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.handler; 019 020import edu.umd.cs.findbugs.annotations.Nullable; 021import java.io.IOException; 022import java.util.concurrent.TimeUnit; 023import org.apache.hadoop.hbase.HConstants; 024import org.apache.hadoop.hbase.ServerName; 025import org.apache.hadoop.hbase.client.RegionReplicaUtil; 026import org.apache.hadoop.hbase.executor.EventHandler; 027import org.apache.hadoop.hbase.executor.EventType; 028import org.apache.hadoop.hbase.regionserver.HRegion; 029import org.apache.hadoop.hbase.regionserver.HRegionServer; 030import org.apache.hadoop.hbase.regionserver.Region; 031import org.apache.hadoop.hbase.regionserver.RegionServerServices.RegionStateTransitionContext; 032import org.apache.hadoop.hbase.util.Bytes; 033import org.apache.hadoop.hbase.util.RetryCounter; 034import org.apache.hadoop.hbase.util.ServerRegionReplicaUtil; 035import org.apache.yetus.audience.InterfaceAudience; 036import org.slf4j.Logger; 037import org.slf4j.LoggerFactory; 038 039import org.apache.hadoop.hbase.shaded.protobuf.generated.RegionServerStatusProtos.RegionStateTransition.TransitionCode; 040 041/** 042 * Handles closing of a region on a region server. 043 * <p/> 044 * Just done the same thing with the old {@link CloseRegionHandler}, with some modifications on 045 * fencing and retrying. But we need to keep the {@link CloseRegionHandler} as is to keep compatible 046 * with the zk less assignment for 1.x, otherwise it is not possible to do rolling upgrade. 047 */ 048@InterfaceAudience.Private 049public class UnassignRegionHandler extends EventHandler { 050 051 private static final Logger LOG = LoggerFactory.getLogger(UnassignRegionHandler.class); 052 053 private final String encodedName; 054 055 private final long closeProcId; 056 // If true, the hosting server is aborting. Region close process is different 057 // when we are aborting. 058 // TODO: not used yet, we still use the old CloseRegionHandler when aborting 059 private final boolean abort; 060 061 private final ServerName destination; 062 063 private final RetryCounter retryCounter; 064 065 public UnassignRegionHandler(HRegionServer server, String encodedName, long closeProcId, 066 boolean abort, @Nullable ServerName destination, EventType eventType) { 067 super(server, eventType); 068 this.encodedName = encodedName; 069 this.closeProcId = closeProcId; 070 this.abort = abort; 071 this.destination = destination; 072 this.retryCounter = HandlerUtil.getRetryCounter(); 073 } 074 075 private HRegionServer getServer() { 076 return (HRegionServer) server; 077 } 078 079 @Override 080 public void process() throws IOException { 081 HRegionServer rs = getServer(); 082 byte[] encodedNameBytes = Bytes.toBytes(encodedName); 083 Boolean previous = rs.getRegionsInTransitionInRS().putIfAbsent(encodedNameBytes, Boolean.FALSE); 084 if (previous != null) { 085 if (previous) { 086 // This could happen as we will update the region state to OPEN when calling 087 // reportRegionStateTransition, so the HMaster will think the region is online, before we 088 // actually open the region, as reportRegionStateTransition is part of the opening process. 089 long backoff = retryCounter.getBackoffTimeAndIncrementAttempts(); 090 LOG.warn( 091 "Received CLOSE for {} which we are already " + "trying to OPEN; try again after {}ms", 092 encodedName, backoff); 093 rs.getExecutorService().delayedSubmit(this, backoff, TimeUnit.MILLISECONDS); 094 } else { 095 LOG.info( 096 "Received CLOSE for {} which we are already trying to CLOSE," + " but not completed yet", 097 encodedName); 098 } 099 return; 100 } 101 HRegion region = rs.getRegion(encodedName); 102 if (region == null) { 103 LOG.debug("Received CLOSE for {} which is not ONLINE and we're not opening/closing.", 104 encodedName); 105 rs.getRegionsInTransitionInRS().remove(encodedNameBytes, Boolean.FALSE); 106 return; 107 } 108 String regionName = region.getRegionInfo().getEncodedName(); 109 LOG.info("Close {}", regionName); 110 if (region.getCoprocessorHost() != null) { 111 // XXX: The behavior is a bit broken. At master side there is no FAILED_CLOSE state, so if 112 // there are exception thrown from the CP, we can not report the error to master, and if 113 // here we just return without calling reportRegionStateTransition, the TRSP at master side 114 // will hang there for ever. So here if the CP throws an exception out, the only way is to 115 // abort the RS... 116 region.getCoprocessorHost().preClose(abort); 117 } 118 if (region.close(abort) == null) { 119 // XXX: Is this still possible? The old comment says about split, but now split is done at 120 // master side, so... 121 LOG.warn("Can't close {}, already closed during close()", regionName); 122 rs.getRegionsInTransitionInRS().remove(encodedNameBytes, Boolean.FALSE); 123 return; 124 } 125 126 rs.removeRegion(region, destination); 127 if ( 128 ServerRegionReplicaUtil.isMetaRegionReplicaReplicationEnabled(rs.getConfiguration(), 129 region.getTableDescriptor().getTableName()) 130 ) { 131 if (RegionReplicaUtil.isDefaultReplica(region.getRegionInfo().getReplicaId())) { 132 // If hbase:meta read replicas enabled, remove replication source for hbase:meta Regions. 133 // See assign region handler where we add the replication source on open. 134 rs.getReplicationSourceService().getReplicationManager() 135 .removeCatalogReplicationSource(region.getRegionInfo()); 136 } 137 } 138 if ( 139 !rs.reportRegionStateTransition(new RegionStateTransitionContext(TransitionCode.CLOSED, 140 HConstants.NO_SEQNUM, closeProcId, -1, region.getRegionInfo())) 141 ) { 142 throw new IOException("Failed to report close to master: " + regionName); 143 } 144 // Cache the close region procedure id after report region transition succeed. 145 rs.finishRegionProcedure(closeProcId); 146 rs.getRegionsInTransitionInRS().remove(encodedNameBytes, Boolean.FALSE); 147 LOG.info("Closed {}", regionName); 148 } 149 150 @Override 151 protected void handleException(Throwable t) { 152 LOG.warn("Fatal error occurred while closing region {}, aborting...", encodedName, t); 153 // Clear any reference in getServer().getRegionsInTransitionInRS() otherwise can hold up 154 // regionserver abort on cluster shutdown. HBASE-23984. 155 getServer().getRegionsInTransitionInRS().remove(Bytes.toBytes(this.encodedName)); 156 getServer().abort("Failed to close region " + encodedName + " and can not recover", t); 157 } 158 159 public static UnassignRegionHandler create(HRegionServer server, String encodedName, 160 long closeProcId, boolean abort, @Nullable ServerName destination) { 161 // Just try our best to determine whether it is for closing meta. It is not the end of the world 162 // if we put the handler into a wrong executor. 163 Region region = server.getRegion(encodedName); 164 EventType eventType = region != null && region.getRegionInfo().isMetaRegion() 165 ? EventType.M_RS_CLOSE_META 166 : EventType.M_RS_CLOSE_REGION; 167 return new UnassignRegionHandler(server, encodedName, closeProcId, abort, destination, 168 eventType); 169 } 170}