/**
*
* 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 java.io.IOException;
import java.util.ArrayList;
import java.util.Arrays;
import java.util.Collections;
import java.util.HashMap;
import java.util.HashSet;
import java.util.Iterator;
import java.util.List;
import java.util.Map;
import java.util.NavigableMap;
import java.util.Set;
import java.util.TreeMap;
import java.util.concurrent.ConcurrentHashMap;
import java.util.concurrent.ConcurrentSkipListSet;
import java.util.concurrent.ThreadFactory;
import java.util.concurrent.TimeUnit;
import java.util.concurrent.atomic.AtomicBoolean;
import java.util.concurrent.atomic.AtomicInteger;
import java.util.concurrent.locks.Lock;
import java.util.concurrent.locks.ReentrantLock;
import org.apache.commons.logging.Log;
import org.apache.commons.logging.LogFactory;
import org.apache.hadoop.classification.InterfaceAudience;
import org.apache.hadoop.conf.Configuration;
import org.apache.hadoop.hbase.Chore;
import org.apache.hadoop.hbase.HBaseIOException;
import org.apache.hadoop.hbase.HConstants;
import org.apache.hadoop.hbase.HRegionInfo;
import org.apache.hadoop.hbase.NotServingRegionException;
import org.apache.hadoop.hbase.RegionTransition;
import org.apache.hadoop.hbase.Server;
import org.apache.hadoop.hbase.ServerName;
import org.apache.hadoop.hbase.Stoppable;
import org.apache.hadoop.hbase.TableName;
import org.apache.hadoop.hbase.TableNotFoundException;
import org.apache.hadoop.hbase.catalog.CatalogTracker;
import org.apache.hadoop.hbase.catalog.MetaReader;
import org.apache.hadoop.hbase.client.Result;
import org.apache.hadoop.hbase.exceptions.DeserializationException;
import org.apache.hadoop.hbase.executor.EventHandler;
import org.apache.hadoop.hbase.executor.EventType;
import org.apache.hadoop.hbase.executor.ExecutorService;
import org.apache.hadoop.hbase.ipc.ServerNotRunningYetException;
import org.apache.hadoop.hbase.master.RegionState.State;
import org.apache.hadoop.hbase.master.balancer.FavoredNodeAssignmentHelper;
import org.apache.hadoop.hbase.master.balancer.FavoredNodeLoadBalancer;
import org.apache.hadoop.hbase.master.handler.ClosedRegionHandler;
import org.apache.hadoop.hbase.master.handler.DisableTableHandler;
import org.apache.hadoop.hbase.master.handler.EnableTableHandler;
import org.apache.hadoop.hbase.master.handler.OpenedRegionHandler;
import org.apache.hadoop.hbase.regionserver.RegionAlreadyInTransitionException;
import org.apache.hadoop.hbase.regionserver.RegionMergeTransaction;
import org.apache.hadoop.hbase.regionserver.RegionOpeningState;
import org.apache.hadoop.hbase.regionserver.RegionServerStoppedException;
import org.apache.hadoop.hbase.regionserver.SplitTransaction;
import org.apache.hadoop.hbase.util.EnvironmentEdgeManager;
import org.apache.hadoop.hbase.util.KeyLocker;
import org.apache.hadoop.hbase.util.Pair;
import org.apache.hadoop.hbase.util.PairOfSameType;
import org.apache.hadoop.hbase.util.Threads;
import org.apache.hadoop.hbase.util.Triple;
import org.apache.hadoop.hbase.zookeeper.MetaRegionTracker;
import org.apache.hadoop.hbase.zookeeper.ZKAssign;
import org.apache.hadoop.hbase.zookeeper.ZKTable;
import org.apache.hadoop.hbase.zookeeper.ZKUtil;
import org.apache.hadoop.hbase.zookeeper.ZooKeeperListener;
import org.apache.hadoop.ipc.RemoteException;
import org.apache.zookeeper.AsyncCallback;
import org.apache.zookeeper.KeeperException;
import org.apache.zookeeper.KeeperException.NoNodeException;
import org.apache.zookeeper.KeeperException.NodeExistsException;
import org.apache.zookeeper.data.Stat;
import com.google.common.base.Preconditions;
import com.google.common.collect.LinkedHashMultimap;
/**
* Manages and performs region assignment.
* <p>
* Monitors ZooKeeper for events related to regions in transition.
* <p>
* Handles existing regions in transition during master failover.
*/
@InterfaceAudience.Private
public class AssignmentManager extends ZooKeeperListener {
private static final Log LOG = LogFactory.getLog(AssignmentManager.class);
public static final ServerName HBCK_CODE_SERVERNAME = ServerName.valueOf(HConstants.HBCK_CODE_NAME,
-1, -1L);
public static final String ASSIGNMENT_TIMEOUT = "hbase.master.assignment.timeoutmonitor.timeout";
public static final int DEFAULT_ASSIGNMENT_TIMEOUT_DEFAULT = 600000;
public static final String ASSIGNMENT_TIMEOUT_MANAGEMENT = "hbase.assignment.timeout.management";
public static final boolean DEFAULT_ASSIGNMENT_TIMEOUT_MANAGEMENT = false;
public static final String ALREADY_IN_TRANSITION_WAITTIME
= "hbase.assignment.already.intransition.waittime";
public static final int DEFAULT_ALREADY_IN_TRANSITION_WAITTIME = 60000; // 1 minute
protected final Server server;
private ServerManager serverManager;
private boolean shouldAssignRegionsWithFavoredNodes;
private CatalogTracker catalogTracker;
protected final TimeoutMonitor timeoutMonitor;
private final TimerUpdater timerUpdater;
private LoadBalancer balancer;
private final MetricsAssignmentManager metricsAssignmentManager;
private final TableLockManager tableLockManager;
private AtomicInteger numRegionsOpened = new AtomicInteger(0);
final private KeyLocker<String> locker = new KeyLocker<String>();
/**
* Map of regions to reopen after the schema of a table is changed. Key -
* encoded region name, value - HRegionInfo
*/
private final Map <String, HRegionInfo> regionsToReopen;
/*
* Maximum times we recurse an assignment/unassignment.
* See below in {@link #assign()} and {@link #unassign()}.
*/
private final int maximumAttempts;
/**
* Map of two merging regions from the region to be created.
*/
private final Map<String, PairOfSameType<HRegionInfo>> mergingRegions
= new HashMap<String, PairOfSameType<HRegionInfo>>();
/**
* The sleep time for which the assignment will wait before retrying in case of hbase:meta assignment
* failure due to lack of availability of region plan
*/
private final long sleepTimeBeforeRetryingMetaAssignment;
/** Plans for region movement. Key is the encoded version of a region name*/
// TODO: When do plans get cleaned out? Ever? In server open and in server
// shutdown processing -- St.Ack
// All access to this Map must be synchronized.
final NavigableMap<String, RegionPlan> regionPlans =
new TreeMap<String, RegionPlan>();
private final ZKTable zkTable;
/**
* Contains the server which need to update timer, these servers will be
* handled by {@link TimerUpdater}
*/
private final ConcurrentSkipListSet<ServerName> serversInUpdatingTimer;
private final ExecutorService executorService;
// For unit tests, keep track of calls to ClosedRegionHandler
private Map<HRegionInfo, AtomicBoolean> closedRegionHandlerCalled = null;
// For unit tests, keep track of calls to OpenedRegionHandler
private Map<HRegionInfo, AtomicBoolean> openedRegionHandlerCalled = null;
//Thread pool executor service for timeout monitor
private java.util.concurrent.ExecutorService threadPoolExecutorService;
// A bunch of ZK events workers. Each is a single thread executor service
private final java.util.concurrent.ExecutorService zkEventWorkers;
private List<EventType> ignoreStatesRSOffline = Arrays.asList(
EventType.RS_ZK_REGION_FAILED_OPEN, EventType.RS_ZK_REGION_CLOSED);
private final RegionStates regionStates;
// The threshold to use bulk assigning. Using bulk assignment
// only if assigning at least this many regions to at least this
// many servers. If assigning fewer regions to fewer servers,
// bulk assigning may be not as efficient.
private final int bulkAssignThresholdRegions;
private final int bulkAssignThresholdServers;
// Should bulk assignment wait till all regions are assigned,
// or it is timed out? This is useful to measure bulk assignment
// performance, but not needed in most use cases.
private final boolean bulkAssignWaitTillAllAssigned;
/**
* Indicator that AssignmentManager has recovered the region states so
* that ServerShutdownHandler can be fully enabled and re-assign regions
* of dead servers. So that when re-assignment happens, AssignmentManager
* has proper region states.
*
* Protected to ease testing.
*/
protected final AtomicBoolean failoverCleanupDone = new AtomicBoolean(false);
/** Is the TimeOutManagement activated **/
private final boolean tomActivated;
/**
* A map to track the count a region fails to open in a row.
* So that we don't try to open a region forever if the failure is
* unrecoverable. We don't put this information in region states
* because we don't expect this to happen frequently; we don't
* want to copy this information over during each state transition either.
*/
private final ConcurrentHashMap<String, AtomicInteger>
failedOpenTracker = new ConcurrentHashMap<String, AtomicInteger>();
/**
* For testing only! Set to true to skip handling of split.
*/
@edu.umd.cs.findbugs.annotations.SuppressWarnings(value="MS_SHOULD_BE_FINAL")
public static boolean TEST_SKIP_SPLIT_HANDLING = false;
/**
* Constructs a new assignment manager.
*
* @param server
* @param serverManager
* @param catalogTracker
* @param service
* @throws KeeperException
* @throws IOException
*/
public AssignmentManager(Server server, ServerManager serverManager,
CatalogTracker catalogTracker, final LoadBalancer balancer,
final ExecutorService service, MetricsMaster metricsMaster,
final TableLockManager tableLockManager) throws KeeperException, IOException {
super(server.getZooKeeper());
this.server = server;
this.serverManager = serverManager;
this.catalogTracker = catalogTracker;
this.executorService = service;
this.regionsToReopen = Collections.synchronizedMap
(new HashMap<String, HRegionInfo> ());
Configuration conf = server.getConfiguration();
// Only read favored nodes if using the favored nodes load balancer.
this.shouldAssignRegionsWithFavoredNodes = conf.getClass(
HConstants.HBASE_MASTER_LOADBALANCER_CLASS, Object.class).equals(
FavoredNodeLoadBalancer.class);
this.tomActivated = conf.getBoolean(
ASSIGNMENT_TIMEOUT_MANAGEMENT, DEFAULT_ASSIGNMENT_TIMEOUT_MANAGEMENT);
if (tomActivated){
this.serversInUpdatingTimer = new ConcurrentSkipListSet<ServerName>();
this.timeoutMonitor = new TimeoutMonitor(
conf.getInt("hbase.master.assignment.timeoutmonitor.period", 30000),
server, serverManager,
conf.getInt(ASSIGNMENT_TIMEOUT, DEFAULT_ASSIGNMENT_TIMEOUT_DEFAULT));
this.timerUpdater = new TimerUpdater(conf.getInt(
"hbase.master.assignment.timerupdater.period", 10000), server);
Threads.setDaemonThreadRunning(timerUpdater.getThread(),
server.getServerName() + ".timerUpdater");
} else {
this.serversInUpdatingTimer = null;
this.timeoutMonitor = null;
this.timerUpdater = null;
}
this.zkTable = new ZKTable(this.watcher);
// This is the max attempts, not retries, so it should be at least 1.
this.maximumAttempts = Math.max(1,
this.server.getConfiguration().getInt("hbase.assignment.maximum.attempts", 10));
this.sleepTimeBeforeRetryingMetaAssignment = this.server.getConfiguration().getLong(
"hbase.meta.assignment.retry.sleeptime", 1000l);
this.balancer = balancer;
int maxThreads = conf.getInt("hbase.assignment.threads.max", 30);
this.threadPoolExecutorService = Threads.getBoundedCachedThreadPool(
maxThreads, 60L, TimeUnit.SECONDS, Threads.newDaemonThreadFactory("AM."));
this.regionStates = new RegionStates(server, serverManager);
this.bulkAssignWaitTillAllAssigned =
conf.getBoolean("hbase.bulk.assignment.waittillallassigned", false);
this.bulkAssignThresholdRegions = conf.getInt("hbase.bulk.assignment.threshold.regions", 7);
this.bulkAssignThresholdServers = conf.getInt("hbase.bulk.assignment.threshold.servers", 3);
int workers = conf.getInt("hbase.assignment.zkevent.workers", 20);
ThreadFactory threadFactory = Threads.newDaemonThreadFactory("AM.ZK.Worker");
zkEventWorkers = Threads.getBoundedCachedThreadPool(workers, 60L,
TimeUnit.SECONDS, threadFactory);
this.tableLockManager = tableLockManager;
this.metricsAssignmentManager = new MetricsAssignmentManager();
}
void startTimeOutMonitor() {
if (tomActivated) {
Threads.setDaemonThreadRunning(timeoutMonitor.getThread(), server.getServerName()
+ ".timeoutMonitor");
}
}
/**
* @return Instance of ZKTable.
*/
public ZKTable getZKTable() {
// These are 'expensive' to make involving trip to zk ensemble so allow
// sharing.
return this.zkTable;
}
/**
* This SHOULD not be public. It is public now
* because of some unit tests.
*
* TODO: make it package private and keep RegionStates in the master package
*/
public RegionStates getRegionStates() {
return regionStates;
}
public RegionPlan getRegionReopenPlan(HRegionInfo hri) {
return new RegionPlan(hri, null, regionStates.getRegionServerOfRegion(hri));
}
/**
* Add a regionPlan for the specified region.
* @param encodedName
* @param plan
*/
public void addPlan(String encodedName, RegionPlan plan) {
synchronized (regionPlans) {
regionPlans.put(encodedName, plan);
}
}
/**
* Add a map of region plans.
*/
public void addPlans(Map<String, RegionPlan> plans) {
synchronized (regionPlans) {
regionPlans.putAll(plans);
}
}
/**
* Set the list of regions that will be reopened
* because of an update in table schema
*
* @param regions
* list of regions that should be tracked for reopen
*/
public void setRegionsToReopen(List <HRegionInfo> regions) {
for(HRegionInfo hri : regions) {
regionsToReopen.put(hri.getEncodedName(), hri);
}
}
/**
* Used by the client to identify if all regions have the schema updates
*
* @param tableName
* @return Pair indicating the status of the alter command
* @throws IOException
*/
public Pair<Integer, Integer> getReopenStatus(TableName tableName)
throws IOException {
List <HRegionInfo> hris =
MetaReader.getTableRegions(this.server.getCatalogTracker(), tableName, true);
Integer pending = 0;
for (HRegionInfo hri : hris) {
String name = hri.getEncodedName();
// no lock concurrent access ok: sequential consistency respected.
if (regionsToReopen.containsKey(name)
|| regionStates.isRegionInTransition(name)) {
pending++;
}
}
return new Pair<Integer, Integer>(pending, hris.size());
}
/**
* Used by ServerShutdownHandler to make sure AssignmentManager has completed
* the failover cleanup before re-assigning regions of dead servers. So that
* when re-assignment happens, AssignmentManager has proper region states.
*/
public boolean isFailoverCleanupDone() {
return failoverCleanupDone.get();
}
/**
* To avoid racing with AM, external entities may need to lock a region,
* for example, when SSH checks what regions to skip re-assigning.
