HBASE-24286: HMaster won't become healthy after after cloning or crea…

hbase-common/src/main/java/org/apache/hadoop/hbase/HConstants.java

@@ -1620,6 +1620,17 @@ public enum OperationStatusCode {
    */
   public static final int BATCH_ROWS_THRESHOLD_DEFAULT = 5000;
 
+  /**
+   * when zookeeper data does not exist on master during meta bootstrap, default to remove the
+   * meta table directory that is considered as partial meta.
+   *
+   * TODO we can remove this feature if we come up a way to define partial meta on during bootstrap
+   *      cluster that does not come with Zookeeper data
+   */
+  public static final String REMOVE_META_ON_RESTART = "hbase.master.remove.meta.on.restart";
+  public static final boolean DEFAULT_REMOVE_META_ON_RESTART = true;
+
+
   private HConstants() {
     // Can't be instantiated with this ctor.
   }

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

@@ -1078,6 +1078,7 @@ private void finishActiveMasterInitialization(MonitoredTask status)
     // complete before we do this next step processing offline regions else it fails reading
     // table states messing up master launch (namespace table, etc., are not assigned).
     this.assignmentManager.processOfflineRegions();
+    this.assignmentManager.processRegionsOnUnknownServers();
     // Initialize after meta is up as below scans meta
     if (favoredNodesManager != null && !maintenanceMode) {
       SnapshotOfRegionAssignmentFromMeta snapshotOfRegionAssignment =

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

@@ -1441,6 +1441,35 @@ public void processOfflineRegions() {
     }
   }
 
+  /**
+   * Create assign procedure for non-offline regions of enabled table that are assigned
+   * to `unknown` servers after hbase:meta is online.
+   *
+   * This is a special case when WAL directory, SCP WALs and ZK data are cleared,
+   * cluster restarts with hbase:meta table and other tables with storefiles.
+   */
+  public void processRegionsOnUnknownServers() {
+    List<RegionInfo> regionsOnUnknownServers = regionStates.getRegionStates().stream()
+      .filter(s -> !s.isOffline())
+      .filter(s -> isTableEnabled(s.getRegion().getTable()))
+      .filter(s -> !regionStates.isRegionInTransition(s.getRegion()))
+      .filter(s -> {
+        ServerName serverName = regionStates.getRegionServerOfRegion(s.getRegion());
+        if (serverName == null) {
+          return false;
+        }
+        return master.getServerManager().isServerKnownAndOnline(serverName)
+          .equals(ServerManager.ServerLiveState.UNKNOWN);
+      })
+      .map(RegionState::getRegion).collect(Collectors.toList());
+    if (!regionsOnUnknownServers.isEmpty()) {
+      LOG.info("Found regions {} on unknown servers, reassign them to online servers",
+        regionsOnUnknownServers);
+      master.getMasterProcedureExecutor().submitProcedures(
+        master.getAssignmentManager().createRoundRobinAssignProcedures(regionsOnUnknownServers));
+    }
+  }
+
   /* AM internal RegionStateStore.RegionStateVisitor implementation. To be used when
    * scanning META table for region rows, using RegionStateStore utility methods. RegionStateStore
    * methods will convert Result into proper RegionInfo instances, but those would still need to be

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

@@ -71,7 +71,11 @@ private static void writeFsLayout(Path rootDir, Configuration conf) throws IOExc
     LOG.info("BOOTSTRAP: creating hbase:meta region");
     FileSystem fs = rootDir.getFileSystem(conf);
     Path tableDir = CommonFSUtils.getTableDir(rootDir, TableName.META_TABLE_NAME);
-    if (fs.exists(tableDir) && !fs.delete(tableDir, true)) {
+    boolean removeMeta = conf.getBoolean(HConstants.REMOVE_META_ON_RESTART,
+      HConstants.DEFAULT_REMOVE_META_ON_RESTART);
+    // we use zookeeper data to tell if this is a partial created meta, if so we should delete
+    // and recreate the meta table.
+    if (removeMeta && fs.exists(tableDir) && !fs.delete(tableDir, true)) {
       LOG.warn("Can not delete partial created meta table, continue...");
     }
     // Bootstrapping, make sure blockcache is off. Else, one will be

