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