*/
public Lock acquireRegionLock(final String encodedName) {
return locker.acquireLock(encodedName);
}
/**
* Now, failover cleanup is completed. Notify server manager to
* process queued up dead servers processing, if any.
*/
void failoverCleanupDone() {
failoverCleanupDone.set(true);
serverManager.processQueuedDeadServers();
}
/**
* Called on startup.
* Figures whether a fresh cluster start of we are joining extant running cluster.
* @throws IOException
* @throws KeeperException
* @throws InterruptedException
*/
void joinCluster() throws IOException,
KeeperException, InterruptedException {
// Concurrency note: In the below the accesses on regionsInTransition are
// outside of a synchronization block where usually all accesses to RIT are
// synchronized. The presumption is that in this case it is safe since this
// method is being played by a single thread on startup.
// TODO: Regions that have a null location and are not in regionsInTransitions
// need to be handled.
// Scan hbase:meta to build list of existing regions, servers, and assignment
// Returns servers who have not checked in (assumed dead) and their regions
Map<ServerName, List<HRegionInfo>> deadServers = rebuildUserRegions();
// This method will assign all user regions if a clean server startup or
// it will reconstruct master state and cleanup any leftovers from
// previous master process.
processDeadServersAndRegionsInTransition(deadServers);
recoverTableInDisablingState();
recoverTableInEnablingState();
}
/**
* Process all regions that are in transition in zookeeper and also
* processes the list of dead servers by scanning the META.
* Used by master joining an cluster. If we figure this is a clean cluster
* startup, will assign all user regions.
* @param deadServers
* Map of dead servers and their regions. Can be null.
* @throws KeeperException
* @throws IOException
* @throws InterruptedException
*/
void processDeadServersAndRegionsInTransition(
final Map<ServerName, List<HRegionInfo>> deadServers)
throws KeeperException, IOException, InterruptedException {
List<String> nodes = ZKUtil.listChildrenNoWatch(watcher,
watcher.assignmentZNode);
if (nodes == null) {
String errorMessage = "Failed to get the children from ZK";
server.abort(errorMessage, new IOException(errorMessage));
return;
}
boolean failover = (!serverManager.getDeadServers().isEmpty() || !serverManager
.getRequeuedDeadServers().isEmpty());
if (!failover) {
// If any one region except meta is assigned, it's a failover.
Map<HRegionInfo, ServerName> regions = regionStates.getRegionAssignments();
for (HRegionInfo hri: regions.keySet()) {
if (!hri.isMetaTable()) {
LOG.debug("Found " + hri + " out on cluster");
failover = true;
break;
}
}
if (!failover) {
// If any one region except meta is in transition, it's a failover.
for (String encodedName: nodes) {
RegionState state = regionStates.getRegionState(encodedName);
if (state != null && !state.getRegion().isMetaRegion()) {
LOG.debug("Found " + state.getRegion().getRegionNameAsString() + " in RITs");
failover = true;
break;
}
}
}
}
// If we found user regions out on cluster, its a failover.
if (failover) {
LOG.info("Found regions out on cluster or in RIT; presuming failover");
// Process list of dead servers and regions in RIT.
// See HBASE-4580 for more information.
processDeadServersAndRecoverLostRegions(deadServers);
} else {
// Fresh cluster startup.
LOG.info("Clean cluster startup. Assigning userregions");
assignAllUserRegions();
}
}
/**
* If region is up in zk in transition, then do fixup and block and wait until
* the region is assigned and out of transition. Used on startup for
* catalog regions.
* @param hri Region to look for.
* @return True if we processed a region in transition else false if region
* was not up in zk in transition.
* @throws InterruptedException
* @throws KeeperException
* @throws IOException
*/
boolean processRegionInTransitionAndBlockUntilAssigned(final HRegionInfo hri)
throws InterruptedException, KeeperException, IOException {
String encodedRegionName = hri.getEncodedName();
if (!processRegionInTransition(encodedRegionName, hri)) {
return false; // The region is not in transition
}
LOG.debug("Waiting on " + HRegionInfo.prettyPrint(encodedRegionName));
while (!this.server.isStopped() &&
this.regionStates.isRegionInTransition(encodedRegionName)) {
RegionState state = this.regionStates.getRegionTransitionState(encodedRegionName);
if (state == null || !serverManager.isServerOnline(state.getServerName())) {
// The region is not in transition, or not in transition on an online
// server. Doesn't help to block here any more. Caller need to
// verify the region is actually assigned.
break;
}
this.regionStates.waitForUpdate(100);
}
return true;
}
/**
* Process failover of new master for region <code>encodedRegionName</code>
* up in zookeeper.
* @param encodedRegionName Region to process failover for.
* @param regionInfo If null we'll go get it from meta table.
* @return True if we processed <code>regionInfo</code> as a RIT.
* @throws KeeperException
* @throws IOException
*/
boolean processRegionInTransition(final String encodedRegionName,
final HRegionInfo regionInfo) throws KeeperException, IOException {
// We need a lock here to ensure that we will not put the same region twice
// It has no reason to be a lock shared with the other operations.
// We can do the lock on the region only, instead of a global lock: what we want to ensure
// is that we don't have two threads working on the same region.
Lock lock = locker.acquireLock(encodedRegionName);
try {
Stat stat = new Stat();
byte [] data = ZKAssign.getDataAndWatch(watcher, encodedRegionName, stat);
if (data == null) return false;
RegionTransition rt;
try {
rt = RegionTransition.parseFrom(data);
} catch (DeserializationException e) {
LOG.warn("Failed parse znode data", e);
return false;
}
HRegionInfo hri = regionInfo;
if (hri == null) {
// The region info is not passed in. We will try to find the region
// from region states map/meta based on the encoded region name. But we
// may not be able to find it. This is valid for online merge that
// the region may have not been created if the merge is not completed.
// Therefore, it is not in meta at master recovery time.
hri = regionStates.getRegionInfo(rt.getRegionName());
EventType et = rt.getEventType();
if (hri == null && et != EventType.RS_ZK_REGION_MERGING
&& et != EventType.RS_ZK_REQUEST_REGION_MERGE) {
LOG.warn("Couldn't find the region in recovering " + rt);
return false;
}
}
return processRegionsInTransition(
rt, hri, stat.getVersion());
} finally {
lock.unlock();
}
}
/**
* This call is invoked only (1) master assign meta;
* (2) during failover mode startup, zk assignment node processing.
* The locker is set in the caller. It returns true if the region
* is in transition for sure, false otherwise.
*
* It should be private but it is used by some test too.
*/
boolean processRegionsInTransition(
final RegionTransition rt, final HRegionInfo regionInfo,
final int expectedVersion) throws KeeperException {
EventType et = rt.getEventType();
// Get ServerName. Could not be null.
final ServerName sn = rt.getServerName();
final byte[] regionName = rt.getRegionName();
final String encodedName = HRegionInfo.encodeRegionName(regionName);
final String prettyPrintedRegionName = HRegionInfo.prettyPrint(encodedName);
LOG.info("Processing " + prettyPrintedRegionName + " in state: " + et);
if (regionStates.isRegionInTransition(encodedName)) {
LOG.info("Processed region " + prettyPrintedRegionName + " in state: "
+ et + ", does nothing since the region is already in transition "
+ regionStates.getRegionTransitionState(encodedName));
// Just return
return true;
}
if (!serverManager.isServerOnline(sn)) {
// It was transitioning on a dead server, so it's closed now.
// Force to OFFLINE and put it in transition, but not assign it
// since log splitting for the dead server is not done yet.
LOG.debug("RIT " + encodedName + " in state=" + rt.getEventType() +
" was on deadserver; forcing offline");
if (regionStates.isRegionOnline(regionInfo)) {
// Meta could still show the region is assigned to the previous
// server. If that server is online, when we reload the meta, the
// region is put back to online, we need to offline it.
regionStates.regionOffline(regionInfo);
}
// Put it back in transition so that SSH can re-assign it
regionStates.updateRegionState(regionInfo, State.OFFLINE, sn);
if (regionInfo.isMetaRegion()) {
// If it's meta region, reset the meta location.
// So that master knows the right meta region server.
MetaRegionTracker.setMetaLocation(watcher, sn);
} else {
// No matter the previous server is online or offline,
// we need to reset the last region server of the region.
regionStates.setLastRegionServerOfRegion(sn, encodedName);
// Make sure we know the server is dead.
if (!serverManager.isServerDead(sn)) {
serverManager.expireServer(sn);
}
}
return false;
}
switch (et) {
case M_ZK_REGION_CLOSING:
// Insert into RIT & resend the query to the region server: may be the previous master
// died before sending the query the first time.
final RegionState rsClosing = regionStates.updateRegionState(rt, State.CLOSING);
this.executorService.submit(
new EventHandler(server, EventType.M_MASTER_RECOVERY) {
@Override
public void process() throws IOException {
ReentrantLock lock = locker.acquireLock(regionInfo.getEncodedName());
try {
unassign(regionInfo, rsClosing, expectedVersion, null, true, null);
if (regionStates.isRegionOffline(regionInfo)) {
assign(regionInfo, true);
}
} finally {
lock.unlock();
}
}
});
break;
case RS_ZK_REGION_CLOSED:
case RS_ZK_REGION_FAILED_OPEN:
// Region is closed, insert into RIT and handle it
regionStates.updateRegionState(regionInfo, State.CLOSED, sn);
invokeAssign(regionInfo);
break;
case M_ZK_REGION_OFFLINE:
// Insert in RIT and resend to the regionserver
regionStates.updateRegionState(rt, State.PENDING_OPEN);
final RegionState rsOffline = regionStates.getRegionState(regionInfo);
this.executorService.submit(
new EventHandler(server, EventType.M_MASTER_RECOVERY) {
@Override
public void process() throws IOException {
ReentrantLock lock = locker.acquireLock(regionInfo.getEncodedName());
try {
RegionPlan plan = new RegionPlan(regionInfo, null, sn);
addPlan(encodedName, plan);
assign(rsOffline, false, false);
} finally {
lock.unlock();
}
}
});
break;
case RS_ZK_REGION_OPENING:
regionStates.updateRegionState(rt, State.OPENING);
break;
case RS_ZK_REGION_OPENED:
// Region is opened, insert into RIT and handle it
// This could be done asynchronously, we would need then to acquire the lock in the
// handler.
regionStates.updateRegionState(rt, State.OPEN);
new OpenedRegionHandler(server, this, regionInfo, sn, expectedVersion).process();
break;
case RS_ZK_REQUEST_REGION_SPLIT:
case RS_ZK_REGION_SPLITTING:
case RS_ZK_REGION_SPLIT:
// Splitting region should be online. We could have skipped it during
// user region rebuilding since we may consider the split is completed.
// Put it in SPLITTING state to avoid complications.
regionStates.regionOnline(regionInfo, sn);
regionStates.updateRegionState(rt, State.SPLITTING);
if (!handleRegionSplitting(
rt, encodedName, prettyPrintedRegionName, sn)) {
deleteSplittingNode(encodedName, sn);
}
break;
case RS_ZK_REQUEST_REGION_MERGE:
case RS_ZK_REGION_MERGING:
case RS_ZK_REGION_MERGED:
if (!handleRegionMerging(
rt, encodedName, prettyPrintedRegionName, sn)) {
deleteMergingNode(encodedName, sn);
}
break;
default:
throw new IllegalStateException("Received region in state:" + et + " is not valid.");
}
LOG.info("Processed region " + prettyPrintedRegionName + " in state "
+ et + ", on " + (serverManager.isServerOnline(sn) ? "" : "dead ")
+ "server: " + sn);
return true;
}
/**
* When a region is closed, it should be removed from the regionsToReopen
* @param hri HRegionInfo of the region which was closed
*/
public void removeClosedRegion(HRegionInfo hri) {
if (regionsToReopen.remove(hri.getEncodedName()) != null) {
LOG.debug("Removed region from reopening regions because it was closed");
}
}
/**
* Handles various states an unassigned node can be in.
* <p>
* Method is called when a state change is suspected for an unassigned node.
* <p>
* This deals with skipped transitions (we got a CLOSED but didn't see CLOSING
* yet).
* @param rt
* @param expectedVersion
*/
void handleRegion(final RegionTransition rt, int expectedVersion) {
if (rt == null) {
LOG.warn("Unexpected NULL input for RegionTransition rt");
return;
}
final ServerName sn = rt.getServerName();
// Check if this is a special HBCK transition
if (sn.equals(HBCK_CODE_SERVERNAME)) {
handleHBCK(rt);
return;
}
final long createTime = rt.getCreateTime();
final byte[] regionName = rt.getRegionName();
String encodedName = HRegionInfo.encodeRegionName(regionName);
String prettyPrintedRegionName = HRegionInfo.prettyPrint(encodedName);
// Verify this is a known server
if (!serverManager.isServerOnline(sn)
&& !ignoreStatesRSOffline.contains(rt.getEventType())) {
LOG.warn("Attempted to handle region transition for server but " +
"it is not online: " + prettyPrintedRegionName + ", " + rt);
return;
}
RegionState regionState =
regionStates.getRegionState(encodedName);
long startTime = System.currentTimeMillis();
if (LOG.isDebugEnabled()) {
boolean lateEvent = createTime < (startTime - 15000);
LOG.debug("Handling " + rt.getEventType() +
", server=" + sn + ", region=" +
(prettyPrintedRegionName == null ? "null" : prettyPrintedRegionName) +
(lateEvent ? ", which is more than 15 seconds late" : "") +
", current_state=" + regionState);
}
// We don't do anything for this event,
// so separate it out, no need to lock/unlock anything
if (rt.getEventType() == EventType.M_ZK_REGION_OFFLINE) {
return;
}
// We need a lock on the region as we could update it
Lock lock = locker.acquireLock(encodedName);
try {
RegionState latestState =
regionStates.getRegionState(encodedName);
if ((regionState == null && latestState != null)
|| (regionState != null && latestState == null)
|| (regionState != null && latestState != null
&& latestState.getState() != regionState.getState())) {
LOG.warn("Region state changed from " + regionState + " to "
+ latestState + ", while acquiring lock");
}
long waitedTime = System.currentTimeMillis() - startTime;
if (waitedTime > 5000) {
LOG.warn("Took " + waitedTime + "ms to acquire the lock");
}
regionState = latestState;
switch (rt.getEventType()) {
case RS_ZK_REQUEST_REGION_SPLIT:
case RS_ZK_REGION_SPLITTING:
case RS_ZK_REGION_SPLIT:
if (!handleRegionSplitting(
rt, encodedName, prettyPrintedRegionName, sn)) {
deleteSplittingNode(encodedName, sn);
}
break;
case RS_ZK_REQUEST_REGION_MERGE:
case RS_ZK_REGION_MERGING:
case RS_ZK_REGION_MERGED:
// Merged region is a new region, we can't find it in the region states now.