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

@@ -0,0 +1,226 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one
+ * or more contributor license agreements.  See the NOTICE file
+ * distributed with this work for additional information
+ * regarding copyright ownership.  The ASF licenses this file
+ * to you under the Apache License, Version 2.0 (the
+ * "License"); you may not use this file except in compliance
+ * with the License.  You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+package org.apache.hadoop.hbase.master;
+
+import static org.junit.Assert.assertEquals;
+import static org.junit.Assert.assertFalse;
+import static org.junit.Assert.assertTrue;
+
+import java.io.IOException;
+import java.time.Duration;
+import java.util.List;
+
+import org.apache.hadoop.fs.Path;
+import org.apache.hadoop.hbase.Cell;
+import org.apache.hadoop.hbase.HBaseClassTestRule;
+import org.apache.hadoop.hbase.HBaseTestingUtility;
+import org.apache.hadoop.hbase.HConstants;
+import org.apache.hadoop.hbase.MiniHBaseCluster;
+import org.apache.hadoop.hbase.ServerName;
+import org.apache.hadoop.hbase.TableName;
+import org.apache.hadoop.hbase.client.Get;
+import org.apache.hadoop.hbase.client.Put;
+import org.apache.hadoop.hbase.client.RegionInfo;
+import org.apache.hadoop.hbase.client.Result;
+import org.apache.hadoop.hbase.client.Table;
+import org.apache.hadoop.hbase.master.region.MasterRegionFactory;
+import org.apache.hadoop.hbase.procedure2.store.wal.WALProcedureStore;
+import org.apache.hadoop.hbase.regionserver.HRegionServer;
+import org.apache.hadoop.hbase.testclassification.LargeTests;
+import org.apache.hadoop.hbase.util.Bytes;
+import org.apache.hadoop.hbase.util.CommonFSUtils;
+import org.apache.hadoop.hbase.zookeeper.ZKUtil;
+
+import org.junit.Before;
+import org.junit.ClassRule;
+import org.junit.Rule;
+import org.junit.Test;
+import org.junit.experimental.categories.Category;
+import org.junit.rules.TestName;
+
+/**
+ * Test reuse storefiles within data directory when cluster failover with a set of new region
+ * servers with different hostnames with or without WALs and Zookeeper ZNodes support. For any
+ * hbase system table and user table can be assigned normally after cluster restart.
+ */
+@Category({ LargeTests.class })
+public class TestRecreateCluster {
+  @ClassRule
+  public static final HBaseClassTestRule CLASS_RULE =
+      HBaseClassTestRule.forClass(TestRecreateCluster.class);
+
+  @Rule
+  public TestName name = new TestName();
+
+  private static final HBaseTestingUtility TEST_UTIL = new HBaseTestingUtility();
+  private static final int NUM_RS = 3;
+  private static final long TIMEOUT_MS = Duration.ofMinutes(2).toMillis();
+
+  @Before
+  public void setup() {
+    TEST_UTIL.getConfiguration().setBoolean(HConstants.REMOVE_META_ON_RESTART, false);
+  }
+
+  @Test
+  public void testRecreateCluster_UserTableDisabled() throws Exception {
+    TEST_UTIL.startMiniCluster(NUM_RS);
+    try {
+      TableName tableName = TableName.valueOf("t1");
+      prepareDataBeforeRecreate(TEST_UTIL, tableName);
+      TEST_UTIL.getAdmin().disableTable(tableName);
+      TEST_UTIL.waitTableDisabled(tableName.getName());
+      restartHBaseCluster(true);
+      TEST_UTIL.getAdmin().enableTable(tableName);
+      validateDataAfterRecreate(TEST_UTIL, tableName);
+    } finally {
+      TEST_UTIL.shutdownMiniCluster();
+    }
+  }
+
+  @Test
+  public void testRecreateCluster_UserTableEnabled() throws Exception {
+    validateRecreateClusterWithUserTableEnabled(true);
+  }
+
+  @Test
+  public void testRecreateCluster_UserTableEnabled_WithoutCleanupWALsAndZNodes() throws Exception {
+    TEST_UTIL.getConfiguration().setBoolean(HConstants.REMOVE_META_ON_RESTART,
+      HConstants.DEFAULT_REMOVE_META_ON_RESTART);
+    validateRecreateClusterWithUserTableEnabled(false);
+  }
+
+  private void validateRecreateClusterWithUserTableEnabled(boolean cleanupWALsAndZNodes)
+    throws Exception {
+    TEST_UTIL.startMiniCluster(NUM_RS);
+    try {
+      TableName tableName = TableName.valueOf("t1");
+      prepareDataBeforeRecreate(TEST_UTIL, tableName);
+      restartHBaseCluster(cleanupWALsAndZNodes);
+      validateDataAfterRecreate(TEST_UTIL, tableName);
+    } finally {
+      TEST_UTIL.shutdownMiniCluster();
+    }
+  }
+
+  private void restartHBaseCluster(boolean cleanUpWALsAndZNodes) throws Exception {
+    // flush cache so that everything is on disk
+    TEST_UTIL.getMiniHBaseCluster().flushcache();
+
+    List<ServerName> oldServers =
+      TEST_UTIL.getHBaseCluster().getMaster().getServerManager().getOnlineServersList();
+
+    // make sure there is no procedures pending
+    TEST_UTIL.waitFor(TIMEOUT_MS, () -> TEST_UTIL.getHBaseCluster().getMaster()
