HBASE-21025 Add cache for TableStateManager

This commit is contained in:
zhangduo 2018-08-10 14:03:28 +08:00
parent 28635d6101
commit 397388316e
5 changed files with 184 additions and 146 deletions

View File

@ -201,7 +201,7 @@ public class BucketCache implements BlockCache, HeapSize {
* Key set of offsets in BucketCache is limited so soft reference is the best choice here.
*/
@VisibleForTesting
final IdReadWriteLock offsetLock = new IdReadWriteLock(ReferenceType.SOFT);
final IdReadWriteLock<Long> offsetLock = new IdReadWriteLock<>(ReferenceType.SOFT);
private final NavigableSet<BlockCacheKey> blocksByHFile =
new ConcurrentSkipListSet<>(new Comparator<BlockCacheKey>() {

View File

@ -20,6 +20,7 @@ package org.apache.hadoop.hbase.master;
import java.io.IOException;
import org.apache.hadoop.hbase.client.TableState;
import org.apache.hadoop.hbase.client.TableState.State;
import org.apache.hadoop.hbase.protobuf.ProtobufUtil;
import org.apache.hadoop.hbase.protobuf.generated.ZooKeeperProtos;
import org.apache.hadoop.hbase.zookeeper.ZKUtil;
@ -45,8 +46,8 @@ public class MirroringTableStateManager extends TableStateManager {
private static final Logger LOG = LoggerFactory.getLogger(MirroringTableStateManager.class);
/**
* Set this key to true in Configuration to enable mirroring of table state out to zookeeper
* so hbase-1.x clients can pick-up table state.
* Set this key to true in Configuration to enable mirroring of table state out to zookeeper so
* hbase-1.x clients can pick-up table state.
*/
static final String MIRROR_TABLE_STATE_TO_ZK_KEY = "hbase.mirror.table.state.to.zookeeper";
@ -54,26 +55,14 @@ public class MirroringTableStateManager extends TableStateManager {
super(master);
}
protected void updateMetaState(TableName tableName, TableState.State newState)
throws IOException {
// Take the lock. Its reentrant. Calls to super will take same lock.
lock.writeLock().lock();
try {
super.updateMetaState(tableName, newState);
@Override
protected void metaStateUpdated(TableName tableName, State newState) throws IOException {
updateZooKeeper(new TableState(tableName, newState));
} finally {
lock.writeLock().unlock();
}
}
public void setDeletedTable(TableName tableName) throws IOException {
lock.writeLock().lock();
try {
super.setDeletedTable(tableName);
@Override
protected void metaStateDeleted(TableName tableName) throws IOException {
deleteZooKeeper(tableName);
} finally {
lock.writeLock().unlock();
}
}
private void updateZooKeeper(TableState tableState) throws IOException {
@ -90,8 +79,8 @@ public class MirroringTableStateManager extends TableStateManager {
// Now set newState
ZooKeeperProtos.DeprecatedTableState.Builder builder =
ZooKeeperProtos.DeprecatedTableState.newBuilder();
builder.setState(ZooKeeperProtos.DeprecatedTableState.State.
valueOf(tableState.getState().toString()));
builder.setState(
ZooKeeperProtos.DeprecatedTableState.State.valueOf(tableState.getState().toString()));
byte[] data = ProtobufUtil.prependPBMagic(builder.build().toByteArray());
ZKUtil.setData(this.master.getZooKeeper(), znode, data);
} catch (KeeperException e) {

