HBASE-11703 Meta region state could be corrupted

hbase-server/src/main/java/org/apache/hadoop/hbase/master/RegionStates.java

@@ -508,7 +508,7 @@ public class RegionStates {
       ServerName oldServerName = regionAssignments.remove(hri);
       if (oldServerName != null && serverHoldings.containsKey(oldServerName)
           && (newState == State.MERGED || newState == State.SPLIT
-            || tableStateManager.isTableState(hri.getTable(),
+            || hri.isMetaRegion() || tableStateManager.isTableState(hri.getTable(),
               ZooKeeperProtos.Table.State.DISABLED, ZooKeeperProtos.Table.State.DISABLING))) {
         // Offline the region only if it's merged/split, or the table is disabled/disabling.
         // Otherwise, offline it from this server only when it is online on a different server.

hbase-server/src/main/java/org/apache/hadoop/hbase/master/handler/ServerShutdownHandler.java

@@ -156,13 +156,7 @@ public class ServerShutdownHandler extends EventHandler {
       while (!this.server.isStopped()) {
         try {
           server.getMetaTableLocator().waitMetaRegionLocation(server.getZooKeeper());
-          // Skip getting user regions if the server is stopped.
-          if (!this.server.isStopped()) {
           hris = am.getRegionStates().getServerRegions(serverName);
-            if (hris != null) {
-              hris.remove(HRegionInfo.FIRST_META_REGIONINFO);
-            }
-          }
           break;
         } catch (InterruptedException e) {
           Thread.currentThread().interrupt();

hbase-server/src/test/java/org/apache/hadoop/hbase/master/TestAssignmentManagerOnCluster.java

@@ -19,6 +19,7 @@ package org.apache.hadoop.hbase.master;
 
 import static org.junit.Assert.assertEquals;
 import static org.junit.Assert.assertFalse;
+import static org.junit.Assert.assertNotEquals;
 import static org.junit.Assert.assertNotNull;
 import static org.junit.Assert.assertNull;
 import static org.junit.Assert.assertTrue;
@@ -100,6 +101,51 @@ public class TestAssignmentManagerOnCluster {
     TEST_UTIL.shutdownMiniCluster();
   }
 
+  /**
+   * This tests restarting meta regionserver
+   */
+  @Test (timeout=180000)
+  public void testRestartMetaRegionServer() throws Exception {
+    MiniHBaseCluster cluster = TEST_UTIL.getHBaseCluster();
+    boolean stoppedARegionServer = false;
+    try {
+      HMaster master = cluster.getMaster();
+      RegionStates regionStates = master.getAssignmentManager().getRegionStates();
+      ServerName metaServerName = regionStates.getRegionServerOfRegion(
+        HRegionInfo.FIRST_META_REGIONINFO);
+      if (master.getServerName().equals(metaServerName)) {
+        // Move meta off master
+        metaServerName = cluster.getLiveRegionServerThreads()
+          .get(0).getRegionServer().getServerName();
+        master.move(HRegionInfo.FIRST_META_REGIONINFO.getEncodedNameAsBytes(),
+          Bytes.toBytes(metaServerName.getServerName()));
+        TEST_UTIL.waitUntilNoRegionsInTransition(60000);
+      }
+      assertNotEquals("Meta should be moved off master",
+        metaServerName, master.getServerName());
+      cluster.killRegionServer(metaServerName);
+      stoppedARegionServer = true;
+      cluster.waitForRegionServerToStop(metaServerName, 60000);
+
+      // Wait for SSH to finish
+      final ServerManager serverManager = master.getServerManager();
+      TEST_UTIL.waitFor(120000, 200, new Waiter.Predicate<Exception>() {
+        @Override
+        public boolean evaluate() throws Exception {
+          return !serverManager.areDeadServersInProgress();
+        }
+      });
+
+      // Now, make sure meta is assigned
+      assertTrue("Meta should be assigned",
+        regionStates.isRegionOnline(HRegionInfo.FIRST_META_REGIONINFO));
+    } finally {
+      if (stoppedARegionServer) {
+        cluster.startRegionServer();
+      }
+    }
+  }
+
   /**
    * This tests region assignment
    */