// However, the two merging regions are not new. They should be in state for merging.
if (!handleRegionMerging(
rt, encodedName, prettyPrintedRegionName, sn)) {
deleteMergingNode(encodedName, sn);
}
break;
case M_ZK_REGION_CLOSING:
// Should see CLOSING after we have asked it to CLOSE or additional
// times after already being in state of CLOSING
if (regionState == null
|| !regionState.isPendingCloseOrClosingOnServer(sn)) {
LOG.warn("Received CLOSING for " + prettyPrintedRegionName
+ " from " + sn + " but the region isn't PENDING_CLOSE/CLOSING here: "
+ regionStates.getRegionState(encodedName));
return;
}
// Transition to CLOSING (or update stamp if already CLOSING)
regionStates.updateRegionState(rt, State.CLOSING);
break;
case RS_ZK_REGION_CLOSED:
// Should see CLOSED after CLOSING but possible after PENDING_CLOSE
if (regionState == null
|| !regionState.isPendingCloseOrClosingOnServer(sn)) {
LOG.warn("Received CLOSED for " + prettyPrintedRegionName
+ " from " + sn + " but the region isn't PENDING_CLOSE/CLOSING here: "
+ regionStates.getRegionState(encodedName));
return;
}
// Handle CLOSED by assigning elsewhere or stopping if a disable
// If we got here all is good. Need to update RegionState -- else
// what follows will fail because not in expected state.
new ClosedRegionHandler(server, this, regionState.getRegion()).process();
updateClosedRegionHandlerTracker(regionState.getRegion());
break;
case RS_ZK_REGION_FAILED_OPEN:
if (regionState == null
|| !regionState.isPendingOpenOrOpeningOnServer(sn)) {
LOG.warn("Received FAILED_OPEN for " + prettyPrintedRegionName
+ " from " + sn + " but the region isn't PENDING_OPEN/OPENING here: "
+ regionStates.getRegionState(encodedName));
return;
}
AtomicInteger failedOpenCount = failedOpenTracker.get(encodedName);
if (failedOpenCount == null) {
failedOpenCount = new AtomicInteger();
// No need to use putIfAbsent, or extra synchronization since
// this whole handleRegion block is locked on the encoded region
// name, and failedOpenTracker is updated only in this block
failedOpenTracker.put(encodedName, failedOpenCount);
}
if (failedOpenCount.incrementAndGet() >= maximumAttempts) {
regionStates.updateRegionState(rt, State.FAILED_OPEN);
// remove the tracking info to save memory, also reset
// the count for next open initiative
failedOpenTracker.remove(encodedName);
} else {
// Handle this the same as if it were opened and then closed.
regionState = regionStates.updateRegionState(rt, State.CLOSED);
if (regionState != null) {
// When there are more than one region server a new RS is selected as the
// destination and the same is updated in the regionplan. (HBASE-5546)
try {
getRegionPlan(regionState.getRegion(), sn, true);
new ClosedRegionHandler(server, this, regionState.getRegion()).process();
} catch (HBaseIOException e) {
LOG.warn("Failed to get region plan", e);
}
}
}
break;
case RS_ZK_REGION_OPENING:
// Should see OPENING after we have asked it to OPEN or additional
// times after already being in state of OPENING
if (regionState == null
|| !regionState.isPendingOpenOrOpeningOnServer(sn)) {
LOG.warn("Received OPENING for " + prettyPrintedRegionName
+ " from " + sn + " but the region isn't PENDING_OPEN/OPENING here: "
+ regionStates.getRegionState(encodedName));
return;
}
// Transition to OPENING (or update stamp if already OPENING)
regionStates.updateRegionState(rt, State.OPENING);
break;
case RS_ZK_REGION_OPENED:
// Should see OPENED after OPENING but possible after PENDING_OPEN.
if (regionState == null
|| !regionState.isPendingOpenOrOpeningOnServer(sn)) {
LOG.warn("Received OPENED for " + prettyPrintedRegionName
+ " from " + sn + " but the region isn't PENDING_OPEN/OPENING here: "
+ regionStates.getRegionState(encodedName));
if (regionState != null) {
// Close it without updating the internal region states,
// so as not to create double assignments in unlucky scenarios
// mentioned in OpenRegionHandler#process
unassign(regionState.getRegion(), null, -1, null, false, sn);
}
return;
}
// Handle OPENED by removing from transition and deleted zk node
regionState = regionStates.updateRegionState(rt, State.OPEN);
if (regionState != null) {
failedOpenTracker.remove(encodedName); // reset the count, if any
new OpenedRegionHandler(
server, this, regionState.getRegion(), sn, expectedVersion).process();
updateOpenedRegionHandlerTracker(regionState.getRegion());
}
break;
default:
throw new IllegalStateException("Received event is not valid.");
}
} finally {
lock.unlock();
}
}
//For unit tests only
boolean wasClosedHandlerCalled(HRegionInfo hri) {
AtomicBoolean b = closedRegionHandlerCalled.get(hri);
//compareAndSet to be sure that unit tests don't see stale values. Means,
//we will return true exactly once unless the handler code resets to true
//this value.
return b == null ? false : b.compareAndSet(true, false);
}
//For unit tests only
boolean wasOpenedHandlerCalled(HRegionInfo hri) {
AtomicBoolean b = openedRegionHandlerCalled.get(hri);
//compareAndSet to be sure that unit tests don't see stale values. Means,
//we will return true exactly once unless the handler code resets to true
//this value.
return b == null ? false : b.compareAndSet(true, false);
}
//For unit tests only
void initializeHandlerTrackers() {
closedRegionHandlerCalled = new HashMap<HRegionInfo, AtomicBoolean>();
openedRegionHandlerCalled = new HashMap<HRegionInfo, AtomicBoolean>();
}
void updateClosedRegionHandlerTracker(HRegionInfo hri) {
if (closedRegionHandlerCalled != null) { //only for unit tests this is true
closedRegionHandlerCalled.put(hri, new AtomicBoolean(true));
}
}
void updateOpenedRegionHandlerTracker(HRegionInfo hri) {
if (openedRegionHandlerCalled != null) { //only for unit tests this is true
openedRegionHandlerCalled.put(hri, new AtomicBoolean(true));
}
}
// TODO: processFavoredNodes might throw an exception, for e.g., if the
// meta could not be contacted/updated. We need to see how seriously to treat
// this problem as. Should we fail the current assignment. We should be able
// to recover from this problem eventually (if the meta couldn't be updated
// things should work normally and eventually get fixed up).
void processFavoredNodes(List<HRegionInfo> regions) throws IOException {
if (!shouldAssignRegionsWithFavoredNodes) return;
// The AM gets the favored nodes info for each region and updates the meta
// table with that info
Map<HRegionInfo, List<ServerName>> regionToFavoredNodes =
new HashMap<HRegionInfo, List<ServerName>>();
for (HRegionInfo region : regions) {
regionToFavoredNodes.put(region,
((FavoredNodeLoadBalancer)this.balancer).getFavoredNodes(region));
}
FavoredNodeAssignmentHelper.updateMetaWithFavoredNodesInfo(regionToFavoredNodes, catalogTracker);
}
/**
* Handle a ZK unassigned node transition triggered by HBCK repair tool.
* <p>
* This is handled in a separate code path because it breaks the normal rules.
* @param rt
*/
private void handleHBCK(RegionTransition rt) {
String encodedName = HRegionInfo.encodeRegionName(rt.getRegionName());
LOG.info("Handling HBCK triggered transition=" + rt.getEventType() +
", server=" + rt.getServerName() + ", region=" +
HRegionInfo.prettyPrint(encodedName));
RegionState regionState = regionStates.getRegionTransitionState(encodedName);
switch (rt.getEventType()) {
case M_ZK_REGION_OFFLINE:
HRegionInfo regionInfo;
if (regionState != null) {
regionInfo = regionState.getRegion();
} else {
try {
byte [] name = rt.getRegionName();
Pair<HRegionInfo, ServerName> p = MetaReader.getRegion(catalogTracker, name);
regionInfo = p.getFirst();
} catch (IOException e) {
LOG.info("Exception reading hbase:meta doing HBCK repair operation", e);
return;
}
}
LOG.info("HBCK repair is triggering assignment of region=" +
regionInfo.getRegionNameAsString());
// trigger assign, node is already in OFFLINE so don't need to update ZK
assign(regionInfo, false);
break;
default:
LOG.warn("Received unexpected region state from HBCK: " + rt.toString());
break;
}
}
// ZooKeeper events
/**
* New unassigned node has been created.
*
* <p>This happens when an RS begins the OPENING or CLOSING of a region by
* creating an unassigned node.
*
* <p>When this happens we must:
* <ol>
* <li>Watch the node for further events</li>
* <li>Read and handle the state in the node</li>
* </ol>
*/
@Override
public void nodeCreated(String path) {
handleAssignmentEvent(path);
}
/**
* Existing unassigned node has had data changed.
*
* <p>This happens when an RS transitions from OFFLINE to OPENING, or between
* OPENING/OPENED and CLOSING/CLOSED.
*
* <p>When this happens we must:
* <ol>
* <li>Watch the node for further events</li>
* <li>Read and handle the state in the node</li>
* </ol>
*/
@Override
public void nodeDataChanged(String path) {
handleAssignmentEvent(path);
}
// We don't want to have two events on the same region managed simultaneously.
// For this reason, we need to wait if an event on the same region is currently in progress.
// So we track the region names of the events in progress, and we keep a waiting list.
private final Set<String> regionsInProgress = new HashSet<String>();
// In a LinkedHashMultimap, the put order is kept when we retrieve the collection back. We need
// this as we want the events to be managed in the same order as we received them.
private final LinkedHashMultimap <String, RegionRunnable>
zkEventWorkerWaitingList = LinkedHashMultimap.create();
/**
* A specific runnable that works only on a region.
*/
private interface RegionRunnable extends Runnable{
/**
* @return - the name of the region it works on.
*/
String getRegionName();
}
/**
* Submit a task, ensuring that there is only one task at a time that working on a given region.
* Order is respected.
*/
protected void zkEventWorkersSubmit(final RegionRunnable regRunnable) {
synchronized (regionsInProgress) {
// If we're there is already a task with this region, we add it to the
// waiting list and return.
if (regionsInProgress.contains(regRunnable.getRegionName())) {
synchronized (zkEventWorkerWaitingList){
zkEventWorkerWaitingList.put(regRunnable.getRegionName(), regRunnable);
}
return;
}
// No event in progress on this region => we can submit a new task immediately.
regionsInProgress.add(regRunnable.getRegionName());
zkEventWorkers.submit(new Runnable() {
@Override
public void run() {
try {
regRunnable.run();
} finally {
// now that we have finished, let's see if there is an event for the same region in the
// waiting list. If it's the case, we can now submit it to the pool.
synchronized (regionsInProgress) {
regionsInProgress.remove(regRunnable.getRegionName());
synchronized (zkEventWorkerWaitingList) {
java.util.Set<RegionRunnable> waiting = zkEventWorkerWaitingList.get(
regRunnable.getRegionName());
if (!waiting.isEmpty()) {
// We want the first object only. The only way to get it is through an iterator.
RegionRunnable toSubmit = waiting.iterator().next();
zkEventWorkerWaitingList.remove(toSubmit.getRegionName(), toSubmit);
zkEventWorkersSubmit(toSubmit);
}
}
}
}
}
});
}
}
@Override
public void nodeDeleted(final String path) {
if (path.startsWith(watcher.assignmentZNode)) {
final String regionName = ZKAssign.getRegionName(watcher, path);
zkEventWorkersSubmit(new RegionRunnable() {
@Override
public String getRegionName() {
return regionName;
}
@Override
public void run() {
Lock lock = locker.acquireLock(regionName);
try {
RegionState rs = regionStates.getRegionTransitionState(regionName);
if (rs == null) {
rs = regionStates.getRegionState(regionName);
if (rs == null || !rs.isMergingNew()) {
// MergingNew is an offline state
return;
}
}
HRegionInfo regionInfo = rs.getRegion();
String regionNameStr = regionInfo.getRegionNameAsString();
LOG.debug("Znode " + regionNameStr + " deleted, state: " + rs);
boolean disabled = getZKTable().isDisablingOrDisabledTable(regionInfo.getTable());
ServerName serverName = rs.getServerName();
if (serverManager.isServerOnline(serverName)) {
if (rs.isOnServer(serverName)
&& (rs.isOpened() || rs.isSplitting())) {
regionOnline(regionInfo, serverName);
if (disabled) {
// if server is offline, no hurt to unassign again
LOG.info("Opened " + regionNameStr
+ "but this table is disabled, triggering close of region");
unassign(regionInfo);
}
} else if (rs.isMergingNew()) {
synchronized (regionStates) {
String p = regionInfo.getEncodedName();
PairOfSameType<HRegionInfo> regions = mergingRegions.get(p);
if (regions != null) {
onlineMergingRegion(disabled, regions.getFirst(), serverName);
onlineMergingRegion(disabled, regions.getSecond(), serverName);
}
}
}
}
} finally {
lock.unlock();
}
}
private void onlineMergingRegion(boolean disabled,
final HRegionInfo hri, final ServerName serverName) {
RegionState regionState = regionStates.getRegionState(hri);
if (regionState != null && regionState.isMerging()
&& regionState.isOnServer(serverName)) {
regionOnline(regionState.getRegion(), serverName);
if (disabled) {
unassign(hri);
}
}
}
});
}
}
/**
* New unassigned node has been created.
*
* <p>This happens when an RS begins the OPENING, SPLITTING or CLOSING of a
* region by creating a znode.
*
* <p>When this happens we must:
* <ol>
* <li>Watch the node for further children changed events</li>
* <li>Watch all new children for changed events</li>
* </ol>
*/
@Override
public void nodeChildrenChanged(String path) {
if (path.equals(watcher.assignmentZNode)) {
zkEventWorkers.submit(new Runnable() {
@Override
public void run() {
try {
// Just make sure we see the changes for the new znodes
List<String> children =
ZKUtil.listChildrenAndWatchForNewChildren(
watcher, watcher.assignmentZNode);
if (children != null) {
Stat stat = new Stat();
for (String child : children) {
// if region is in transition, we already have a watch
// on it, so no need to watch it again. So, as I know for now,
// this is needed to watch splitting nodes only.
if (!regionStates.isRegionInTransition(child)) {
ZKAssign.getDataAndWatch(watcher, child, stat);
}
}
}
} catch (KeeperException e) {
server.abort("Unexpected ZK exception reading unassigned children", e);
}
}
});
}
}
/**
* Marks the region as online. Removes it from regions in transition and
* updates the in-memory assignment information.
* <p>
* Used when a region has been successfully opened on a region server.
* @param regionInfo
* @param sn
*/
void regionOnline(HRegionInfo regionInfo, ServerName sn) {
numRegionsOpened.incrementAndGet();
regionStates.regionOnline(regionInfo, sn);
// Remove plan if one.
clearRegionPlan(regionInfo);
// Add the server to serversInUpdatingTimer
addToServersInUpdatingTimer(sn);
}
/**
* Pass the assignment event to a worker for processing.
* Each worker is a single thread executor service. The reason
* for just one thread is to make sure all events for a given
* region are processed in order.
*
* @param path
*/
private void handleAssignmentEvent(final String path) {
if (path.startsWith(watcher.assignmentZNode)) {
final String regionName = ZKAssign.getRegionName(watcher, path);
zkEventWorkersSubmit(new RegionRunnable() {
@Override
public String getRegionName() {
return regionName;
}
@Override
public void run() {
try {
Stat stat = new Stat();
byte [] data = ZKAssign.getDataAndWatch(watcher, path, stat);
if (data == null) return;
RegionTransition rt = RegionTransition.parseFrom(data);
handleRegion(rt, stat.getVersion());
} catch (KeeperException e) {
server.abort("Unexpected ZK exception reading unassigned node data", e);
} catch (DeserializationException e) {
server.abort("Unexpected exception deserializing node data", e);
}
}
});
}
}
/**
* Add the server to the set serversInUpdatingTimer, then {@link TimerUpdater}
* will update timers for this server in background
* @param sn
*/
private void addToServersInUpdatingTimer(final ServerName sn) {
if (tomActivated){
this.serversInUpdatingTimer.add(sn);
}
}
/**
* Touch timers for all regions in transition that have the passed
* <code>sn</code> in common.
