001/* 002 * Licensed to the Apache Software Foundation (ASF) under one 003 * or more contributor license agreements. See the NOTICE file 004 * distributed with this work for additional information 005 * regarding copyright ownership. The ASF licenses this file 006 * to you under the Apache License, Version 2.0 (the 007 * "License"); you may not use this file except in compliance 008 * with the License. You may obtain a copy of the License at 009 * 010 * http://www.apache.org/licenses/LICENSE-2.0 011 * 012 * Unless required by applicable law or agreed to in writing, software 013 * distributed under the License is distributed on an "AS IS" BASIS, 014 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. 015 * See the License for the specific language governing permissions and 016 * limitations under the License. 017 */ 018package org.apache.hadoop.hbase.regionserver; 019 020import static org.junit.jupiter.api.Assertions.assertEquals; 021import static org.junit.jupiter.api.Assertions.assertTrue; 022 023import java.io.IOException; 024import java.util.List; 025import java.util.Map; 026import java.util.concurrent.atomic.AtomicBoolean; 027import java.util.concurrent.atomic.AtomicReference; 028import org.apache.hadoop.conf.Configuration; 029import org.apache.hadoop.hbase.HBaseConfiguration; 030import org.apache.hadoop.hbase.HBaseTestingUtil; 031import org.apache.hadoop.hbase.LocalHBaseCluster; 032import org.apache.hadoop.hbase.ServerName; 033import org.apache.hadoop.hbase.SingleProcessHBaseCluster; 034import org.apache.hadoop.hbase.client.RegionInfo; 035import org.apache.hadoop.hbase.master.HMaster; 036import org.apache.hadoop.hbase.master.ServerListener; 037import org.apache.hadoop.hbase.master.ServerManager; 038import org.apache.hadoop.hbase.testclassification.MediumTests; 039import org.apache.hadoop.hbase.testclassification.RegionServerTests; 040import org.apache.hadoop.hbase.util.Bytes; 041import org.apache.hadoop.hbase.util.JVMClusterUtil.MasterThread; 042import org.apache.hadoop.hbase.util.Threads; 043import org.junit.jupiter.api.Disabled; 044import org.junit.jupiter.api.Tag; 045import org.junit.jupiter.api.Test; 046import org.slf4j.Logger; 047import org.slf4j.LoggerFactory; 048 049import org.apache.hadoop.hbase.shaded.protobuf.generated.RegionServerStatusProtos.RegionServerStartupResponse; 050 051/** 052 * Tests that a regionserver that dies after reporting for duty gets removed from list of online 053 * regions. See HBASE-9593. 054 */ 055@Tag(RegionServerTests.TAG) 056@Tag(MediumTests.TAG) 057@Disabled("See HBASE-19515") 058public class TestRSKilledWhenInitializing { 059 060 private static final Logger LOG = LoggerFactory.getLogger(TestRSKilledWhenInitializing.class); 061 062 // This boolean needs to be globally available. It is used below in our 063 // mocked up regionserver so it knows when to die. 064 private static AtomicBoolean masterActive = new AtomicBoolean(false); 065 // Ditto for this variable. It also is used in the mocked regionserver class. 066 private static final AtomicReference<ServerName> killedRS = new AtomicReference<ServerName>(); 067 068 private static final int NUM_MASTERS = 1; 069 private static final int NUM_RS = 2; 070 071 /** 072 * Test verifies whether a region server is removed from online servers list in master if it went 073 * down after registering with master. Test will TIMEOUT if an error!!!! 074 */ 075 @Test 076 public void testRSTerminationAfterRegisteringToMasterBeforeCreatingEphemeralNode() 077 throws Exception { 078 // Create config to use for this cluster 079 Configuration conf = HBaseConfiguration.create(); 080 conf.setInt(ServerManager.WAIT_ON_REGIONSERVERS_MINTOSTART, 1); 081 // Start the cluster 082 final HBaseTestingUtil TEST_UTIL = new HBaseTestingUtil(conf); 083 TEST_UTIL.startMiniDFSCluster(3); 084 TEST_UTIL.startMiniZKCluster(); 085 TEST_UTIL.createRootDir(); 086 final LocalHBaseCluster cluster = new LocalHBaseCluster(conf, NUM_MASTERS, NUM_RS, 087 HMaster.class, RegisterAndDieRegionServer.class); 088 final MasterThread master = startMaster(cluster.getMasters().get(0)); 089 try { 090 // Master is up waiting on RegionServers to check in. Now start RegionServers. 