View File

@ -1,4 +1,4 @@
/*
/**
* 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
@ -17,65 +17,68 @@
*/
package org.apache.hadoop.hbase.master;
import edu.umd.cs.findbugs.annotations.NonNull;
import edu.umd.cs.findbugs.annotations.Nullable;
import java.io.IOException;
import java.util.HashMap;
import java.util.Map;
import java.util.Set;
import java.util.concurrent.ConcurrentHashMap;
import java.util.concurrent.ConcurrentMap;
import java.util.concurrent.locks.ReadWriteLock;
import java.util.concurrent.locks.ReentrantReadWriteLock;
import org.apache.hadoop.hbase.client.TableDescriptor;
import org.apache.hadoop.hbase.exceptions.IllegalArgumentIOException;
import org.apache.hadoop.hbase.util.ZKDataMigrator;
import org.apache.hadoop.hbase.zookeeper.ZKUtil;
import org.apache.hadoop.hbase.zookeeper.ZNodePaths;
import org.apache.hbase.thirdparty.com.google.common.collect.Sets;
import edu.umd.cs.findbugs.annotations.NonNull;
import edu.umd.cs.findbugs.annotations.Nullable;
import org.apache.hadoop.hbase.MetaTableAccessor;
import org.apache.hadoop.hbase.TableDescriptors;
import org.apache.hadoop.hbase.TableName;
import org.apache.hadoop.hbase.TableNotFoundException;
import org.apache.hadoop.hbase.client.Connection;
import org.apache.hadoop.hbase.client.Result;
import org.apache.hadoop.hbase.client.TableDescriptor;
import org.apache.hadoop.hbase.client.TableState;
import org.apache.hadoop.hbase.exceptions.IllegalArgumentIOException;
import org.apache.hadoop.hbase.util.IdReadWriteLock;
import org.apache.hadoop.hbase.util.ZKDataMigrator;
import org.apache.hadoop.hbase.zookeeper.ZKUtil;
import org.apache.hadoop.hbase.zookeeper.ZNodePaths;
import org.apache.yetus.audience.InterfaceAudience;
import org.apache.zookeeper.KeeperException;
import org.slf4j.Logger;
import org.slf4j.LoggerFactory;
import org.apache.hadoop.hbase.client.Connection;
import org.apache.hadoop.hbase.client.Result;
import org.apache.hadoop.hbase.client.TableState;
import org.apache.hbase.thirdparty.com.google.common.collect.Sets;
/**
* This is a helper class used to manage table states.
* This class uses hbase:meta as its store for table state so hbase:meta must be online before
* {@link #start()} is called.
* TODO: Cache state. Cut down on meta looksups.
* This is a helper class used to manage table states. This class uses hbase:meta as its store for
* table state so hbase:meta must be online before {@link #start()} is called.
*/
// TODO: Make this a guava Service
@InterfaceAudience.Private
public class TableStateManager {
private static final Logger LOG = LoggerFactory.getLogger(TableStateManager.class);
/**
* Set this key to false in Configuration to disable migrating table state from zookeeper
* so hbase:meta table.
* Set this key to false in Configuration to disable migrating table state from zookeeper so
* hbase:meta table.
*/
static final String MIGRATE_TABLE_STATE_FROM_ZK_KEY = "hbase.migrate.table.state.from.zookeeper";
private static final String MIGRATE_TABLE_STATE_FROM_ZK_KEY =
"hbase.migrate.table.state.from.zookeeper";
final ReadWriteLock lock = new ReentrantReadWriteLock();
final MasterServices master;
private final IdReadWriteLock<TableName> tnLock = new IdReadWriteLock<>();
protected final MasterServices master;
private final ConcurrentMap<TableName, TableState.State> tableName2State =
new ConcurrentHashMap<>();
public TableStateManager(MasterServices master) {
this.master = master;
}
/**
* Set table state to provided.
* Caller should lock table on write.
* Set table state to provided. Caller should lock table on write.
* @param tableName table to change state for
* @param newState new state
* @throws IOException
*/
public void setTableState(TableName tableName, TableState.State newState) throws IOException {
ReadWriteLock lock = tnLock.getLock(tableName);
lock.writeLock().lock();
try {
updateMetaState(tableName, newState);
@ -85,18 +88,16 @@ public class TableStateManager {
}
/**
* Set table state to provided but only if table in specified states
* Caller should lock table on write.
* Set table state to provided but only if table in specified states Caller should lock table on
* write.
* @param tableName table to change state for
* @param newState new state
* @param states states to check against
* @return null if succeed or table state if failed
* @throws IOException
*/
public TableState setTableStateIfInStates(TableName tableName,
TableState.State newState,
TableState.State... states)
throws IOException {
public TableState setTableStateIfInStates(TableName tableName, TableState.State newState,
TableState.State... states) throws IOException {
ReadWriteLock lock = tnLock.getLock(tableName);
lock.writeLock().lock();