* Call this method whenever a server checks in. Doing so helps the case where
* a new regionserver has joined the cluster and its been given 1k regions to
* open. If this method is tickled every time the region reports in a
* successful open then the 1k-th region won't be timed out just because its
* sitting behind the open of 999 other regions. This method is NOT used
* as part of bulk assign -- there we have a different mechanism for extending
* the regions in transition timer (we turn it off temporarily -- because
* there is no regionplan involved when bulk assigning.
* @param sn
*/
private void updateTimers(final ServerName sn) {
Preconditions.checkState(tomActivated);
if (sn == null) return;
// This loop could be expensive.
// First make a copy of current regionPlan rather than hold sync while
// looping because holding sync can cause deadlock. Its ok in this loop
// if the Map we're going against is a little stale
List<Map.Entry<String, RegionPlan>> rps;
synchronized(this.regionPlans) {
rps = new ArrayList<Map.Entry<String, RegionPlan>>(regionPlans.entrySet());
}
for (Map.Entry<String, RegionPlan> e : rps) {
if (e.getValue() != null && e.getKey() != null && sn.equals(e.getValue().getDestination())) {
RegionState regionState = regionStates.getRegionTransitionState(e.getKey());
if (regionState != null) {
regionState.updateTimestampToNow();
}
}
}
}
/**
* Marks the region as offline. Removes it from regions in transition and
* removes in-memory assignment information.
* <p>
* Used when a region has been closed and should remain closed.
* @param regionInfo
*/
public void regionOffline(final HRegionInfo regionInfo) {
regionOffline(regionInfo, null);
}
public void offlineDisabledRegion(HRegionInfo regionInfo) {
// Disabling so should not be reassigned, just delete the CLOSED node
LOG.debug("Table being disabled so deleting ZK node and removing from " +
"regions in transition, skipping assignment of region " +
regionInfo.getRegionNameAsString());
String encodedName = regionInfo.getEncodedName();
deleteNodeInStates(encodedName, "closed", null,
EventType.RS_ZK_REGION_CLOSED, EventType.M_ZK_REGION_OFFLINE);
regionOffline(regionInfo);
}
// Assignment methods
/**
* Assigns the specified region.
* <p>
* If a RegionPlan is available with a valid destination then it will be used
* to determine what server region is assigned to. If no RegionPlan is
* available, region will be assigned to a random available server.
* <p>
* Updates the RegionState and sends the OPEN RPC.
* <p>
* This will only succeed if the region is in transition and in a CLOSED or
* OFFLINE state or not in transition (in-memory not zk), and of course, the
* chosen server is up and running (It may have just crashed!). If the
* in-memory checks pass, the zk node is forced to OFFLINE before assigning.
*
* @param region server to be assigned
* @param setOfflineInZK whether ZK node should be created/transitioned to an
* OFFLINE state before assigning the region
*/
public void assign(HRegionInfo region, boolean setOfflineInZK) {
assign(region, setOfflineInZK, false);
}
/**
* Use care with forceNewPlan. It could cause double assignment.
*/
public void assign(HRegionInfo region,
boolean setOfflineInZK, boolean forceNewPlan) {
if (isDisabledorDisablingRegionInRIT(region)) {
return;
}
if (this.serverManager.isClusterShutdown()) {
LOG.info("Cluster shutdown is set; skipping assign of " +
region.getRegionNameAsString());
return;
}
String encodedName = region.getEncodedName();
Lock lock = locker.acquireLock(encodedName);
try {
RegionState state = forceRegionStateToOffline(region, forceNewPlan);
if (state != null) {
if (regionStates.wasRegionOnDeadServer(encodedName)) {
LOG.info("Skip assigning " + region.getRegionNameAsString()
+ ", it's host " + regionStates.getLastRegionServerOfRegion(encodedName)
+ " is dead but not processed yet");
return;
}
assign(state, setOfflineInZK, forceNewPlan);
}
} finally {
lock.unlock();
}
}
/**
* Bulk assign regions to <code>destination</code>.
* @param destination
* @param regions Regions to assign.
* @return true if successful
*/
boolean assign(final ServerName destination, final List<HRegionInfo> regions) {
long startTime = EnvironmentEdgeManager.currentTimeMillis();
try {
int regionCount = regions.size();
if (regionCount == 0) {
return true;
}
LOG.debug("Assigning " + regionCount + " region(s) to " + destination.toString());
Set<String> encodedNames = new HashSet<String>(regionCount);
for (HRegionInfo region : regions) {
encodedNames.add(region.getEncodedName());
}
List<HRegionInfo> failedToOpenRegions = new ArrayList<HRegionInfo>();
Map<String, Lock> locks = locker.acquireLocks(encodedNames);
try {
AtomicInteger counter = new AtomicInteger(0);
Map<String, Integer> offlineNodesVersions = new ConcurrentHashMap<String, Integer>();
OfflineCallback cb = new OfflineCallback(
watcher, destination, counter, offlineNodesVersions);
Map<String, RegionPlan> plans = new HashMap<String, RegionPlan>(regions.size());
List<RegionState> states = new ArrayList<RegionState>(regions.size());
for (HRegionInfo region : regions) {
String encodedName = region.getEncodedName();
if (!isDisabledorDisablingRegionInRIT(region)) {
RegionState state = forceRegionStateToOffline(region, false);
boolean onDeadServer = false;
if (state != null) {
if (regionStates.wasRegionOnDeadServer(encodedName)) {
LOG.info("Skip assigning " + region.getRegionNameAsString()
+ ", it's host " + regionStates.getLastRegionServerOfRegion(encodedName)
+ " is dead but not processed yet");
onDeadServer = true;
} else if (asyncSetOfflineInZooKeeper(state, cb, destination)) {
RegionPlan plan = new RegionPlan(region, state.getServerName(), destination);
plans.put(encodedName, plan);
states.add(state);
continue;
}
}
// Reassign if the region wasn't on a dead server
if (!onDeadServer) {
LOG.info("failed to force region state to offline or "
+ "failed to set it offline in ZK, will reassign later: " + region);
failedToOpenRegions.add(region); // assign individually later
}
}
// Release the lock, this region is excluded from bulk assign because
// we can't update its state, or set its znode to offline.
Lock lock = locks.remove(encodedName);
lock.unlock();
}
// Wait until all unassigned nodes have been put up and watchers set.
int total = states.size();
for (int oldCounter = 0; !server.isStopped();) {
int count = counter.get();
if (oldCounter != count) {
LOG.info(destination.toString() + " unassigned znodes=" + count +
" of total=" + total);
oldCounter = count;
}
if (count >= total) break;
Threads.sleep(5);
}
if (server.isStopped()) {
return false;
}
// Add region plans, so we can updateTimers when one region is opened so
// that unnecessary timeout on RIT is reduced.
this.addPlans(plans);
List<Triple<HRegionInfo, Integer, List<ServerName>>> regionOpenInfos =
new ArrayList<Triple<HRegionInfo, Integer, List<ServerName>>>(states.size());
for (RegionState state: states) {
HRegionInfo region = state.getRegion();
String encodedRegionName = region.getEncodedName();
Integer nodeVersion = offlineNodesVersions.get(encodedRegionName);
if (nodeVersion == null || nodeVersion == -1) {
LOG.warn("failed to offline in zookeeper: " + region);
failedToOpenRegions.add(region); // assign individually later
Lock lock = locks.remove(encodedRegionName);
lock.unlock();
} else {
regionStates.updateRegionState(
region, State.PENDING_OPEN, destination);
List<ServerName> favoredNodes = ServerName.EMPTY_SERVER_LIST;
if (this.shouldAssignRegionsWithFavoredNodes) {
favoredNodes = ((FavoredNodeLoadBalancer)this.balancer).getFavoredNodes(region);
}
regionOpenInfos.add(new Triple<HRegionInfo, Integer, List<ServerName>>(
region, nodeVersion, favoredNodes));
}
}
// Move on to open regions.
try {
// Send OPEN RPC. If it fails on a IOE or RemoteException,
// regions will be assigned individually.
long maxWaitTime = System.currentTimeMillis() +
this.server.getConfiguration().
getLong("hbase.regionserver.rpc.startup.waittime", 60000);
for (int i = 1; i <= maximumAttempts && !server.isStopped(); i++) {
try {
List<RegionOpeningState> regionOpeningStateList = serverManager
.sendRegionOpen(destination, regionOpenInfos);
if (regionOpeningStateList == null) {
// Failed getting RPC connection to this server
return false;
}
for (int k = 0, n = regionOpeningStateList.size(); k < n; k++) {
RegionOpeningState openingState = regionOpeningStateList.get(k);
if (openingState != RegionOpeningState.OPENED) {
HRegionInfo region = regionOpenInfos.get(k).getFirst();
if (openingState == RegionOpeningState.ALREADY_OPENED) {
processAlreadyOpenedRegion(region, destination);
} else if (openingState == RegionOpeningState.FAILED_OPENING) {
// Failed opening this region, reassign it later
failedToOpenRegions.add(region);
} else {
LOG.warn("THIS SHOULD NOT HAPPEN: unknown opening state "
+ openingState + " in assigning region " + region);
}
}
}
break;
} catch (IOException e) {
if (e instanceof RemoteException) {
e = ((RemoteException)e).unwrapRemoteException();
}
if (e instanceof RegionServerStoppedException) {
LOG.warn("The region server was shut down, ", e);
// No need to retry, the region server is a goner.
return false;
} else if (e instanceof ServerNotRunningYetException) {
long now = System.currentTimeMillis();
if (now < maxWaitTime) {
LOG.debug("Server is not yet up; waiting up to " +
(maxWaitTime - now) + "ms", e);
Thread.sleep(100);
i--; // reset the try count
continue;
}
} else if (e instanceof java.net.SocketTimeoutException
&& this.serverManager.isServerOnline(destination)) {
// In case socket is timed out and the region server is still online,
// the openRegion RPC could have been accepted by the server and
// just the response didn't go through. So we will retry to
// open the region on the same server.
if (LOG.isDebugEnabled()) {
LOG.debug("Bulk assigner openRegion() to " + destination
+ " has timed out, but the regions might"
+ " already be opened on it.", e);
}
continue;
}
throw e;
}
}
} catch (IOException e) {
// Can be a socket timeout, EOF, NoRouteToHost, etc
LOG.info("Unable to communicate with " + destination
+ " in order to assign regions, ", e);
return false;
} catch (InterruptedException e) {
throw new RuntimeException(e);
}
} finally {
for (Lock lock : locks.values()) {
lock.unlock();
}
}
if (!failedToOpenRegions.isEmpty()) {
for (HRegionInfo region : failedToOpenRegions) {
if (!regionStates.isRegionOnline(region)) {
invokeAssign(region);
}
}
}
LOG.debug("Bulk assigning done for " + destination);
return true;
} finally {
metricsAssignmentManager.updateBulkAssignTime(EnvironmentEdgeManager.currentTimeMillis() - startTime);
}
}
/**
* Send CLOSE RPC if the server is online, otherwise, offline the region.
*
* The RPC will be sent only to the region sever found in the region state
* if it is passed in, otherwise, to the src server specified. If region
* state is not specified, we don't update region state at all, instead
* we just send the RPC call. This is useful for some cleanup without
* messing around the region states (see handleRegion, on region opened
* on an unexpected server scenario, for an example)
*/
private void unassign(final HRegionInfo region,
final RegionState state, final int versionOfClosingNode,
final ServerName dest, final boolean transitionInZK,
final ServerName src) {
ServerName server = src;
if (state != null) {
server = state.getServerName();
}
long maxWaitTime = -1;
for (int i = 1; i <= this.maximumAttempts; i++) {
if (this.server.isStopped() || this.server.isAborted()) {
LOG.debug("Server stopped/aborted; skipping unassign of " + region);
return;
}
// ClosedRegionhandler can remove the server from this.regions
if (!serverManager.isServerOnline(server)) {
LOG.debug("Offline " + region.getRegionNameAsString()
+ ", no need to unassign since it's on a dead server: " + server);
if (transitionInZK) {
// delete the node. if no node exists need not bother.
deleteClosingOrClosedNode(region, server);
}
if (state != null) {
regionOffline(region);
}
return;
}
try {
// Send CLOSE RPC
if (serverManager.sendRegionClose(server, region,
versionOfClosingNode, dest, transitionInZK)) {
LOG.debug("Sent CLOSE to " + server + " for region " +
region.getRegionNameAsString());
if (!transitionInZK && state != null) {
// Retry to make sure the region is
// closed so as to avoid double assignment.
unassign(region, state, versionOfClosingNode,
dest, transitionInZK,src);
}
return;
}
// This never happens. Currently regionserver close always return true.
// Todo; this can now happen (0.96) if there is an exception in a coprocessor
LOG.warn("Server " + server + " region CLOSE RPC returned false for " +
region.getRegionNameAsString());
} catch (Throwable t) {
if (t instanceof RemoteException) {
t = ((RemoteException)t).unwrapRemoteException();
}
if (t instanceof NotServingRegionException
|| t instanceof RegionServerStoppedException) {
LOG.debug("Offline " + region.getRegionNameAsString()
+ ", it's not any more on " + server, t);
if (transitionInZK) {
deleteClosingOrClosedNode(region, server);
}
if (state != null) {
regionOffline(region);
}
return;
} else if (state != null
&& t instanceof RegionAlreadyInTransitionException) {
// RS is already processing this region, only need to update the timestamp
LOG.debug("update " + state + " the timestamp.");
state.updateTimestampToNow();
if (maxWaitTime < 0) {
maxWaitTime = EnvironmentEdgeManager.currentTimeMillis()
+ this.server.getConfiguration().getLong(ALREADY_IN_TRANSITION_WAITTIME,
DEFAULT_ALREADY_IN_TRANSITION_WAITTIME);
}
try {
long now = EnvironmentEdgeManager.currentTimeMillis();
if (now < maxWaitTime) {
LOG.debug("Region is already in transition; "
+ "waiting up to " + (maxWaitTime - now) + "ms", t);
Thread.sleep(100);
i--; // reset the try count
}
} catch (InterruptedException ie) {
LOG.warn("Failed to unassign "
+ region.getRegionNameAsString() + " since interrupted", ie);
Thread.currentThread().interrupt();
if (!tomActivated) {
regionStates.updateRegionState(region, State.FAILED_CLOSE);
}
return;
}
} else {
LOG.info("Server " + server + " returned " + t + " for "
+ region.getRegionNameAsString() + ", try=" + i
+ " of " + this.maximumAttempts, t);
// Presume retry or server will expire.
}
}
}
// Run out of attempts
if (!tomActivated && state != null) {
regionStates.updateRegionState(region, State.FAILED_CLOSE);
}
}
/**
* Set region to OFFLINE unless it is opening and forceNewPlan is false.
*/
private RegionState forceRegionStateToOffline(
final HRegionInfo region, final boolean forceNewPlan) {
RegionState state = regionStates.getRegionState(region);
if (state == null) {
LOG.warn("Assigning a region not in region states: " + region);
state = regionStates.createRegionState(region);
}
ServerName sn = state.getServerName();
if (forceNewPlan && LOG.isDebugEnabled()) {
LOG.debug("Force region state offline " + state);
}
switch (state.getState()) {
case OPEN:
case OPENING:
case PENDING_OPEN:
case CLOSING:
case PENDING_CLOSE:
if (!forceNewPlan) {
LOG.debug("Skip assigning " +
region + ", it is already " + state);
return null;
}
case FAILED_CLOSE:
case FAILED_OPEN:
unassign(region, state, -1, null, false, null);
state = regionStates.getRegionState(region);
if (state.isFailedClose()) {
// If we can't close the region, we can't re-assign
// it so as to avoid possible double assignment/data loss.
