending IBR request.
+   *    b) The storage of the block fails.
+   *    c) DN first sends HB, NN will mark the storage FAILED.
+   *    d) DN then sends the pending IBR request.
+   * 2) SBN processes block request from pendingDNMessages.
+   *    It is possible to have messages in pendingDNMessages that refer
+   *    to some failed storage.
+   *    a) SBN receives a IBR and put it in pendingDNMessages.
+   *    b) The storage of the block fails.
+   *    c) Edit log replay get the IBR from pendingDNMessages.
+   * Alternatively, we can resolve these scenarios with the following approaches.
+   * A. Make sure DN don't deliver IBR for failed storage.
+   * B. Remove all blocks in PendingDataNodeMessages for the failed storage
+   *    when we remove all blocks from BlocksMap for that storage.
    */
   void heartbeatCheck() {
     final DatanodeManager dm = blockManager.getDatanodeManager();
@@ -254,6 +273,10 @@ class HeartbeatManager implements DatanodeStatistics {
     while (!allAlive) {
       // locate the first dead node.
       DatanodeID dead = null;
+
+      // locate the first failed storage that isn't on a dead node.
+      DatanodeStorageInfo failedStorage = null;
+
       // check the number of stale nodes
       int numOfStaleNodes = 0;
       int numOfStaleStorages = 0;
@@ -271,7 +294,14 @@ class HeartbeatManager implements DatanodeStatistics {
             if (storageInfo.areBlockContentsStale()) {
               numOfStaleStorages++;
             }
+
+            if (failedStorage == null &&
+                storageInfo.areBlocksOnFailedStorage() &&
+                d != dead) {
+              failedStorage = storageInfo;
+            }
           }
+
         }
         
         // Set the number of stale nodes in the DatanodeManager
@@ -279,8 +309,8 @@ class HeartbeatManager implements DatanodeStatistics {
         dm.setNumStaleStorages(numOfStaleStorages);
       }
 
-      allAlive = dead == null;
+      allAlive = dead == null && failedStorage == null;
-      if (!allAlive) {
+      if (dead != null) {
         // acquire the fsnamesystem lock, and then remove the dead node.
         namesystem.writeLock();
         try {
@@ -294,6 +324,20 @@ class HeartbeatManager implements DatanodeStatistics {
           namesystem.writeUnlock();
         }
       }
+      if (failedStorage != null) {
+        // acquire the fsnamesystem lock, and remove blocks on the storage.
+        namesystem.writeLock();
+        try {
+          if (namesystem.isInStartupSafeMode()) {
+            return;
+          }
+          synchronized(this) {
+            blockManager.removeBlocksAssociatedTo(failedStorage);
+          }
+        } finally {
+          namesystem.writeUnlock();
+        }
+      }
     }
   }
 

hadoop-hdfs-project/hadoop-hdfs/src/main/java/org/apache/hadoop/hdfs/server/protocol/DatanodeStorage.java

@@ -39,7 +39,9 @@ public class DatanodeStorage {
      * property should be used for debugging purposes only.
      * </p> 
      */
-    READ_ONLY_SHARED;
+    READ_ONLY_SHARED,
+
+    FAILED;
   }
   
   private final String storageID;

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

@@ -214,6 +214,25 @@ public class BlockManagerTestUtil {
     bm.getDatanodeManager().getHeartbeatManager().heartbeatCheck();
   }
 
+  /**
+   * Call heartbeat check function of HeartbeatManager and get
+   * under replicated blocks count within write lock to make sure
+   * computeDatanodeWork doesn't interfere.
+   * @param namesystem the FSNamesystem
+   * @param bm the BlockManager to manipulate
+   * @return the number of under replicated blocks
+   */
+  public static int checkHeartbeatAndGetUnderReplicatedBlocksCount(
+      FSNamesystem namesystem, BlockManager bm) {
+    namesystem.writeLock();
+    try {
+      bm.getDatanodeManager().getHeartbeatManager().heartbeatCheck();
+      return bm.getUnderReplicatedNotMissingBlocks();
+    } finally {
+      namesystem.writeUnlock();
+    }
+  }
+
   public static DatanodeStorageInfo updateStorage(DatanodeDescriptor dn,
       DatanodeStorage s) {
     return dn.updateStorage(s);

hadoop-hdfs-project/hadoop-hdfs/src/test/java/org/apache/hadoop/hdfs/server/datanode/TestDataNodeVolumeFailure.java