try {
TableState currentState = readMetaState(tableName);
@ -115,17 +116,17 @@ public class TableStateManager {
}
/**
* Set table state to provided but only if table not in specified states
* Caller should lock table on write.
* Set table state to provided but only if table not in specified states Caller should lock table
* on write.
* @param tableName table to change state for
* @param newState new state
* @param states states to check against
* @throws IOException
*/
public boolean setTableStateIfNotInStates(TableName tableName,
TableState.State newState,
TableState.State... states)
throws IOException {
public boolean setTableStateIfNotInStates(TableName tableName, TableState.State newState,
TableState.State... states) throws IOException {
ReadWriteLock lock = tnLock.getLock(tableName);
lock.writeLock().lock();
try {
TableState currentState = readMetaState(tableName);
if (currentState == null) {
throw new TableNotFoundException(tableName);
@ -136,6 +137,9 @@ public class TableStateManager {
} else {
return false;
}
} finally {
lock.writeLock().unlock();
}
}
public boolean isTableState(TableName tableName, TableState.State... states) {
@ -144,6 +148,7 @@ public class TableStateManager {
return tableState.isInStates(states);
} catch (IOException e) {
LOG.error("Unable to get table " + tableName + " state", e);
// XXX: is it safe to just return false here?
return false;
}
}
@ -152,28 +157,42 @@ public class TableStateManager {
if (tableName.equals(TableName.META_TABLE_NAME)) {
return;
}
ReadWriteLock lock = tnLock.getLock(tableName);
lock.writeLock().lock();
try {
MetaTableAccessor.deleteTableState(master.getConnection(), tableName);
metaStateDeleted(tableName);
} finally {
tableName2State.remove(tableName);
lock.writeLock().unlock();
}
}
public boolean isTablePresent(TableName tableName) throws IOException {
ReadWriteLock lock = tnLock.getLock(tableName);
lock.readLock().lock();
try {
return readMetaState(tableName) != null;
} finally {
lock.readLock().unlock();
}
}
/**
* Return all tables in given states.
*
* @param states filter by states
* @return tables in given states
* @throws IOException
*/
public Set<TableName> getTablesInStates(final TableState.State... states) throws IOException {
public Set<TableName> getTablesInStates(TableState.State... states) throws IOException {
// Only be called in region normalizer, will not use cache.
final Set<TableName> rv = Sets.newHashSet();
MetaTableAccessor.fullScanTables(master.getConnection(), new MetaTableAccessor.Visitor() {
@Override
public boolean visit(Result r) throws IOException {
TableState tableState = MetaTableAccessor.getTableState(r);
if (tableState != null && tableState.inStates(states))
if (tableState != null && tableState.inStates(states)) {
rv.add(tableState.getTableName());
}
return true;
}
});
@ -188,15 +207,20 @@ public class TableStateManager {
@NonNull
public TableState getTableState(TableName tableName) throws IOException {
ReadWriteLock lock = tnLock.getLock(tableName);
lock.readLock().lock();
try {
TableState currentState = readMetaState(tableName);
if (currentState == null) {
throw new TableStateNotFoundException(tableName);
}
return currentState;
} finally {
lock.readLock().unlock();
}
}
protected void updateMetaState(TableName tableName, TableState.State newState)
throws IOException {
private void updateMetaState(TableName tableName, TableState.State newState) throws IOException {
if (tableName.equals(TableName.META_TABLE_NAME)) {
if (TableState.State.DISABLING.equals(newState) ||
TableState.State.DISABLED.equals(newState)) {
@ -205,12 +229,36 @@ public class TableStateManager {
// Otherwise, just return; no need to set ENABLED on meta -- it is always ENABLED.
return;
}
boolean succ = false;
try {
MetaTableAccessor.updateTableState(master.getConnection(), tableName, newState);
tableName2State.put(tableName, newState);
} finally {
if (!succ) {
tableName2State.remove(tableName);
}
}
metaStateUpdated(tableName, newState);
}
protected void metaStateUpdated(TableName tableName, TableState.State newState)
throws IOException {
}
protected void metaStateDeleted(TableName tableName) throws IOException {
}
@Nullable
protected TableState readMetaState(TableName tableName) throws IOException {
return MetaTableAccessor.getTableState(master.getConnection(), tableName);
private TableState readMetaState(TableName tableName) throws IOException {
TableState.State state = tableName2State.get(tableName);
if (state != null) {
return new TableState(tableName, state);
}
TableState tableState = MetaTableAccessor.getTableState(master.getConnection(), tableName);
if (tableState != null) {
tableName2State.putIfAbsent(tableName, tableState.getState());