LOG.info("Skip assigning " +
region + ", we couldn't close it: " + state);
return null;
}
case OFFLINE:
// This region could have been open on this server
// for a while. If the server is dead and not processed
// yet, we can move on only if the meta shows the
// region is not on this server actually, or on a server
// not dead, or dead and processed already.
if (regionStates.isServerDeadAndNotProcessed(sn)
&& wasRegionOnDeadServerByMeta(region, sn)) {
LOG.info("Skip assigning " + region.getRegionNameAsString()
+ ", it is on a dead but not processed yet server");
return null;
}
case CLOSED:
break;
default:
LOG.error("Trying to assign region " + region
+ ", which is " + state);
return null;
}
return state;
}
private boolean wasRegionOnDeadServerByMeta(
final HRegionInfo region, final ServerName sn) {
try {
if (region.isMetaRegion()) {
ServerName server = catalogTracker.getMetaLocation();
return regionStates.isServerDeadAndNotProcessed(server);
}
while (!server.isStopped()) {
try {
catalogTracker.waitForMeta();
Pair<HRegionInfo, ServerName> r =
MetaReader.getRegion(catalogTracker, region.getRegionName());
ServerName server = r == null ? null : r.getSecond();
return regionStates.isServerDeadAndNotProcessed(server);
} catch (IOException ioe) {
LOG.info("Received exception accessing hbase:meta during force assign "
+ region.getRegionNameAsString() + ", retrying", ioe);
}
}
} catch (InterruptedException e) {
Thread.currentThread().interrupt();
LOG.info("Interrupted accessing hbase:meta", e);
}
// Call is interrupted or server is stopped.
return regionStates.isServerDeadAndNotProcessed(sn);
}
/**
* Caller must hold lock on the passed <code>state</code> object.
* @param state
* @param setOfflineInZK
* @param forceNewPlan
*/
private void assign(RegionState state,
final boolean setOfflineInZK, final boolean forceNewPlan) {
long startTime = EnvironmentEdgeManager.currentTimeMillis();
try {
RegionState currentState = state;
int versionOfOfflineNode = -1;
RegionPlan plan = null;
long maxWaitTime = -1;
HRegionInfo region = state.getRegion();
RegionOpeningState regionOpenState;
for (int i = 1; i <= maximumAttempts; i++) {
if (server.isStopped() || server.isAborted()) {
LOG.info("Skip assigning " + region.getRegionNameAsString()
+ ", the server is stopped/aborted");
return;
}
if (plan == null) { // Get a server for the region at first
try {
plan = getRegionPlan(region, forceNewPlan);
} catch (HBaseIOException e) {
LOG.warn("Failed to get region plan", e);
}
}
if (plan == null) {
LOG.warn("Unable to determine a plan to assign " + region);
if (tomActivated){
this.timeoutMonitor.setAllRegionServersOffline(true);
} else {
if (region.isMetaRegion()) {
try {
Thread.sleep(this.sleepTimeBeforeRetryingMetaAssignment);
if (i == maximumAttempts) i = 1;
continue;
} catch (InterruptedException e) {
LOG.error("Got exception while waiting for hbase:meta assignment");
Thread.currentThread().interrupt();
}
}
regionStates.updateRegionState(region, State.FAILED_OPEN);
}
return;
}
if (setOfflineInZK && versionOfOfflineNode == -1) {
// get the version of the znode after setting it to OFFLINE.
// versionOfOfflineNode will be -1 if the znode was not set to OFFLINE
versionOfOfflineNode = setOfflineInZooKeeper(currentState, plan.getDestination());
if (versionOfOfflineNode != -1) {
if (isDisabledorDisablingRegionInRIT(region)) {
return;
}
// In case of assignment from EnableTableHandler table state is ENABLING. Any how
// EnableTableHandler will set ENABLED after assigning all the table regions. If we
// try to set to ENABLED directly then client API may think table is enabled.
// When we have a case such as all the regions are added directly into hbase:meta and we call
// assignRegion then we need to make the table ENABLED. Hence in such case the table
// will not be in ENABLING or ENABLED state.
TableName tableName = region.getTable();
if (!zkTable.isEnablingTable(tableName) && !zkTable.isEnabledTable(tableName)) {
LOG.debug("Setting table " + tableName + " to ENABLED state.");
setEnabledTable(tableName);
}
}
}
if (setOfflineInZK && versionOfOfflineNode == -1) {
LOG.info("Unable to set offline in ZooKeeper to assign " + region);
// Setting offline in ZK must have been failed due to ZK racing or some
// exception which may make the server to abort. If it is ZK racing,
// we should retry since we already reset the region state,
// existing (re)assignment will fail anyway.
if (!server.isAborted()) {
continue;
}
}
LOG.info("Assigning " + region.getRegionNameAsString() +
" to " + plan.getDestination().toString());
// Transition RegionState to PENDING_OPEN
currentState = regionStates.updateRegionState(region,
State.PENDING_OPEN, plan.getDestination());
boolean needNewPlan;
final String assignMsg = "Failed assignment of " + region.getRegionNameAsString() +
" to " + plan.getDestination();
try {
List<ServerName> favoredNodes = ServerName.EMPTY_SERVER_LIST;
if (this.shouldAssignRegionsWithFavoredNodes) {
favoredNodes = ((FavoredNodeLoadBalancer)this.balancer).getFavoredNodes(region);
}
regionOpenState = serverManager.sendRegionOpen(
plan.getDestination(), region, versionOfOfflineNode, favoredNodes);
if (regionOpenState == RegionOpeningState.FAILED_OPENING) {
// Failed opening this region, looping again on a new server.
needNewPlan = true;
LOG.warn(assignMsg + ", regionserver says 'FAILED_OPENING', " +
" trying to assign elsewhere instead; " +
"try=" + i + " of " + this.maximumAttempts);
} else {
// we're done
if (regionOpenState == RegionOpeningState.ALREADY_OPENED) {
processAlreadyOpenedRegion(region, plan.getDestination());
}
return;
}
} catch (Throwable t) {
if (t instanceof RemoteException) {
t = ((RemoteException) t).unwrapRemoteException();
}
// Should we wait a little before retrying? If the server is starting it's yes.
// If the region is already in transition, it's yes as well: we want to be sure that
// the region will get opened but we don't want a double assignment.
boolean hold = (t instanceof RegionAlreadyInTransitionException ||
t instanceof ServerNotRunningYetException);
// In case socket is timed out and the region server is still online,
// the openRegion RPC could have been accepted by the server and
// just the response didn't go through. So we will retry to
// open the region on the same server to avoid possible
// double assignment.
boolean retry = !hold && (t instanceof java.net.SocketTimeoutException
&& this.serverManager.isServerOnline(plan.getDestination()));
if (hold) {
LOG.warn(assignMsg + ", waiting a little before trying on the same region server " +
"try=" + i + " of " + this.maximumAttempts, t);
if (maxWaitTime < 0) {
if (t instanceof RegionAlreadyInTransitionException) {
maxWaitTime = EnvironmentEdgeManager.currentTimeMillis()
+ this.server.getConfiguration().getLong(ALREADY_IN_TRANSITION_WAITTIME,
DEFAULT_ALREADY_IN_TRANSITION_WAITTIME);
} else {
maxWaitTime = this.server.getConfiguration().
getLong("hbase.regionserver.rpc.startup.waittime", 60000);
}
}
try {
needNewPlan = false;
long now = EnvironmentEdgeManager.currentTimeMillis();
if (now < maxWaitTime) {
LOG.debug("Server is not yet up or region is already in transition; "
+ "waiting up to " + (maxWaitTime - now) + "ms", t);
Thread.sleep(100);
i--; // reset the try count
} else if (!(t instanceof RegionAlreadyInTransitionException)) {
LOG.debug("Server is not up for a while; try a new one", t);
needNewPlan = true;
}
} catch (InterruptedException ie) {
LOG.warn("Failed to assign "
+ region.getRegionNameAsString() + " since interrupted", ie);
Thread.currentThread().interrupt();
if (!tomActivated) {
regionStates.updateRegionState(region, State.FAILED_OPEN);
}
return;
}
} else if (retry) {
needNewPlan = false;
LOG.warn(assignMsg + ", trying to assign to the same region server " +
"try=" + i + " of " + this.maximumAttempts, t);
} else {
needNewPlan = true;
LOG.warn(assignMsg + ", trying to assign elsewhere instead;" +
" try=" + i + " of " + this.maximumAttempts, t);
}
}
if (i == this.maximumAttempts) {
// Don't reset the region state or get a new plan any more.
// This is the last try.
continue;
}
// If region opened on destination of present plan, reassigning to new
// RS may cause double assignments. In case of RegionAlreadyInTransitionException
// reassigning to same RS.
if (needNewPlan) {
// Force a new plan and reassign. Will return null if no servers.
// The new plan could be the same as the existing plan since we don't
// exclude the server of the original plan, which should not be
// excluded since it could be the only server up now.
RegionPlan newPlan = null;
try {
newPlan = getRegionPlan(region, true);
} catch (HBaseIOException e) {
LOG.warn("Failed to get region plan", e);
}
if (newPlan == null) {
if (tomActivated) {
this.timeoutMonitor.setAllRegionServersOffline(true);
} else {
regionStates.updateRegionState(region, State.FAILED_OPEN);
}
LOG.warn("Unable to find a viable location to assign region " +
region.getRegionNameAsString());
return;
}
if (plan != newPlan && !plan.getDestination().equals(newPlan.getDestination())) {
// Clean out plan we failed execute and one that doesn't look like it'll
// succeed anyways; we need a new plan!
// Transition back to OFFLINE
currentState = regionStates.updateRegionState(region, State.OFFLINE);
versionOfOfflineNode = -1;
plan = newPlan;
}
}
}
// Run out of attempts
if (!tomActivated) {
regionStates.updateRegionState(region, State.FAILED_OPEN);
}
} finally {
metricsAssignmentManager.updateAssignmentTime(EnvironmentEdgeManager.currentTimeMillis() - startTime);
}
}
private void processAlreadyOpenedRegion(HRegionInfo region, ServerName sn) {
// Remove region from in-memory transition and unassigned node from ZK
// While trying to enable the table the regions of the table were
// already enabled.
LOG.debug("ALREADY_OPENED " + region.getRegionNameAsString()
+ " to " + sn);
String encodedName = region.getEncodedName();
deleteNodeInStates(encodedName, "offline", sn, EventType.M_ZK_REGION_OFFLINE);
regionStates.regionOnline(region, sn);
}
private boolean isDisabledorDisablingRegionInRIT(final HRegionInfo region) {
TableName tableName = region.getTable();
boolean disabled = this.zkTable.isDisabledTable(tableName);
if (disabled || this.zkTable.isDisablingTable(tableName)) {
LOG.info("Table " + tableName + (disabled ? " disabled;" : " disabling;") +
" skipping assign of " + region.getRegionNameAsString());
offlineDisabledRegion(region);
return true;
}
return false;
}
/**
* Set region as OFFLINED up in zookeeper
*
* @param state
* @return the version of the offline node if setting of the OFFLINE node was
* successful, -1 otherwise.
*/
private int setOfflineInZooKeeper(final RegionState state, final ServerName destination) {
if (!state.isClosed() && !state.isOffline()) {
String msg = "Unexpected state : " + state + " .. Cannot transit it to OFFLINE.";
this.server.abort(msg, new IllegalStateException(msg));
return -1;
}
regionStates.updateRegionState(state.getRegion(), State.OFFLINE);
int versionOfOfflineNode;
try {
// get the version after setting the znode to OFFLINE
versionOfOfflineNode = ZKAssign.createOrForceNodeOffline(watcher,
state.getRegion(), destination);
if (versionOfOfflineNode == -1) {
LOG.warn("Attempted to create/force node into OFFLINE state before "
+ "completing assignment but failed to do so for " + state);
return -1;
}
} catch (KeeperException e) {
server.abort("Unexpected ZK exception creating/setting node OFFLINE", e);
return -1;
}
return versionOfOfflineNode;
}
/**
* @param region the region to assign
* @return Plan for passed <code>region</code> (If none currently, it creates one or
* if no servers to assign, it returns null).
*/
private RegionPlan getRegionPlan(final HRegionInfo region,
final boolean forceNewPlan) throws HBaseIOException {
return getRegionPlan(region, null, forceNewPlan);
}
/**
* @param region the region to assign
* @param serverToExclude Server to exclude (we know its bad). Pass null if
* all servers are thought to be assignable.
* @param forceNewPlan If true, then if an existing plan exists, a new plan
* will be generated.
* @return Plan for passed <code>region</code> (If none currently, it creates one or
* if no servers to assign, it returns null).
*/
private RegionPlan getRegionPlan(final HRegionInfo region,
final ServerName serverToExclude, final boolean forceNewPlan) throws HBaseIOException {
// Pickup existing plan or make a new one
final String encodedName = region.getEncodedName();
final List<ServerName> destServers =
serverManager.createDestinationServersList(serverToExclude);
if (destServers.isEmpty()){
LOG.warn("Can't move " + encodedName +
", there is no destination server available.");
return null;
}
RegionPlan randomPlan = null;
boolean newPlan = false;
RegionPlan existingPlan;
synchronized (this.regionPlans) {
existingPlan = this.regionPlans.get(encodedName);
if (existingPlan != null && existingPlan.getDestination() != null) {
LOG.debug("Found an existing plan for " + region.getRegionNameAsString()
+ " destination server is " + existingPlan.getDestination() +
" accepted as a dest server = " + destServers.contains(existingPlan.getDestination()));
}
if (forceNewPlan
|| existingPlan == null
|| existingPlan.getDestination() == null
|| !destServers.contains(existingPlan.getDestination())) {
newPlan = true;
randomPlan = new RegionPlan(region, null,
balancer.randomAssignment(region, destServers));
if (!region.isMetaTable() && shouldAssignRegionsWithFavoredNodes) {
List<HRegionInfo> regions = new ArrayList<HRegionInfo>(1);
regions.add(region);
try {
processFavoredNodes(regions);
} catch (IOException ie) {
LOG.warn("Ignoring exception in processFavoredNodes " + ie);
}
}
this.regionPlans.put(encodedName, randomPlan);
}
}
if (newPlan) {
if (randomPlan.getDestination() == null) {
LOG.warn("Can't find a destination for " + encodedName);
return null;
}
LOG.debug("No previous transition plan found (or ignoring " +
"an existing plan) for " + region.getRegionNameAsString() +
"; generated random plan=" + randomPlan + "; " +
serverManager.countOfRegionServers() +
" (online=" + serverManager.getOnlineServers().size() +
", available=" + destServers.size() + ") available servers" +
", forceNewPlan=" + forceNewPlan);
return randomPlan;
}
LOG.debug("Using pre-existing plan for " +
region.getRegionNameAsString() + "; plan=" + existingPlan);
return existingPlan;
}
/**
* Unassigns the specified region.
* <p>
* Updates the RegionState and sends the CLOSE RPC unless region is being
* split by regionserver; then the unassign fails (silently) because we
* presume the region being unassigned no longer exists (its been split out
* of existence). TODO: What to do if split fails and is rolled back and
* parent is revivified?