+      .getProcedures().stream().filter(p -> p.isFinished()).findAny().isPresent());
+
+    // shutdown and delete data if needed
+    Path walRootDirPath = TEST_UTIL.getMiniHBaseCluster().getMaster().getWALRootDir();
+    Path rootDirPath = CommonFSUtils.getRootDir(TEST_UTIL.getConfiguration());
+    TEST_UTIL.shutdownMiniHBaseCluster();
+
+    if (cleanUpWALsAndZNodes) {
+      TEST_UTIL.getDFSCluster().getFileSystem()
+        .delete(new Path(rootDirPath, MasterRegionFactory.MASTER_STORE_DIR), true);
+      TEST_UTIL.getDFSCluster().getFileSystem()
+        .delete(new Path(walRootDirPath, MasterRegionFactory.MASTER_STORE_DIR), true);
+      TEST_UTIL.getDFSCluster().getFileSystem()
+        .delete(new Path(walRootDirPath, WALProcedureStore.MASTER_PROCEDURE_LOGDIR), true);
+
+      TEST_UTIL.getDFSCluster().getFileSystem()
+        .delete(new Path(walRootDirPath, HConstants.HREGION_LOGDIR_NAME), true);
+      TEST_UTIL.getDFSCluster().getFileSystem()
+        .delete(new Path(walRootDirPath, HConstants.HREGION_OLDLOGDIR_NAME), true);
+      // delete all zk data
+      // we cannot keep ZK data because it will hold the meta region states as open and
+      // didn't submit a InitMetaProcedure
+      ZKUtil.deleteChildrenRecursively(TEST_UTIL.getZooKeeperWatcher(),
+        TEST_UTIL.getZooKeeperWatcher().getZNodePaths().baseZNode);
+      TEST_UTIL.shutdownMiniZKCluster();
+      TEST_UTIL.startMiniZKCluster();
+    }
+
+    TEST_UTIL.restartHBaseCluster(NUM_RS);
+    TEST_UTIL.waitFor(TIMEOUT_MS,
+      () -> TEST_UTIL.getMiniHBaseCluster().getNumLiveRegionServers() == NUM_RS);
+
+    // make sure we have a new set of region servers with different hostnames and ports
+    List<ServerName> newServers =
+      TEST_UTIL.getHBaseCluster().getMaster().getServerManager().getOnlineServersList();
+    assertFalse(newServers.stream().filter(newServer -> oldServers.contains(newServer)).findAny()
+      .isPresent());
+  }
+
+  private void prepareDataBeforeRecreate(
+      HBaseTestingUtility testUtil, TableName tableName) throws Exception {
+    Table table = testUtil.createTable(tableName, "f");
+    Put put = new Put(Bytes.toBytes("r1"));
+    put.addColumn(Bytes.toBytes("f"), Bytes.toBytes("c"), Bytes.toBytes("v"));
+    table.put(put);
+
+    ensureTableNotColocatedWithSystemTable(tableName, TableName.NAMESPACE_TABLE_NAME);
+  }
+
+  private void ensureTableNotColocatedWithSystemTable(TableName userTable, TableName systemTable)
+      throws IOException, InterruptedException {
+    MiniHBaseCluster hbaseCluster = TEST_UTIL.getHBaseCluster();
+    assertTrue("Please start more than 1 regionserver",
+        hbaseCluster.getRegionServerThreads().size() > 1);
+
+    int userTableServerNum = getServerNumForTableWithOnlyOneRegion(userTable);
+    int systemTableServerNum = getServerNumForTableWithOnlyOneRegion(systemTable);
+
+    if (userTableServerNum != systemTableServerNum) {
+      // no-ops if user table and system are already on a different host
+      return;
+    }
+
+    int destServerNum = (systemTableServerNum + 1) % NUM_RS;
+    assertTrue(systemTableServerNum != destServerNum);
+
+    HRegionServer systemTableServer = hbaseCluster.getRegionServer(systemTableServerNum);
+    HRegionServer destServer = hbaseCluster.getRegionServer(destServerNum);
+    assertTrue(!systemTableServer.equals(destServer));
+    // make sure the dest server is live before moving region
+    hbaseCluster.waitForRegionServerToStart(destServer.getServerName().getHostname(),
+        destServer.getServerName().getPort(), TIMEOUT_MS);
+    // move region of userTable to a different regionserver not co-located with system table
+    TEST_UTIL.moveRegionAndWait(TEST_UTIL.getAdmin().getRegions(userTable).get(0),
+        destServer.getServerName());
+  }
+
+  private int getServerNumForTableWithOnlyOneRegion(TableName tableName) throws IOException {
+    List<RegionInfo> tableRegionInfos = TEST_UTIL.getAdmin().getRegions(tableName);
+    assertEquals(1, tableRegionInfos.size());
+    return TEST_UTIL.getHBaseCluster()
+        .getServerWith(tableRegionInfos.get(0).getRegionName());
+  }
+
+  private void validateDataAfterRecreate(
+      HBaseTestingUtility testUtil, TableName tableName) throws Exception {
+    Table t1 = testUtil.getConnection().getTable(tableName);
+    Get get = new Get(Bytes.toBytes("r1"));
+    get.addColumn(Bytes.toBytes("f"), Bytes.toBytes("c"));
+    Result result = t1.get(get);
+    assertTrue(result.advance());
+    Cell cell = result.current();
+    assertEquals("v", Bytes.toString(cell.getValueArray(),
+        cell.getValueOffset(), cell.getValueLength()));
+    assertFalse(result.advance());
+  }
+
+}