}
return tableState;
}
public void start() throws IOException {
@ -222,8 +270,8 @@ public class TableStateManager {
private void fixTableStates(TableDescriptors tableDescriptors, Connection connection)
throws IOException {
final Map<String, TableDescriptor> allDescriptors = tableDescriptors.getAll();
final Map<String, TableState> states = new HashMap<>();
Map<String, TableDescriptor> allDescriptors = tableDescriptors.getAll();
Map<String, TableState> states = new HashMap<>();
// NOTE: Ful hbase:meta table scan!
MetaTableAccessor.fullScanTables(connection, new MetaTableAccessor.Visitor() {
@Override
@ -234,21 +282,22 @@ public class TableStateManager {
}
});
for (Map.Entry<String, TableDescriptor> entry : allDescriptors.entrySet()) {
String table = entry.getKey();
if (table.equals(TableName.META_TABLE_NAME.getNameAsString())) {
TableName tableName = TableName.valueOf(entry.getKey());
if (TableName.isMetaTableName(tableName)) {
// This table is always enabled. No fixup needed. No entry in hbase:meta needed.
// Call through to fixTableState though in case a super class wants to do something.
fixTableState(new TableState(TableName.valueOf(table), TableState.State.ENABLED));
fixTableState(new TableState(tableName, TableState.State.ENABLED));
continue;
}
TableState tableState = states.get(table);
TableState tableState = states.get(entry.getKey());
if (tableState == null) {
LOG.warn(table + " has no table state in hbase:meta, assuming ENABLED");
MetaTableAccessor.updateTableState(connection, TableName.valueOf(table),
TableState.State.ENABLED);
fixTableState(new TableState(TableName.valueOf(table), TableState.State.ENABLED));
LOG.warn(tableName + " has no table state in hbase:meta, assuming ENABLED");
MetaTableAccessor.updateTableState(connection, tableName, TableState.State.ENABLED);
fixTableState(new TableState(tableName, TableState.State.ENABLED));
tableName2State.put(tableName, TableState.State.ENABLED);
} else {
fixTableState(tableState);
tableName2State.put(tableName, tableState.getState());
}
}
}
@ -256,27 +305,27 @@ public class TableStateManager {
/**
* For subclasses in case they want to do fixup post hbase:meta.
*/
protected void fixTableState(TableState tableState) throws IOException {}
protected void fixTableState(TableState tableState) throws IOException {
}
/**
* This code is for case where a hbase2 Master is starting for the first time. ZooKeeper is
* where we used to keep table state. On first startup, read zookeeper and update hbase:meta
* with the table states found in zookeeper. This is tricky as we'll do this check every time we
* startup until mirroring is disabled. See the {@link #MIGRATE_TABLE_STATE_FROM_ZK_KEY} flag.
* Original form of this migration came in with HBASE-13032. It deleted all znodes when done.
* We can't do that if we want to support hbase-1.x clients who need to be able to read table
* state out of zk. See {@link MirroringTableStateManager}.
* This code is for case where a hbase2 Master is starting for the first time. ZooKeeper is where
* we used to keep table state. On first startup, read zookeeper and update hbase:meta with the
* table states found in zookeeper. This is tricky as we'll do this check every time we startup
* until mirroring is disabled. See the {@link #MIGRATE_TABLE_STATE_FROM_ZK_KEY} flag. Original
* form of this migration came in with HBASE-13032. It deleted all znodes when done. We can't do
* that if we want to support hbase-1.x clients who need to be able to read table state out of zk.
* See {@link MirroringTableStateManager}.
* @deprecated Since 2.0.0. Remove in hbase-3.0.0.
*/
@Deprecated
private void migrateZooKeeper() throws IOException {
if (!this.master.getConfiguration().getBoolean(MIGRATE_TABLE_STATE_FROM_ZK_KEY,
true)) {
if (!this.master.getConfiguration().getBoolean(MIGRATE_TABLE_STATE_FROM_ZK_KEY, true)) {
return;
}
try {
for (Map.Entry<TableName, TableState.State> entry:
ZKDataMigrator.queryForTableStates(this.master.getZooKeeper()).entrySet()) {
for (Map.Entry<TableName, TableState.State> entry : ZKDataMigrator
.queryForTableStates(this.master.getZooKeeper()).entrySet()) {
if (this.master.getTableDescriptors().get(entry.getKey()) == null) {
deleteZooKeeper(entry.getKey());
LOG.info("Purged table state entry from zookeepr for table not in hbase:meta: " +
@ -313,8 +362,8 @@ public class TableStateManager {
}
/**
* Utility method that knows how to delete the old hbase-1.x table state znode.
* Used also by the Mirroring subclass.
* Utility method that knows how to delete the old hbase-1.x table state znode. Used also by the
* Mirroring subclass.
* @deprecated Since 2.0.0. To be removed in hbase-3.0.0.
*/
@Deprecated

