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.conf.Configuration;
024import org.apache.hadoop.hbase.HConstants;
025import org.apache.hadoop.hbase.TableName;
026import org.apache.hadoop.hbase.client.RegionInfo;
027import org.apache.hadoop.hbase.client.RegionReplicaUtil;
028import org.apache.hadoop.hbase.client.TableDescriptor;
029import org.apache.hadoop.hbase.executor.EventHandler;
030import org.apache.hadoop.hbase.executor.EventType;
031import org.apache.hadoop.hbase.regionserver.HRegion;
032import org.apache.hadoop.hbase.regionserver.HRegionServer;
033import org.apache.hadoop.hbase.regionserver.Region;
034import org.apache.hadoop.hbase.regionserver.RegionServerServices.PostOpenDeployContext;
035import org.apache.hadoop.hbase.regionserver.RegionServerServices.RegionStateTransitionContext;
036import org.apache.hadoop.hbase.util.RetryCounter;
037import org.apache.hadoop.hbase.util.ServerRegionReplicaUtil;
038import org.apache.yetus.audience.InterfaceAudience;
039import org.slf4j.Logger;
040import org.slf4j.LoggerFactory;
041
042import org.apache.hadoop.hbase.shaded.protobuf.generated.RegionServerStatusProtos.RegionStateTransition.TransitionCode;
043
044/**
045 * Handles opening of a region on a region server.
046 * <p/>
047 * Just done the same thing with the old {@link OpenRegionHandler}, with some modifications on
048 * fencing and retrying. But we need to keep the {@link OpenRegionHandler} as is to keep compatible
049 * with the zk less assignment for 1.x, otherwise it is not possible to do rolling upgrade.
050 */
051@InterfaceAudience.Private
052public class AssignRegionHandler extends EventHandler {
053
054  private static final Logger LOG = LoggerFactory.getLogger(AssignRegionHandler.class);
055
056  private final RegionInfo regionInfo;
057
058  private final long openProcId;
059
060  private final TableDescriptor tableDesc;
061
062  private final long masterSystemTime;
063
064  private final RetryCounter retryCounter;
065
066  public AssignRegionHandler(HRegionServer server, RegionInfo regionInfo, long openProcId,
067    @Nullable TableDescriptor tableDesc, long masterSystemTime, EventType eventType) {
068    super(server, eventType);
069    this.regionInfo = regionInfo;
070    this.openProcId = openProcId;
071    this.tableDesc = tableDesc;
072    this.masterSystemTime = masterSystemTime;
073    this.retryCounter = HandlerUtil.getRetryCounter();
074  }
075
076  private HRegionServer getServer() {
077    return (HRegionServer) server;
078  }
079
080  private void cleanUpAndReportFailure(IOException error) throws IOException {
081    LOG.warn("Failed to open region {}, will report to master", regionInfo.getRegionNameAsString(),
082      error);
083    HRegionServer rs = getServer();
084    rs.getRegionsInTransitionInRS().remove(regionInfo.getEncodedNameAsBytes(), Boolean.TRUE);
085    if (
086      !rs.reportRegionStateTransition(new RegionStateTransitionContext(TransitionCode.FAILED_OPEN,
087        HConstants.NO_SEQNUM, openProcId, masterSystemTime, regionInfo))
088    ) {
089      throw new IOException(
090        "Failed to report failed open to master: " + regionInfo.getRegionNameAsString());
091    }
092  }
093
094  @Override
095  public void process() throws IOException {
096    HRegionServer rs = getServer();
097    String encodedName = regionInfo.getEncodedName();
098    byte[] encodedNameBytes = regionInfo.getEncodedNameAsBytes();
099    String regionName = regionInfo.getRegionNameAsString();
100    Region onlineRegion = rs.getRegion(encodedName);
101    if (onlineRegion != null) {
102      LOG.warn("Received OPEN for {} which is already online", regionName);
103      // Just follow the old behavior, do we need to call reportRegionStateTransition? Maybe not?
104      // For normal case, it could happen that the rpc call to schedule this handler is succeeded,
105      // but before returning to master the connection is broken. And when master tries again, we
106      // have already finished the opening. For this case we do not need to call
107      // reportRegionStateTransition any more.
