Java Code Examples for org.apache.hadoop.hbase.master.HMaster#abort()

/**
 * Cause a master to exit without shutting down entire cluster.
 * @param serverNumber  Used as index into a list.
 */
public String abortMaster(int serverNumber) {
  HMaster server = getMaster(serverNumber);
  LOG.info("Aborting " + server.toString());
  server.abort("Aborting for tests", new Exception("Trace info"));
  return server.toString();
}
 

/**
 * Master recovery when the znode already exists. Internally, this
 *  test differs from {@link #testMasterSessionExpired} because here
 *  the master znode will exist in ZK.
 */
@Test
public void testMasterZKSessionRecoveryFailure() throws Exception {
  LOG.info("Starting " + name.getMethodName());
  MiniHBaseCluster cluster = TEST_UTIL.getHBaseCluster();
  HMaster m = cluster.getMaster();
  m.abort("Test recovery from zk session expired",
      new KeeperException.SessionExpiredException());
  assertTrue(m.isStopped()); // Master doesn't recover any more
  testSanity(name.getMethodName());
}
 

/**
 * Tests that the master does not call retainAssignment after recovery from expired zookeeper
 * session. Without the HBASE-6046 fix master always tries to assign all the user regions by
 * calling retainAssignment.
 */
@Test
public void testRegionAssignmentAfterMasterRecoveryDueToZKExpiry() throws Exception {
  MiniHBaseCluster cluster = TEST_UTIL.getHBaseCluster();
  cluster.startRegionServer();
  cluster.waitForActiveAndReadyMaster(10000);
  HMaster m = cluster.getMaster();
  final ZKWatcher zkw = m.getZooKeeper();
  // now the cluster is up. So assign some regions.
  try (Admin admin = TEST_UTIL.getAdmin()) {
    byte[][] SPLIT_KEYS = new byte[][] { Bytes.toBytes("a"), Bytes.toBytes("b"),
        Bytes.toBytes("c"), Bytes.toBytes("d"), Bytes.toBytes("e"), Bytes.toBytes("f"),
        Bytes.toBytes("g"), Bytes.toBytes("h"), Bytes.toBytes("i"), Bytes.toBytes("j") };
    TableDescriptor htd =
        TableDescriptorBuilder.newBuilder(TableName.valueOf(name.getMethodName()))
            .setColumnFamily(ColumnFamilyDescriptorBuilder.of(HConstants.CATALOG_FAMILY)).build();
    admin.createTable(htd, SPLIT_KEYS);
    TEST_UTIL.waitUntilNoRegionsInTransition(60000);
    m.getZooKeeper().close();
    MockLoadBalancer.retainAssignCalled = false;
    final int expectedNumOfListeners = countPermanentListeners(zkw);
    m.abort("Test recovery from zk session expired",
        new KeeperException.SessionExpiredException());
    assertTrue(m.isStopped()); // Master doesn't recover any more
    // The recovered master should not call retainAssignment, as it is not a
    // clean startup.
    assertFalse("Retain assignment should not be called", MockLoadBalancer.retainAssignCalled);
    // number of listeners should be same as the value before master aborted
    // wait for new master is initialized
    cluster.waitForActiveAndReadyMaster(120000);
    final HMaster newMaster = cluster.getMasterThread().getMaster();
    assertEquals(expectedNumOfListeners, countPermanentListeners(newMaster.getZooKeeper()));
  }
}
 

/**
 * Tests whether the logs are split when master recovers from a expired zookeeper session and an
 * RS goes down.
 */
@Test
public void testLogSplittingAfterMasterRecoveryDueToZKExpiry() throws Exception {
  MiniHBaseCluster cluster = TEST_UTIL.getHBaseCluster();
  cluster.startRegionServer();
  TableName tableName = TableName.valueOf(name.getMethodName());
  byte[] family = Bytes.toBytes("col");
  try (Admin admin = TEST_UTIL.getAdmin()) {
    byte[][] SPLIT_KEYS = new byte[][] { Bytes.toBytes("1"), Bytes.toBytes("2"),
      Bytes.toBytes("3"), Bytes.toBytes("4"), Bytes.toBytes("5") };
    TableDescriptor htd = TableDescriptorBuilder.newBuilder(tableName)
        .setColumnFamily(ColumnFamilyDescriptorBuilder.of(family)).build();
    admin.createTable(htd, SPLIT_KEYS);
  }
  TEST_UTIL.waitUntilNoRegionsInTransition(60000);
  HMaster m = cluster.getMaster();
  try (Table table = TEST_UTIL.getConnection().getTable(tableName)) {
    int numberOfPuts;
    for (numberOfPuts = 0; numberOfPuts < 6; numberOfPuts++) {
      Put p = new Put(Bytes.toBytes(numberOfPuts));
      p.addColumn(Bytes.toBytes("col"), Bytes.toBytes("ql"),
        Bytes.toBytes("value" + numberOfPuts));
      table.put(p);
    }
    m.abort("Test recovery from zk session expired",
      new KeeperException.SessionExpiredException());
    assertTrue(m.isStopped()); // Master doesn't recover any more
    cluster.killRegionServer(TEST_UTIL.getRSForFirstRegionInTable(tableName).getServerName());
    // Without patch for HBASE-6046 this test case will always timeout
    // with patch the test case should pass.
    int numberOfRows = 0;
    try (ResultScanner scanner = table.getScanner(new Scan())) {
      while (scanner.next() != null) {
        numberOfRows++;
      }
    }
    assertEquals("Number of rows should be equal to number of puts.", numberOfPuts, numberOfRows);
  }
}
 

@Test
public void test() throws Exception {
  RegionInfo region = UTIL.getMiniHBaseCluster().getRegions(NAME).get(0).getRegionInfo();
  AssignmentManager am = UTIL.getMiniHBaseCluster().getMaster().getAssignmentManager();
  RegionStateNode rsn = am.getRegionStates().getRegionStateNode(region);

  ServerName sn = rsn.getRegionLocation();
  ARRIVE = new CountDownLatch(1);
  HALT = true;
  am.moveAsync(new RegionPlan(region, sn, sn));
  ARRIVE.await();

  // let's restart the master
  EXCLUDE_SERVERS.add(rsn.getRegionLocation());
  KILL = true;
  HMaster activeMaster = UTIL.getMiniHBaseCluster().getMaster();
  activeMaster.abort("For testing");
  activeMaster.join();
  KILL = false;

  // sleep a while to reproduce the problem, as after the fix in HBASE-21472 the execution logic
  // is changed so the old code to reproduce the problem can not compile...
  Thread.sleep(10000);
  HALT = false;
  Thread.sleep(5000);

  HRegionServer rs = UTIL.getMiniHBaseCluster().getRegionServer(sn);
  assertNotNull(rs.getRegion(region.getEncodedName()));
  assertNull(UTIL.getOtherRegionServer(rs).getRegion(region.getEncodedName()));
}