* <p>
* If a RegionPlan is already set, it will remain.
*
* @param region server to be unassigned
*/
public void unassign(HRegionInfo region) {
unassign(region, false);
}
/**
* Unassigns the specified region.
* <p>
* Updates the RegionState and sends the CLOSE RPC unless region is being
* split by regionserver; then the unassign fails (silently) because we
* presume the region being unassigned no longer exists (its been split out
* of existence). TODO: What to do if split fails and is rolled back and
* parent is revivified?
* <p>
* If a RegionPlan is already set, it will remain.
*
* @param region server to be unassigned
* @param force if region should be closed even if already closing
*/
public void unassign(HRegionInfo region, boolean force, ServerName dest) {
// TODO: Method needs refactoring. Ugly buried returns throughout. Beware!
LOG.debug("Starting unassign of " + region.getRegionNameAsString()
+ " (offlining), current state: " + regionStates.getRegionState(region));
String encodedName = region.getEncodedName();
// Grab the state of this region and synchronize on it
int versionOfClosingNode = -1;
// We need a lock here as we're going to do a put later and we don't want multiple states
// creation
ReentrantLock lock = locker.acquireLock(encodedName);
RegionState state = regionStates.getRegionTransitionState(encodedName);
boolean reassign = true;
try {
if (state == null) {
// Region is not in transition.
// We can unassign it only if it's not SPLIT/MERGED.
state = regionStates.getRegionState(encodedName);
if (state != null && state.isUnassignable()) {
LOG.info("Attempting to unassign " + state + ", ignored");
// Offline region will be reassigned below
return;
}
// Create the znode in CLOSING state
try {
if (state == null || state.getServerName() == null) {
// We don't know where the region is, offline it.
// No need to send CLOSE RPC
LOG.warn("Attempting to unassign a region not in RegionStates"
+ region.getRegionNameAsString() + ", offlined");
regionOffline(region);
return;
}
versionOfClosingNode = ZKAssign.createNodeClosing(
watcher, region, state.getServerName());
if (versionOfClosingNode == -1) {
LOG.info("Attempting to unassign " +
region.getRegionNameAsString() + " but ZK closing node "
+ "can't be created.");
reassign = false; // not unassigned at all
return;
}
} catch (KeeperException e) {
if (e instanceof NodeExistsException) {
// Handle race between master initiated close and regionserver
// orchestrated splitting. See if existing node is in a
// SPLITTING or SPLIT state. If so, the regionserver started
// an op on node before we could get our CLOSING in. Deal.
NodeExistsException nee = (NodeExistsException)e;
String path = nee.getPath();
try {
if (isSplitOrSplittingOrMergedOrMerging(path)) {
LOG.debug(path + " is SPLIT or SPLITTING or MERGED or MERGING; " +
"skipping unassign because region no longer exists -- its split or merge");
reassign = false; // no need to reassign for split/merged region
return;
}
} catch (KeeperException.NoNodeException ke) {
LOG.warn("Failed getData on SPLITTING/SPLIT at " + path +
"; presuming split and that the region to unassign, " +
encodedName + ", no longer exists -- confirm", ke);
return;
} catch (KeeperException ke) {
LOG.error("Unexpected zk state", ke);
} catch (DeserializationException de) {
LOG.error("Failed parse", de);
}
}
// If we get here, don't understand whats going on -- abort.
server.abort("Unexpected ZK exception creating node CLOSING", e);
reassign = false; // heading out already
return;
}
state = regionStates.updateRegionState(region, State.PENDING_CLOSE);
} else if (state.isFailedOpen()) {
// The region is not open yet
regionOffline(region);
return;
} else if (force && state.isPendingCloseOrClosing()) {
LOG.debug("Attempting to unassign " + region.getRegionNameAsString() +
" which is already " + state.getState() +
" but forcing to send a CLOSE RPC again ");
if (state.isFailedClose()) {
state = regionStates.updateRegionState(region, State.PENDING_CLOSE);
}
state.updateTimestampToNow();
} else {
LOG.debug("Attempting to unassign " +
region.getRegionNameAsString() + " but it is " +
"already in transition (" + state.getState() + ", force=" + force + ")");
return;
}
unassign(region, state, versionOfClosingNode, dest, true, null);
} finally {
lock.unlock();
// Region is expected to be reassigned afterwards
if (reassign && regionStates.isRegionOffline(region)) {
assign(region, true);
}
}
}
public void unassign(HRegionInfo region, boolean force){
unassign(region, force, null);
}
/**
* @param region regioninfo of znode to be deleted.
*/
public void deleteClosingOrClosedNode(HRegionInfo region, ServerName sn) {
String encodedName = region.getEncodedName();
deleteNodeInStates(encodedName, "closing", sn, EventType.M_ZK_REGION_CLOSING,
EventType.RS_ZK_REGION_CLOSED);
}
/**
* @param path
* @return True if znode is in SPLIT or SPLITTING or MERGED or MERGING state.
* @throws KeeperException Can happen if the znode went away in meantime.
* @throws DeserializationException
*/
private boolean isSplitOrSplittingOrMergedOrMerging(final String path)
throws KeeperException, DeserializationException {
boolean result = false;
// This may fail if the SPLIT or SPLITTING or MERGED or MERGING znode gets
// cleaned up before we can get data from it.
byte [] data = ZKAssign.getData(watcher, path);
if (data == null) {
LOG.info("Node " + path + " is gone");
return false;
}
RegionTransition rt = RegionTransition.parseFrom(data);
switch (rt.getEventType()) {
case RS_ZK_REQUEST_REGION_SPLIT:
case RS_ZK_REGION_SPLIT:
case RS_ZK_REGION_SPLITTING:
case RS_ZK_REQUEST_REGION_MERGE:
case RS_ZK_REGION_MERGED:
case RS_ZK_REGION_MERGING:
result = true;
break;
default:
LOG.info("Node " + path + " is in " + rt.getEventType());
break;
}
return result;
}
/**
* Used by unit tests. Return the number of regions opened so far in the life
* of the master. Increases by one every time the master opens a region
* @return the counter value of the number of regions opened so far
*/
public int getNumRegionsOpened() {
return numRegionsOpened.get();
}
/**
* Waits until the specified region has completed assignment.
* <p>
* If the region is already assigned, returns immediately. Otherwise, method
* blocks until the region is assigned.
* @param regionInfo region to wait on assignment for
* @throws InterruptedException
*/
public boolean waitForAssignment(HRegionInfo regionInfo)
throws InterruptedException {
while (!regionStates.isRegionOnline(regionInfo)) {
if (regionStates.isRegionInState(regionInfo, State.FAILED_OPEN)
|| this.server.isStopped()) {
return false;
}
// We should receive a notification, but it's
// better to have a timeout to recheck the condition here:
// it lowers the impact of a race condition if any
regionStates.waitForUpdate(100);
}
return true;
}
/**
* Assigns the hbase:meta region.
* <p>
* Assumes that hbase:meta is currently closed and is not being actively served by
* any RegionServer.
* <p>
* Forcibly unsets the current meta region location in ZooKeeper and assigns
* hbase:meta to a random RegionServer.
* @throws KeeperException
*/
public void assignMeta() throws KeeperException {
MetaRegionTracker.deleteMetaLocation(this.watcher);
assign(HRegionInfo.FIRST_META_REGIONINFO, true);
}
/**
* Assigns specified regions retaining assignments, if any.
* <p>
* This is a synchronous call and will return once every region has been
* assigned. If anything fails, an exception is thrown
* @throws InterruptedException
* @throws IOException
*/
public void assign(Map<HRegionInfo, ServerName> regions)
throws IOException, InterruptedException {
if (regions == null || regions.isEmpty()) {
return;
}
List<ServerName> servers = serverManager.createDestinationServersList();
if (servers == null || servers.isEmpty()) {
throw new IOException("Found no destination server to assign region(s)");
}
// Reuse existing assignment info
Map<ServerName, List<HRegionInfo>> bulkPlan =
balancer.retainAssignment(regions, servers);
assign(regions.size(), servers.size(),
"retainAssignment=true", bulkPlan);
}
/**
* Assigns specified regions round robin, if any.
* <p>
* This is a synchronous call and will return once every region has been
* assigned. If anything fails, an exception is thrown
* @throws InterruptedException
* @throws IOException
*/
public void assign(List<HRegionInfo> regions)
throws IOException, InterruptedException {
if (regions == null || regions.isEmpty()) {
return;
}
List<ServerName> servers = serverManager.createDestinationServersList();
if (servers == null || servers.isEmpty()) {
throw new IOException("Found no destination server to assign region(s)");
}
// Generate a round-robin bulk assignment plan
Map<ServerName, List<HRegionInfo>> bulkPlan
= balancer.roundRobinAssignment(regions, servers);
processFavoredNodes(regions);
assign(regions.size(), servers.size(),
"round-robin=true", bulkPlan);
}
private void assign(int regions, int totalServers,
String message, Map<ServerName, List<HRegionInfo>> bulkPlan)
throws InterruptedException, IOException {
int servers = bulkPlan.size();
if (servers == 1 || (regions < bulkAssignThresholdRegions
&& servers < bulkAssignThresholdServers)) {
// Not use bulk assignment. This could be more efficient in small
// cluster, especially mini cluster for testing, so that tests won't time out
if (LOG.isTraceEnabled()) {
LOG.trace("Not using bulk assignment since we are assigning only " + regions +
" region(s) to " + servers + " server(s)");
}
for (Map.Entry<ServerName, List<HRegionInfo>> plan: bulkPlan.entrySet()) {
if (!assign(plan.getKey(), plan.getValue())) {
for (HRegionInfo region: plan.getValue()) {
if (!regionStates.isRegionOnline(region)) {
invokeAssign(region);
}
}
}
}
} else {
LOG.info("Bulk assigning " + regions + " region(s) across "
+ totalServers + " server(s), " + message);
// Use fixed count thread pool assigning.
BulkAssigner ba = new GeneralBulkAssigner(
this.server, bulkPlan, this, bulkAssignWaitTillAllAssigned);
ba.bulkAssign();
LOG.info("Bulk assigning done");
}
}
/**
* Assigns all user regions, if any exist. Used during cluster startup.
* <p>
* This is a synchronous call and will return once every region has been
* assigned. If anything fails, an exception is thrown and the cluster
* should be shutdown.
* @throws InterruptedException
* @throws IOException
* @throws KeeperException
*/
private void assignAllUserRegions()
throws IOException, InterruptedException, KeeperException {
// Cleanup any existing ZK nodes and start watching
ZKAssign.deleteAllNodes(watcher);
ZKUtil.listChildrenAndWatchForNewChildren(this.watcher,
this.watcher.assignmentZNode);
failoverCleanupDone();
// Skip assignment for regions of tables in DISABLING state because during clean cluster startup
// no RS is alive and regions map also doesn't have any information about the regions.
// See HBASE-6281.
Set<TableName> disabledOrDisablingOrEnabling = ZKTable.getDisabledOrDisablingTables(watcher);
disabledOrDisablingOrEnabling.addAll(ZKTable.getEnablingTables(watcher));
// Scan hbase:meta for all user regions, skipping any disabled tables
Map<HRegionInfo, ServerName> allRegions;
SnapshotOfRegionAssignmentFromMeta snapshotOfRegionAssignment =
new SnapshotOfRegionAssignmentFromMeta(catalogTracker, disabledOrDisablingOrEnabling, true);
snapshotOfRegionAssignment.initialize();
allRegions = snapshotOfRegionAssignment.getRegionToRegionServerMap();
if (allRegions == null || allRegions.isEmpty()) return;
// Determine what type of assignment to do on startup
boolean retainAssignment = server.getConfiguration().
getBoolean("hbase.master.startup.retainassign", true);
if (retainAssignment) {
assign(allRegions);
} else {
List<HRegionInfo> regions = new ArrayList<HRegionInfo>(allRegions.keySet());
assign(regions);
}
for (HRegionInfo hri : allRegions.keySet()) {
TableName tableName = hri.getTable();
if (!zkTable.isEnabledTable(tableName)) {
setEnabledTable(tableName);
}
}
}
/**
* Wait until no regions in transition.
* @param timeout How long to wait.
* @return True if nothing in regions in transition.
* @throws InterruptedException
*/
boolean waitUntilNoRegionsInTransition(final long timeout)
throws InterruptedException {
// Blocks until there are no regions in transition. It is possible that
// there
// are regions in transition immediately after this returns but guarantees
// that if it returns without an exception that there was a period of time
// with no regions in transition from the point-of-view of the in-memory
// state of the Master.
final long endTime = System.currentTimeMillis() + timeout;
while (!this.server.isStopped() && regionStates.isRegionsInTransition()
&& endTime > System.currentTimeMillis()) {
regionStates.waitForUpdate(100);
}
return !regionStates.isRegionsInTransition();
}
/**
* Rebuild the list of user regions and assignment information.
* <p>
* Returns a map of servers that are not found to be online and the regions
* they were hosting.
* @return map of servers not online to their assigned regions, as stored
* in META
* @throws IOException
*/
Map<ServerName, List<HRegionInfo>> rebuildUserRegions() throws IOException, KeeperException {
Set<TableName> enablingTables = ZKTable.getEnablingTables(watcher);
Set<TableName> disabledOrEnablingTables = ZKTable.getDisabledTables(watcher);
disabledOrEnablingTables.addAll(enablingTables);
Set<TableName> disabledOrDisablingOrEnabling = ZKTable.getDisablingTables(watcher);
disabledOrDisablingOrEnabling.addAll(disabledOrEnablingTables);
// Region assignment from META
List<Result> results = MetaReader.fullScan(this.catalogTracker);
// Get any new but slow to checkin region server that joined the cluster
Set<ServerName> onlineServers = serverManager.getOnlineServers().keySet();
// Map of offline servers and their regions to be returned
Map<ServerName, List<HRegionInfo>> offlineServers =
new TreeMap<ServerName, List<HRegionInfo>>();
// Iterate regions in META
for (Result result : results) {
Pair<HRegionInfo, ServerName> region = HRegionInfo.getHRegionInfoAndServerName(result);
if (region == null) continue;
HRegionInfo regionInfo = region.getFirst();
ServerName regionLocation = region.getSecond();
if (regionInfo == null) continue;
regionStates.createRegionState(regionInfo);
if (regionStates.isRegionInState(regionInfo, State.SPLIT)) {
// Split is considered to be completed. If the split znode still
// exists, the region will be put back to SPLITTING state later
LOG.debug("Region " + regionInfo.getRegionNameAsString()
+ " split is completed. Hence need not add to regions list");
continue;
}
TableName tableName = regionInfo.getTable();
if (regionLocation == null) {
// regionLocation could be null if createTable didn't finish properly.
// When createTable is in progress, HMaster restarts.
// Some regions have been added to hbase:meta, but have not been assigned.
// When this happens, the region's table must be in ENABLING state.
// It can't be in ENABLED state as that is set when all regions are
// assigned.
// It can't be in DISABLING state, because DISABLING state transitions
// from ENABLED state when application calls disableTable.