@@ -19,6 +19,8 @@ package org.apache.hadoop.hdfs.server.datanode;
 
 import static org.junit.Assert.assertEquals;
 import static org.junit.Assert.assertNotNull;
+import static org.junit.Assert.assertTrue;
+import static org.junit.Assume.assumeTrue;
 
 import java.io.File;
 import java.io.FilenameFilter;
@@ -52,6 +54,7 @@ import org.apache.hadoop.hdfs.protocol.DatanodeInfo;
 import org.apache.hadoop.hdfs.protocol.ExtendedBlock;
 import org.apache.hadoop.hdfs.protocol.LocatedBlock;
 import org.apache.hadoop.hdfs.security.token.block.BlockTokenIdentifier;
+import org.apache.hadoop.hdfs.server.blockmanagement.BlockManager;
 import org.apache.hadoop.hdfs.server.blockmanagement.BlockManagerTestUtil;
 import org.apache.hadoop.hdfs.server.common.HdfsServerConstants;
 import org.apache.hadoop.hdfs.server.namenode.FSNamesystem;
@@ -79,6 +82,7 @@ public class TestDataNodeVolumeFailure {
   File dataDir = null;
   File data_fail = null;
   File failedDir = null;
+  private FileSystem fs;
 
   // mapping blocks to Meta files(physical files) and locs(NameNode locations)
   private class BlockLocs {
@@ -97,6 +101,8 @@ public class TestDataNodeVolumeFailure {
     conf.setInt(DFSConfigKeys.DFS_DATANODE_FAILED_VOLUMES_TOLERATED_KEY, 1);
     cluster = new MiniDFSCluster.Builder(conf).numDataNodes(dn_num).build();
     cluster.waitActive();
+    fs = cluster.getFileSystem();
+    dataDir = new File(cluster.getDataDirectory());
   }
 
   @After
@@ -110,6 +116,10 @@ public class TestDataNodeVolumeFailure {
     if(cluster != null) {
       cluster.shutdown();
     }
+    for (int i = 0; i < 3; i++) {
+      FileUtil.setExecutable(new File(dataDir, "data"+(2*i+1)), true);
+      FileUtil.setExecutable(new File(dataDir, "data"+(2*i+2)), true);
+    }
   }
   
   /*
@@ -119,8 +129,6 @@ public class TestDataNodeVolumeFailure {
    */
   @Test
   public void testVolumeFailure() throws Exception {
-    FileSystem fs = cluster.getFileSystem();
-    dataDir = new File(cluster.getDataDirectory());
     System.out.println("Data dir: is " +  dataDir.getPath());
    
     
@@ -192,6 +200,39 @@ public class TestDataNodeVolumeFailure {
         " is created and replicated");
   }
 
+  /**
+   * Test that there are under replication blocks after vol failures
+   */
+  @Test
+  public void testUnderReplicationAfterVolFailure() throws Exception {
+    // Bring up one more datanode
+    cluster.startDataNodes(conf, 1, true, null, null);
+    cluster.waitActive();
+
+    final BlockManager bm = cluster.getNamesystem().getBlockManager();
+
+    Path file1 = new Path("/test1");
+    DFSTestUtil.createFile(fs, file1, 1024, (short)3, 1L);
+    DFSTestUtil.waitReplication(fs, file1, (short)3);
+
+    // Fail the first volume on both datanodes
+    File dn1Vol1 = new File(dataDir, "data"+(2*0+1));
+    File dn2Vol1 = new File(dataDir, "data"+(2*1+1));
+    assertTrue("Couldn't chmod local vol", FileUtil.setExecutable(dn1Vol1, false));
+    assertTrue("Couldn't chmod local vol", FileUtil.setExecutable(dn2Vol1, false));
+
+    Path file2 = new Path("/test2");
+    DFSTestUtil.createFile(fs, file2, 1024, (short)3, 1L);
+    DFSTestUtil.waitReplication(fs, file2, (short)3);
+
+    // underReplicatedBlocks are due to failed volumes
+    int underReplicatedBlocks =
+        BlockManagerTestUtil.checkHeartbeatAndGetUnderReplicatedBlocksCount(
+            cluster.getNamesystem(), bm);
+    assertTrue("There is no under replicated block after volume failure",
+        underReplicatedBlocks > 0);
+  }
+
   /**
    * verifies two things:
    *  1. number of locations of each block in the name node