HDFS-7409. Allow dead nodes to finish decommissioning if all files are fully replicated.

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

@@ -375,6 +375,9 @@ Release 2.7.0 - UNRELEASED
     HDFS-7398. Reset cached thread-local FSEditLogOp's on every
     FSEditLog#logEdit. (Gera Shegalov via cnauroth)
 
+    HDFS-7409. Allow dead nodes to finish decommissioning if all files are
+    fully replicated. (wang)
+
   OPTIMIZATIONS
 
   BUG FIXES

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

@@ -3272,11 +3272,19 @@ public class BlockManager {
     }
 
     if (!status && !srcNode.isAlive) {
-      LOG.warn("srcNode " + srcNode + " is dead " +
-          "when decommission is in progress. Continue to mark " +
-          "it as decommission in progress. In that way, when it rejoins the " +
-          "cluster it can continue the decommission process.");
-      status = true;
+      updateState();
+      if (pendingReplicationBlocksCount == 0 &&
+          underReplicatedBlocksCount == 0) {
+        LOG.info("srcNode {} is dead and there are no under-replicated" +
+            " blocks or blocks pending replication. Marking as " +
+            "decommissioned.");
+      } else {
+        LOG.warn("srcNode " + srcNode + " is dead " +
+            "while decommission is in progress. Continuing to mark " +
+            "it as decommission in progress so when it rejoins the " +
+            "cluster it can continue the decommission process.");
+        status = true;
+      }
     }
 
     srcNode.decommissioningStatus.set(underReplicatedBlocks,

hadoop-hdfs-project/hadoop-hdfs/src/test/java/org/apache/hadoop/hdfs/server/namenode/TestDecommissioningStatus.java

@@ -18,6 +18,7 @@
 package org.apache.hadoop.hdfs.server.namenode;
 
 import static org.junit.Assert.assertEquals;
+import static org.junit.Assert.assertFalse;
 import static org.junit.Assert.assertTrue;
 
 import java.io.IOException;
@@ -347,8 +348,15 @@ public class TestDecommissioningStatus {
     BlockManagerTestUtil.checkDecommissionState(dm, dead.get(0));
 
     // Verify that the DN remains in DECOMMISSION_INPROGRESS state.
-    assertTrue("the node is in decommissioned state ",
-        !dead.get(0).isDecommissioned());
+    assertTrue("the node should be DECOMMISSION_IN_PROGRESSS",
+        dead.get(0).isDecommissionInProgress());
+
+    // Delete the under-replicated file, which should let the 
+    // DECOMMISSION_IN_PROGRESS node become DECOMMISSIONED
+    cleanupFile(fileSys, f);
+    BlockManagerTestUtil.checkDecommissionState(dm, dead.get(0));
+    assertTrue("the node should be decommissioned",
+        dead.get(0).isDecommissioned());
 
     // Add the node back
     cluster.restartDataNode(dataNodeProperties, true);
@@ -359,7 +367,6 @@ public class TestDecommissioningStatus {
     // make them available again.
     writeConfigFile(localFileSys, excludeFile, null);
     dm.refreshNodes(conf);
-    cleanupFile(fileSys, f);
   }
 
   /**