091 for (int i = 0; i < NUM_RS; i++) { 092 cluster.getRegionServers().get(i).start(); 093 } 094 // Expected total regionservers depends on whether Master can host regions or not. 095 int expectedTotalRegionServers = NUM_RS; 096 List<ServerName> onlineServersList = null; 097 do { 098 onlineServersList = master.getMaster().getServerManager().getOnlineServersList(); 099 } while (onlineServersList.size() < expectedTotalRegionServers); 100 // Wait until killedRS is set. Means RegionServer is starting to go down. 101 while (killedRS.get() == null) { 102 Threads.sleep(1); 103 } 104 // Wait on the RegionServer to fully die. 105 while (cluster.getLiveRegionServers().size() >= expectedTotalRegionServers) { 106 Threads.sleep(1); 107 } 108 // Make sure Master is fully up before progressing. Could take a while if regions 109 // being reassigned. 110 while (!master.getMaster().isInitialized()) { 111 Threads.sleep(1); 112 } 113 114 // Now in steady state. How many regions open? Master should have too many regionservers 115 // showing still. The downed RegionServer should still be showing as registered. 116 assertTrue(master.getMaster().getServerManager().isServerOnline(killedRS.get())); 117 // Find non-meta region (namespace?) and assign to the killed server. That'll trigger cleanup. 118 Map<RegionInfo, ServerName> assignments = null; 119 do { 120 assignments = 121 master.getMaster().getAssignmentManager().getRegionStates().getRegionAssignments(); 122 } while (assignments == null || assignments.size() < 2); 123 RegionInfo hri = null; 124 for (Map.Entry<RegionInfo, ServerName> e : assignments.entrySet()) { 125 if (e.getKey().isMetaRegion()) continue; 126 hri = e.getKey(); 127 break; 128 } 129 // Try moving region to the killed server. It will fail. As by-product, we will 130 // remove the RS from Master online list because no corresponding znode. 131 assertEquals(expectedTotalRegionServers, 132 master.getMaster().getServerManager().getOnlineServersList().size()); 133 LOG.info("Move " + hri.getEncodedName() + " to " + killedRS.get()); 134 master.getMaster().move(hri.getEncodedNameAsBytes(), 135 Bytes.toBytes(killedRS.get().toString())); 136 137 // TODO: This test could do more to verify fix. It could create a table 138 // and do round-robin assign. It should fail if zombie RS. HBASE-19515. 139 140 // Wait until the RS no longer shows as registered in Master. 141 while (onlineServersList.size() > (NUM_RS + 1)) { 142 Thread.sleep(100); 143 onlineServersList = master.getMaster().getServerManager().getOnlineServersList(); 144 } 145 } finally { 146 // Shutdown is messy with complaints about fs being closed. Why? TODO. 147 cluster.shutdown(); 148 cluster.join(); 149 TEST_UTIL.shutdownMiniDFSCluster(); 150 TEST_UTIL.shutdownMiniZKCluster(); 151 TEST_UTIL.cleanupTestDir(); 152 } 153 } 154 155 /** 156 * Start Master. Get as far as the state where Master is waiting on RegionServers to check in, 157 * then return. 158 */ 159 private MasterThread startMaster(MasterThread master) { 160 master.start(); 161 // It takes a while until ServerManager creation to happen inside Master startup. 162 while (master.getMaster().getServerManager() == null) { 163 continue; 164 } 165 // Set a listener for the waiting-on-RegionServers state. We want to wait 166 // until this condition before we leave this method and start regionservers. 167 final AtomicBoolean waiting = new AtomicBoolean(false); 168 if (master.getMaster().getServerManager() == null) throw new NullPointerException("SM"); 169 master.getMaster().getServerManager().registerListener(new ServerListener() { 170 @Override 171 public void waiting() { 172 waiting.set(true); 173 } 174 }); 175 // Wait until the Master gets to place where it is waiting on RegionServers to check in. 176 while (!waiting.get()) { 177 continue; 178 } 179 // Set the global master-is-active; gets picked up by regionservers later. 180 masterActive.set(true); 181 return master; 182 } 183 184 /** 185 * A RegionServer that reports for duty and then immediately dies if it is the first to receive 186 * the response to a reportForDuty. When it dies, it clears its ephemeral znode which the master 187 * notices and so removes the region from its set of online regionservers. 188 */ 189 static class RegisterAndDieRegionServer 190 extends SingleProcessHBaseCluster.MiniHBaseClusterRegionServer { 191 public RegisterAndDieRegionServer(Configuration conf) throws IOException, InterruptedException { 192 super(conf); 193 } 194 195 @Override 196 protected void handleReportForDutyResponse(RegionServerStartupResponse c) throws IOException { 197 if (killedRS.compareAndSet(null, getServerName())) { 198 // Make sure Master is up so it will see the removal of the ephemeral znode for this RS. 199 while (!masterActive.get()) { 200 Threads.sleep(100); 201 } 202 super.kill(); 203 } else { 204 super.handleReportForDutyResponse(c); 205 } 206 } 207 } 208}