// It can't be in DISABLED state, because DISABLED states transitions
// from DISABLING state.
if (!enablingTables.contains(tableName)) {
LOG.warn("Region " + regionInfo.getEncodedName() +
" has null regionLocation." + " But its table " + tableName +
" isn't in ENABLING state.");
}
} else if (!onlineServers.contains(regionLocation)) {
// Region is located on a server that isn't online
List<HRegionInfo> offlineRegions = offlineServers.get(regionLocation);
if (offlineRegions == null) {
offlineRegions = new ArrayList<HRegionInfo>(1);
offlineServers.put(regionLocation, offlineRegions);
}
offlineRegions.add(regionInfo);
// need to enable the table if not disabled or disabling or enabling
// this will be used in rolling restarts
if (!disabledOrDisablingOrEnabling.contains(tableName)
&& !getZKTable().isEnabledTable(tableName)) {
setEnabledTable(tableName);
}
} else {
// Region is being served and on an active server
// add only if region not in disabled or enabling table
if (!disabledOrEnablingTables.contains(tableName)) {
regionStates.updateRegionState(regionInfo, State.OPEN, regionLocation);
regionStates.regionOnline(regionInfo, regionLocation);
}
// need to enable the table if not disabled or disabling or enabling
// this will be used in rolling restarts
if (!disabledOrDisablingOrEnabling.contains(tableName)
&& !getZKTable().isEnabledTable(tableName)) {
setEnabledTable(tableName);
}
}
}
return offlineServers;
}
/**
* Recover the tables that were not fully moved to DISABLED state. These
* tables are in DISABLING state when the master restarted/switched.
*
* @throws KeeperException
* @throws TableNotFoundException
* @throws IOException
*/
private void recoverTableInDisablingState()
throws KeeperException, TableNotFoundException, IOException {
Set<TableName> disablingTables = ZKTable.getDisablingTables(watcher);
if (disablingTables.size() != 0) {
for (TableName tableName : disablingTables) {
// Recover by calling DisableTableHandler
LOG.info("The table " + tableName
+ " is in DISABLING state. Hence recovering by moving the table"
+ " to DISABLED state.");
new DisableTableHandler(this.server, tableName, catalogTracker,
this, tableLockManager, true).prepare().process();
}
}
}
/**
* Recover the tables that are not fully moved to ENABLED state. These tables
* are in ENABLING state when the master restarted/switched
*
* @throws KeeperException
* @throws org.apache.hadoop.hbase.TableNotFoundException
* @throws IOException
*/
private void recoverTableInEnablingState()
throws KeeperException, TableNotFoundException, IOException {
Set<TableName> enablingTables = ZKTable.getEnablingTables(watcher);
if (enablingTables.size() != 0) {
for (TableName tableName : enablingTables) {
// Recover by calling EnableTableHandler
LOG.info("The table " + tableName
+ " is in ENABLING state. Hence recovering by moving the table"
+ " to ENABLED state.");
// enableTable in sync way during master startup,
// no need to invoke coprocessor
EnableTableHandler eth = new EnableTableHandler(this.server, tableName,
catalogTracker, this, tableLockManager, true);
try {
eth.prepare();
} catch (TableNotFoundException e) {
LOG.warn("Table " + tableName + " not found in hbase:meta to recover.");
continue;
}
eth.process();
}
}
}
/**
* Processes list of dead servers from result of hbase:meta scan and regions in RIT
* <p>
* This is used for failover to recover the lost regions that belonged to
* RegionServers which failed while there was no active master or regions
* that were in RIT.
* <p>
*
*
* @param deadServers
* The list of dead servers which failed while there was no active
* master. Can be null.
* @throws IOException
* @throws KeeperException
*/
private void processDeadServersAndRecoverLostRegions(
Map<ServerName, List<HRegionInfo>> deadServers)
throws IOException, KeeperException {
if (deadServers != null) {
for (Map.Entry<ServerName, List<HRegionInfo>> server: deadServers.entrySet()) {
ServerName serverName = server.getKey();
// We need to keep such info even if the server is known dead
regionStates.setLastRegionServerOfRegions(serverName, server.getValue());
if (!serverManager.isServerDead(serverName)) {
serverManager.expireServer(serverName); // Let SSH do region re-assign
}
}
}
List<String> nodes = ZKUtil.listChildrenAndWatchForNewChildren(
this.watcher, this.watcher.assignmentZNode);
if (!nodes.isEmpty()) {
for (String encodedRegionName : nodes) {
processRegionInTransition(encodedRegionName, null);
}
}
// Now we can safely claim failover cleanup completed and enable
// ServerShutdownHandler for further processing. The nodes (below)
// in transition, if any, are for regions not related to those
// dead servers at all, and can be done in parallel to SSH.
failoverCleanupDone();
}
/**
* Set Regions in transitions metrics.
* This takes an iterator on the RegionInTransition map (CLSM), and is not synchronized.
* This iterator is not fail fast, which may lead to stale read; but that's better than
* creating a copy of the map for metrics computation, as this method will be invoked
* on a frequent interval.
*/
public void updateRegionsInTransitionMetrics() {
long currentTime = System.currentTimeMillis();
int totalRITs = 0;
int totalRITsOverThreshold = 0;
long oldestRITTime = 0;
int ritThreshold = this.server.getConfiguration().
getInt(HConstants.METRICS_RIT_STUCK_WARNING_THRESHOLD, 60000);
for (RegionState state: regionStates.getRegionsInTransition().values()) {
totalRITs++;
long ritTime = currentTime - state.getStamp();
if (ritTime > ritThreshold) { // more than the threshold
totalRITsOverThreshold++;
}
if (oldestRITTime < ritTime) {
oldestRITTime = ritTime;
}
}
if (this.metricsAssignmentManager != null) {
this.metricsAssignmentManager.updateRITOldestAge(oldestRITTime);
this.metricsAssignmentManager.updateRITCount(totalRITs);
this.metricsAssignmentManager.updateRITCountOverThreshold(totalRITsOverThreshold);
}
}
/**
* @param region Region whose plan we are to clear.
*/
void clearRegionPlan(final HRegionInfo region) {
synchronized (this.regionPlans) {
this.regionPlans.remove(region.getEncodedName());
}
}
/**
* Wait on region to clear regions-in-transition.
* @param hri Region to wait on.
* @throws IOException
*/
public void waitOnRegionToClearRegionsInTransition(final HRegionInfo hri)
throws IOException, InterruptedException {
waitOnRegionToClearRegionsInTransition(hri, -1L);
}
/**
* Wait on region to clear regions-in-transition or time out
* @param hri
* @param timeOut Milliseconds to wait for current region to be out of transition state.
* @return True when a region clears regions-in-transition before timeout otherwise false
* @throws InterruptedException
*/
public boolean waitOnRegionToClearRegionsInTransition(final HRegionInfo hri, long timeOut)
throws InterruptedException {
if (!regionStates.isRegionInTransition(hri)) return true;
long end = (timeOut <= 0) ? Long.MAX_VALUE : EnvironmentEdgeManager.currentTimeMillis()
+ timeOut;
// There is already a timeout monitor on regions in transition so I
// should not have to have one here too?
LOG.info("Waiting for " + hri.getEncodedName() +
" to leave regions-in-transition, timeOut=" + timeOut + " ms.");
while (!this.server.isStopped() && regionStates.isRegionInTransition(hri)) {
regionStates.waitForUpdate(100);
if (EnvironmentEdgeManager.currentTimeMillis() > end) {
LOG.info("Timed out on waiting for " + hri.getEncodedName() + " to be assigned.");
return false;
}
}
if (this.server.isStopped()) {
LOG.info("Giving up wait on regions in transition because stoppable.isStopped is set");
return false;
}
return true;
}
/**
* Update timers for all regions in transition going against the server in the
* serversInUpdatingTimer.
*/
public class TimerUpdater extends Chore {
public TimerUpdater(final int period, final Stoppable stopper) {
super("AssignmentTimerUpdater", period, stopper);
}
@Override
protected void chore() {
Preconditions.checkState(tomActivated);
ServerName serverToUpdateTimer = null;
while (!serversInUpdatingTimer.isEmpty() && !stopper.isStopped()) {
if (serverToUpdateTimer == null) {
serverToUpdateTimer = serversInUpdatingTimer.first();
} else {
serverToUpdateTimer = serversInUpdatingTimer
.higher(serverToUpdateTimer);
}
if (serverToUpdateTimer == null) {
break;
}
updateTimers(serverToUpdateTimer);
serversInUpdatingTimer.remove(serverToUpdateTimer);
}
}
}
/**
* Monitor to check for time outs on region transition operations
*/
public class TimeoutMonitor extends Chore {
private boolean allRegionServersOffline = false;
private ServerManager serverManager;
private final int timeout;
/**
* Creates a periodic monitor to check for time outs on region transition
* operations. This will deal with retries if for some reason something
* doesn't happen within the specified timeout.
* @param period
* @param stopper When {@link Stoppable#isStopped()} is true, this thread will
* cleanup and exit cleanly.
* @param timeout
*/
public TimeoutMonitor(final int period, final Stoppable stopper,
ServerManager serverManager,
final int timeout) {
super("AssignmentTimeoutMonitor", period, stopper);
this.timeout = timeout;
this.serverManager = serverManager;
}
private synchronized void setAllRegionServersOffline(
boolean allRegionServersOffline) {
this.allRegionServersOffline = allRegionServersOffline;
}
@Override
protected void chore() {
Preconditions.checkState(tomActivated);
boolean noRSAvailable = this.serverManager.createDestinationServersList().isEmpty();
// Iterate all regions in transition checking for time outs
long now = System.currentTimeMillis();
// no lock concurrent access ok: we will be working on a copy, and it's java-valid to do
// a copy while another thread is adding/removing items
for (String regionName : regionStates.getRegionsInTransition().keySet()) {
RegionState regionState = regionStates.getRegionTransitionState(regionName);
if (regionState == null) continue;
if (regionState.getStamp() + timeout <= now) {
// decide on action upon timeout
actOnTimeOut(regionState);
} else if (this.allRegionServersOffline && !noRSAvailable) {
RegionPlan existingPlan = regionPlans.get(regionName);
if (existingPlan == null
|| !this.serverManager.isServerOnline(existingPlan
.getDestination())) {
// if some RSs just came back online, we can start the assignment
// right away
actOnTimeOut(regionState);
}
}
}
setAllRegionServersOffline(noRSAvailable);
}
private void actOnTimeOut(RegionState regionState) {
HRegionInfo regionInfo = regionState.getRegion();
LOG.info("Regions in transition timed out: " + regionState);
// Expired! Do a retry.
switch (regionState.getState()) {
case CLOSED:
LOG.info("Region " + regionInfo.getEncodedName()
+ " has been CLOSED for too long, waiting on queued "
+ "ClosedRegionHandler to run or server shutdown");
// Update our timestamp.
regionState.updateTimestampToNow();
break;
case OFFLINE:
LOG.info("Region has been OFFLINE for too long, " + "reassigning "
+ regionInfo.getRegionNameAsString() + " to a random server");
invokeAssign(regionInfo);
break;
case PENDING_OPEN:
LOG.info("Region has been PENDING_OPEN for too "
+ "long, reassigning region=" + regionInfo.getRegionNameAsString());
invokeAssign(regionInfo);
break;
case OPENING:
processOpeningState(regionInfo);
break;
case OPEN:
LOG.error("Region has been OPEN for too long, " +
"we don't know where region was opened so can't do anything");
regionState.updateTimestampToNow();
break;
case PENDING_CLOSE:
LOG.info("Region has been PENDING_CLOSE for too "
+ "long, running forced unassign again on region="
+ regionInfo.getRegionNameAsString());
invokeUnassign(regionInfo);
break;
case CLOSING:
LOG.info("Region has been CLOSING for too " +
"long, this should eventually complete or the server will " +
"expire, send RPC again");
invokeUnassign(regionInfo);
break;
case SPLIT:
case SPLITTING:
case FAILED_OPEN:
case FAILED_CLOSE:
case MERGING:
break;
default:
throw new IllegalStateException("Received event is not valid.");
}
}
}
private void processOpeningState(HRegionInfo regionInfo) {
LOG.info("Region has been OPENING for too long, reassigning region="
+ regionInfo.getRegionNameAsString());
// Should have a ZK node in OPENING state
try {
String node = ZKAssign.getNodeName(watcher, regionInfo.getEncodedName());
Stat stat = new Stat();
byte [] data = ZKAssign.getDataNoWatch(watcher, node, stat);
if (data == null) {
LOG.warn("Data is null, node " + node + " no longer exists");
return;
}
RegionTransition rt = RegionTransition.parseFrom(data);
EventType et = rt.getEventType();
if (et == EventType.RS_ZK_REGION_OPENED) {
LOG.debug("Region has transitioned to OPENED, allowing "
+ "watched event handlers to process");
return;
} else if (et != EventType.RS_ZK_REGION_OPENING && et != EventType.RS_ZK_REGION_FAILED_OPEN ) {
LOG.warn("While timing out a region, found ZK node in unexpected state: " + et);
return;
}
invokeAssign(regionInfo);
} catch (KeeperException ke) {
LOG.error("Unexpected ZK exception timing out CLOSING region", ke);
} catch (DeserializationException e) {
LOG.error("Unexpected exception parsing CLOSING region", e);
}
}
void invokeAssign(HRegionInfo regionInfo) {
threadPoolExecutorService.submit(new AssignCallable(this, regionInfo));
}
private void invokeUnassign(HRegionInfo regionInfo) {
threadPoolExecutorService.submit(new UnAssignCallable(this, regionInfo));
}
public boolean isCarryingMeta(ServerName serverName) {
return isCarryingRegion(serverName, HRegionInfo.FIRST_META_REGIONINFO);
}
/**
* Check if the shutdown server carries the specific region.
* We have a bunch of places that store region location
* Those values aren't consistent. There is a delay of notification.
* The location from zookeeper unassigned node has the most recent data;
* but the node could be deleted after the region is opened by AM.
* The AM's info could be old when OpenedRegionHandler
* processing hasn't finished yet when server shutdown occurs.
* @return whether the serverName currently hosts the region
*/
private boolean isCarryingRegion(ServerName serverName, HRegionInfo hri) {
RegionTransition rt = null;
try {
byte [] data = ZKAssign.getData(watcher, hri.getEncodedName());
// This call can legitimately come by null
rt = data == null? null: RegionTransition.parseFrom(data);
} catch (KeeperException e) {
server.abort("Exception reading unassigned node for region=" + hri.getEncodedName(), e);
} catch (DeserializationException e) {
server.abort("Exception parsing unassigned node for region=" + hri.getEncodedName(), e);
}
ServerName addressFromZK = rt != null? rt.getServerName(): null;
if (addressFromZK != null) {
// if we get something from ZK, we will use the data
boolean matchZK = addressFromZK.equals(serverName);
LOG.debug("Checking region=" + hri.getRegionNameAsString() + ", zk server=" + addressFromZK +
" current=" + serverName + ", matches=" + matchZK);
return matchZK;
}
ServerName addressFromAM = regionStates.getRegionServerOfRegion(hri);
boolean matchAM = (addressFromAM != null &&
addressFromAM.equals(serverName));
LOG.debug("based on AM, current region=" + hri.getRegionNameAsString() +
" is on server=" + (addressFromAM != null ? addressFromAM : "null") +
" server being checked: " + serverName);
return matchAM;
}
/**
* Process shutdown server removing any assignments.
* @param sn Server that went down.