108      return;
109    }
110    Boolean previous = rs.getRegionsInTransitionInRS().putIfAbsent(encodedNameBytes, Boolean.TRUE);
111    if (previous != null) {
112      if (previous) {
113        // The region is opening and this maybe a retry on the rpc call, it is safe to ignore it.
114        LOG.info("Receiving OPEN for {} which we are already trying to OPEN"
115          + " - ignoring this new request for this region.", regionName);
116      } else {
117        // The region is closing. This is possible as we will update the region state to CLOSED when
118        // calling reportRegionStateTransition, so the HMaster will think the region is offline,
119        // before we actually close the region, as reportRegionStateTransition is part of the
120        // closing process.
121        long backoff = retryCounter.getBackoffTimeAndIncrementAttempts();
122        LOG.info("Receiving OPEN for {} which we are trying to close, try again after {}ms",
123          regionName, backoff);
124        rs.getExecutorService().delayedSubmit(this, backoff, TimeUnit.MILLISECONDS);
125      }
126      return;
127    }
128    LOG.info("Open {}", regionName);
129    HRegion region;
130    try {
131      TableDescriptor htd =
132        tableDesc != null ? tableDesc : rs.getTableDescriptors().get(regionInfo.getTable());
133      if (htd == null) {
134        throw new IOException("Missing table descriptor for " + regionName);
135      }
136      // pass null for the last parameter, which used to be a CancelableProgressable, as now the
137      // opening can not be interrupted by a close request any more.
138      Configuration conf = rs.getConfiguration();
139      TableName tn = htd.getTableName();
140      if (ServerRegionReplicaUtil.isMetaRegionReplicaReplicationEnabled(conf, tn)) {
141        if (RegionReplicaUtil.isDefaultReplica(this.regionInfo.getReplicaId())) {
142          // Add the hbase:meta replication source on replica zero/default.
143          rs.getReplicationSourceService().getReplicationManager()
144            .addCatalogReplicationSource(this.regionInfo);
145        }
146      }
147      region = HRegion.openHRegion(regionInfo, htd, rs.getWAL(regionInfo), conf, rs, null);
148    } catch (IOException e) {
149      cleanUpAndReportFailure(e);
150      return;
151    }
152    // From here on out, this is PONR. We can not revert back. The only way to address an
153    // exception from here on out is to abort the region server.
154    rs.postOpenDeployTasks(new PostOpenDeployContext(region, openProcId, masterSystemTime));
155    rs.addRegion(region);
156    LOG.info("Opened {}", regionName);
157    // Cache the open region procedure id after report region transition succeed.
158    rs.finishRegionProcedure(openProcId);
159    Boolean current = rs.getRegionsInTransitionInRS().remove(regionInfo.getEncodedNameAsBytes());
160    if (current == null) {
161      // Should NEVER happen, but let's be paranoid.
162      LOG.error("Bad state: we've just opened {} which was NOT in transition", regionName);
163    } else if (!current) {
164      // Should NEVER happen, but let's be paranoid.
165      LOG.error("Bad state: we've just opened {} which was closing", regionName);
166    }
167  }
168
169  @Override
170  protected void handleException(Throwable t) {
171    LOG.warn("Fatal error occurred while opening region {}, aborting...",
172      regionInfo.getRegionNameAsString(), t);
173    // Clear any reference in getServer().getRegionsInTransitionInRS() otherwise can hold up
174    // regionserver abort on cluster shutdown. HBASE-23984.
175    getServer().getRegionsInTransitionInRS().remove(regionInfo.getEncodedNameAsBytes());
176    getServer().abort(
177      "Failed to open region " + regionInfo.getRegionNameAsString() + " and can not recover", t);
178  }
179
180  public static AssignRegionHandler create(HRegionServer server, RegionInfo regionInfo,
181    long openProcId, TableDescriptor tableDesc, long masterSystemTime) {
182    EventType eventType;
183    if (regionInfo.isMetaRegion()) {
184      eventType = EventType.M_RS_OPEN_META;
185    } else if (
186      regionInfo.getTable().isSystemTable()
187        || (tableDesc != null && tableDesc.getPriority() >= HConstants.ADMIN_QOS)
188    ) {
189      eventType = EventType.M_RS_OPEN_PRIORITY_REGION;
190    } else {
191      eventType = EventType.M_RS_OPEN_REGION;
192    }
193    return new AssignRegionHandler(server, regionInfo, openProcId, tableDesc, masterSystemTime,
194      eventType);
195  }
196}