* @return list of regions in transition on this server
*/
public List<HRegionInfo> processServerShutdown(final ServerName sn) {
// Clean out any existing assignment plans for this server
synchronized (this.regionPlans) {
for (Iterator <Map.Entry<String, RegionPlan>> i =
this.regionPlans.entrySet().iterator(); i.hasNext();) {
Map.Entry<String, RegionPlan> e = i.next();
ServerName otherSn = e.getValue().getDestination();
// The name will be null if the region is planned for a random assign.
if (otherSn != null && otherSn.equals(sn)) {
// Use iterator's remove else we'll get CME
i.remove();
}
}
}
List<HRegionInfo> regions = regionStates.serverOffline(watcher, sn);
for (Iterator<HRegionInfo> it = regions.iterator(); it.hasNext(); ) {
HRegionInfo hri = it.next();
String encodedName = hri.getEncodedName();
// We need a lock on the region as we could update it
Lock lock = locker.acquireLock(encodedName);
try {
RegionState regionState =
regionStates.getRegionTransitionState(encodedName);
if (regionState == null
|| (regionState.getServerName() != null && !regionState.isOnServer(sn))
|| !(regionState.isFailedClose() || regionState.isOffline()
|| regionState.isPendingOpenOrOpening())) {
LOG.info("Skip " + regionState + " since it is not opening/failed_close"
+ " on the dead server any more: " + sn);
it.remove();
} else {
try {
// Delete the ZNode if exists
ZKAssign.deleteNodeFailSilent(watcher, hri);
} catch (KeeperException ke) {
server.abort("Unexpected ZK exception deleting node " + hri, ke);
}
if (zkTable.isDisablingOrDisabledTable(hri.getTable())) {
regionStates.regionOffline(hri);
it.remove();
continue;
}
// Mark the region offline and assign it again by SSH
regionStates.updateRegionState(hri, State.OFFLINE);
}
} finally {
lock.unlock();
}
}
return regions;
}
/**
* @param plan Plan to execute.
*/
public void balance(final RegionPlan plan) {
HRegionInfo hri = plan.getRegionInfo();
TableName tableName = hri.getTable();
if (zkTable.isDisablingOrDisabledTable(tableName)) {
LOG.info("Ignored moving region of disabling/disabled table "
+ tableName);
return;
}
// Move the region only if it's assigned
String encodedName = hri.getEncodedName();
ReentrantLock lock = locker.acquireLock(encodedName);
try {
if (!regionStates.isRegionOnline(hri)) {
RegionState state = regionStates.getRegionState(encodedName);
LOG.info("Ignored moving region not assigned: " + hri + ", "
+ (state == null ? "not in region states" : state));
return;
}
synchronized (this.regionPlans) {
this.regionPlans.put(plan.getRegionName(), plan);
}
unassign(hri, false, plan.getDestination());
} finally {
lock.unlock();
}
}
public void stop() {
if (tomActivated){
this.timeoutMonitor.interrupt();
this.timerUpdater.interrupt();
}
}
/**
* Shutdown the threadpool executor service
*/
public void shutdown() {
// It's an immediate shutdown, so we're clearing the remaining tasks.
synchronized (zkEventWorkerWaitingList){
zkEventWorkerWaitingList.clear();
}
threadPoolExecutorService.shutdownNow();
zkEventWorkers.shutdownNow();
}
protected void setEnabledTable(TableName tableName) {
try {
this.zkTable.setEnabledTable(tableName);
} catch (KeeperException e) {
// here we can abort as it is the start up flow
String errorMsg = "Unable to ensure that the table " + tableName
+ " will be" + " enabled because of a ZooKeeper issue";
LOG.error(errorMsg);
this.server.abort(errorMsg, e);
}
}
/**
* Set region as OFFLINED up in zookeeper asynchronously.
* @param state
* @return True if we succeeded, false otherwise (State was incorrect or failed
* updating zk).
*/
private boolean asyncSetOfflineInZooKeeper(final RegionState state,
final AsyncCallback.StringCallback cb, final ServerName destination) {
if (!state.isClosed() && !state.isOffline()) {
this.server.abort("Unexpected state trying to OFFLINE; " + state,
new IllegalStateException());
return false;
}
regionStates.updateRegionState(state.getRegion(), State.OFFLINE);
try {
ZKAssign.asyncCreateNodeOffline(watcher, state.getRegion(),
destination, cb, state);
} catch (KeeperException e) {
if (e instanceof NodeExistsException) {
LOG.warn("Node for " + state.getRegion() + " already exists");
} else {
server.abort("Unexpected ZK exception creating/setting node OFFLINE", e);
}
return false;
}
return true;
}
private boolean deleteNodeInStates(String encodedName,
String desc, ServerName sn, EventType... types) {
try {
for (EventType et: types) {
if (ZKAssign.deleteNode(watcher, encodedName, et, sn)) {
return true;
}
}
LOG.info("Failed to delete the " + desc + " node for "
+ encodedName + ". The node type may not match");
} catch (NoNodeException e) {
if (LOG.isDebugEnabled()) {
LOG.debug("The " + desc + " node for " + encodedName + " already deleted");
}
} catch (KeeperException ke) {
server.abort("Unexpected ZK exception deleting " + desc
+ " node for the region " + encodedName, ke);
}
return false;
}
private void deleteMergingNode(String encodedName, ServerName sn) {
deleteNodeInStates(encodedName, "merging", sn, EventType.RS_ZK_REGION_MERGING,
EventType.RS_ZK_REQUEST_REGION_MERGE, EventType.RS_ZK_REGION_MERGED);
}
private void deleteSplittingNode(String encodedName, ServerName sn) {
deleteNodeInStates(encodedName, "splitting", sn, EventType.RS_ZK_REGION_SPLITTING,
EventType.RS_ZK_REQUEST_REGION_SPLIT, EventType.RS_ZK_REGION_SPLIT);
}
/**
* A helper to handle region merging transition event.
* It transitions merging regions to MERGING state.
*/
private boolean handleRegionMerging(final RegionTransition rt, final String encodedName,
final String prettyPrintedRegionName, final ServerName sn) {
if (!serverManager.isServerOnline(sn)) {
LOG.warn("Dropped merging! ServerName=" + sn + " unknown.");
return false;
}
byte [] payloadOfMerging = rt.getPayload();
List<HRegionInfo> mergingRegions;
try {
mergingRegions = HRegionInfo.parseDelimitedFrom(
payloadOfMerging, 0, payloadOfMerging.length);
} catch (IOException e) {
LOG.error("Dropped merging! Failed reading " + rt.getEventType()
+ " payload for " + prettyPrintedRegionName);
return false;
}
assert mergingRegions.size() == 3;
HRegionInfo p = mergingRegions.get(0);
HRegionInfo hri_a = mergingRegions.get(1);
HRegionInfo hri_b = mergingRegions.get(2);
RegionState rs_p = regionStates.getRegionState(p);
RegionState rs_a = regionStates.getRegionState(hri_a);
RegionState rs_b = regionStates.getRegionState(hri_b);
if (!((rs_a == null || rs_a.isOpenOrMergingOnServer(sn))
&& (rs_b == null || rs_b.isOpenOrMergingOnServer(sn))
&& (rs_p == null || rs_p.isOpenOrMergingNewOnServer(sn)))) {
LOG.warn("Dropped merging! Not in state good for MERGING; rs_p="
+ rs_p + ", rs_a=" + rs_a + ", rs_b=" + rs_b);
return false;
}
EventType et = rt.getEventType();
if (et == EventType.RS_ZK_REQUEST_REGION_MERGE) {
try {
if (RegionMergeTransaction.transitionMergingNode(watcher, p,
hri_a, hri_b, sn, -1, EventType.RS_ZK_REQUEST_REGION_MERGE,
EventType.RS_ZK_REGION_MERGING) == -1) {
byte[] data = ZKAssign.getData(watcher, encodedName);
EventType currentType = null;
if (data != null) {
RegionTransition newRt = RegionTransition.parseFrom(data);
currentType = newRt.getEventType();
}
if (currentType == null || (currentType != EventType.RS_ZK_REGION_MERGED
&& currentType != EventType.RS_ZK_REGION_MERGING)) {
LOG.warn("Failed to transition pending_merge node "
+ encodedName + " to merging, it's now " + currentType);
return false;
}
}
} catch (Exception e) {
LOG.warn("Failed to transition pending_merge node "
+ encodedName + " to merging", e);
return false;
}
}
synchronized (regionStates) {
regionStates.updateRegionState(hri_a, State.MERGING);
regionStates.updateRegionState(hri_b, State.MERGING);
regionStates.updateRegionState(p, State.MERGING_NEW, sn);
if (et != EventType.RS_ZK_REGION_MERGED) {
regionStates.regionOffline(p, State.MERGING_NEW);
this.mergingRegions.put(encodedName,
new PairOfSameType<HRegionInfo>(hri_a, hri_b));
} else {
this.mergingRegions.remove(encodedName);
regionOffline(hri_a, State.MERGED);
regionOffline(hri_b, State.MERGED);
regionOnline(p, sn);
}
}
if (et == EventType.RS_ZK_REGION_MERGED) {
LOG.debug("Handling MERGED event for " + encodedName + "; deleting node");
// Remove region from ZK
try {
boolean successful = false;
while (!successful) {
// It's possible that the RS tickles in between the reading of the
// znode and the deleting, so it's safe to retry.
successful = ZKAssign.deleteNode(watcher, encodedName,
EventType.RS_ZK_REGION_MERGED, sn);
}
} catch (KeeperException e) {
if (e instanceof NoNodeException) {
String znodePath = ZKUtil.joinZNode(watcher.splitLogZNode, encodedName);
LOG.debug("The znode " + znodePath + " does not exist. May be deleted already.");
} else {
server.abort("Error deleting MERGED node " + encodedName, e);
}
}
LOG.info("Handled MERGED event; merged=" + p.getRegionNameAsString()
+ ", region_a=" + hri_a.getRegionNameAsString() + ", region_b="
+ hri_b.getRegionNameAsString() + ", on " + sn);
// User could disable the table before master knows the new region.
if (zkTable.isDisablingOrDisabledTable(p.getTable())) {
unassign(p);
}
}
return true;
}
/**
* A helper to handle region splitting transition event.
*/
private boolean handleRegionSplitting(final RegionTransition rt, final String encodedName,
final String prettyPrintedRegionName, final ServerName sn) {
if (!serverManager.isServerOnline(sn)) {
LOG.warn("Dropped splitting! ServerName=" + sn + " unknown.");
return false;
}
byte [] payloadOfSplitting = rt.getPayload();
List<HRegionInfo> splittingRegions;
try {
splittingRegions = HRegionInfo.parseDelimitedFrom(
payloadOfSplitting, 0, payloadOfSplitting.length);
} catch (IOException e) {
LOG.error("Dropped splitting! Failed reading " + rt.getEventType()
+ " payload for " + prettyPrintedRegionName);
return false;
}
assert splittingRegions.size() == 2;
HRegionInfo hri_a = splittingRegions.get(0);
HRegionInfo hri_b = splittingRegions.get(1);
RegionState rs_p = regionStates.getRegionState(encodedName);
RegionState rs_a = regionStates.getRegionState(hri_a);
RegionState rs_b = regionStates.getRegionState(hri_b);
if (!((rs_p == null || rs_p.isOpenOrSplittingOnServer(sn))
&& (rs_a == null || rs_a.isOpenOrSplittingNewOnServer(sn))
&& (rs_b == null || rs_b.isOpenOrSplittingNewOnServer(sn)))) {
LOG.warn("Dropped splitting! Not in state good for SPLITTING; rs_p="
+ rs_p + ", rs_a=" + rs_a + ", rs_b=" + rs_b);
return false;
}
if (rs_p == null) {
// Splitting region should be online
rs_p = regionStates.updateRegionState(rt, State.OPEN);
if (rs_p == null) {
LOG.warn("Received splitting for region " + prettyPrintedRegionName
+ " from server " + sn + " but it doesn't exist anymore,"
+ " probably already processed its split");
return false;
}
regionStates.regionOnline(rs_p.getRegion(), sn);
}
HRegionInfo p = rs_p.getRegion();
EventType et = rt.getEventType();
if (et == EventType.RS_ZK_REQUEST_REGION_SPLIT) {
try {
if (SplitTransaction.transitionSplittingNode(watcher, p,
hri_a, hri_b, sn, -1, EventType.RS_ZK_REQUEST_REGION_SPLIT,
EventType.RS_ZK_REGION_SPLITTING) == -1) {
byte[] data = ZKAssign.getData(watcher, encodedName);
EventType currentType = null;
if (data != null) {
RegionTransition newRt = RegionTransition.parseFrom(data);
currentType = newRt.getEventType();
}
if (currentType == null || (currentType != EventType.RS_ZK_REGION_SPLIT
&& currentType != EventType.RS_ZK_REGION_SPLITTING)) {
LOG.warn("Failed to transition pending_split node "
+ encodedName + " to splitting, it's now " + currentType);
return false;
}
}
} catch (Exception e) {
LOG.warn("Failed to transition pending_split node "
+ encodedName + " to splitting", e);
return false;
}
}
synchronized (regionStates) {
regionStates.updateRegionState(hri_a, State.SPLITTING_NEW, sn);
regionStates.updateRegionState(hri_b, State.SPLITTING_NEW, sn);
regionStates.regionOffline(hri_a, State.SPLITTING_NEW);
regionStates.regionOffline(hri_b, State.SPLITTING_NEW);
regionStates.updateRegionState(rt, State.SPLITTING);
// The below is for testing ONLY! We can't do fault injection easily, so
// resort to this kinda uglyness -- St.Ack 02/25/2011.
if (TEST_SKIP_SPLIT_HANDLING) {
LOG.warn("Skipping split message, TEST_SKIP_SPLIT_HANDLING is set");
return true; // return true so that the splitting node stays
}
if (et == EventType.RS_ZK_REGION_SPLIT) {
regionOffline(p, State.SPLIT);
regionOnline(hri_a, sn);
regionOnline(hri_b, sn);
}
}
if (et == EventType.RS_ZK_REGION_SPLIT) {
LOG.debug("Handling SPLIT event for " + encodedName + "; deleting node");
// Remove region from ZK
try {
boolean successful = false;
while (!successful) {
// It's possible that the RS tickles in between the reading of the
// znode and the deleting, so it's safe to retry.
successful = ZKAssign.deleteNode(watcher, encodedName,
EventType.RS_ZK_REGION_SPLIT, sn);
}
} catch (KeeperException e) {
if (e instanceof NoNodeException) {
String znodePath = ZKUtil.joinZNode(watcher.splitLogZNode, encodedName);
LOG.debug("The znode " + znodePath + " does not exist. May be deleted already.");
} else {
server.abort("Error deleting SPLIT node " + encodedName, e);
}
}
LOG.info("Handled SPLIT event; parent=" + p.getRegionNameAsString()
+ ", daughter a=" + hri_a.getRegionNameAsString() + ", daughter b="
+ hri_b.getRegionNameAsString() + ", on " + sn);
// User could disable the table before master knows the new region.
if (zkTable.isDisablingOrDisabledTable(p.getTable())) {
unassign(hri_a);
unassign(hri_b);
}
}
return true;
}
/**
* A region is offline. The new state should be the specified one,
* if not null. If the specified state is null, the new state is Offline.
* The specified state can be Split/Merged/Offline/null only.
*/
private void regionOffline(final HRegionInfo regionInfo, final State state) {
regionStates.regionOffline(regionInfo, state);
removeClosedRegion(regionInfo);
// remove the region plan as well just in case.
clearRegionPlan(regionInfo);
}
/**
* @return Instance of load balancer
*/
public LoadBalancer getBalancer() {
return this.balancer;
}
}