View File

@ -42,14 +42,14 @@ import org.apache.hbase.thirdparty.com.google.common.annotations.VisibleForTesti
* For write lock, use lock.writeLock()
*/
@InterfaceAudience.Private
public class IdReadWriteLock {
public class IdReadWriteLock<T> {
// The number of lock we want to easily support. It's not a maximum.
private static final int NB_CONCURRENT_LOCKS = 1000;
/**
* The pool to get entry from, entries are mapped by {@link Reference} and will be automatically
* garbage-collected by JVM
*/
private final ObjectPool<Long, ReentrantReadWriteLock> lockPool;
private final ObjectPool<T, ReentrantReadWriteLock> lockPool;
private final ReferenceType refType;
public IdReadWriteLock() {
@ -66,18 +66,18 @@ public class IdReadWriteLock {
this.refType = referenceType;
switch (referenceType) {
case SOFT:
lockPool = new SoftObjectPool<>(new ObjectPool.ObjectFactory<Long, ReentrantReadWriteLock>() {
lockPool = new SoftObjectPool<>(new ObjectPool.ObjectFactory<T, ReentrantReadWriteLock>() {
@Override
public ReentrantReadWriteLock createObject(Long id) {
public ReentrantReadWriteLock createObject(T id) {
return new ReentrantReadWriteLock();
}
}, NB_CONCURRENT_LOCKS);
break;
case WEAK:
default:
lockPool = new WeakObjectPool<>(new ObjectPool.ObjectFactory<Long, ReentrantReadWriteLock>() {
lockPool = new WeakObjectPool<>(new ObjectPool.ObjectFactory<T, ReentrantReadWriteLock>() {
@Override
public ReentrantReadWriteLock createObject(Long id) {
public ReentrantReadWriteLock createObject(T id) {
return new ReentrantReadWriteLock();
}
}, NB_CONCURRENT_LOCKS);
@ -92,7 +92,7 @@ public class IdReadWriteLock {
* Get the ReentrantReadWriteLock corresponding to the given id
* @param id an arbitrary number to identify the lock
*/
public ReentrantReadWriteLock getLock(long id) {
public ReentrantReadWriteLock getLock(T id) {
lockPool.purge();
ReentrantReadWriteLock readWriteLock = lockPool.get(id);
return readWriteLock;
@ -113,7 +113,7 @@ public class IdReadWriteLock {
}
@VisibleForTesting
public void waitForWaiters(long id, int numWaiters) throws InterruptedException {
public void waitForWaiters(T id, int numWaiters) throws InterruptedException {
for (ReentrantReadWriteLock readWriteLock;;) {
readWriteLock = lockPool.get(id);
if (readWriteLock != null) {

View File

@ -60,12 +60,12 @@ public class TestIdReadWriteLock {
private static final int NUM_SECONDS = 15;
@Parameterized.Parameter
public IdReadWriteLock idLock;
public IdReadWriteLock<Long> idLock;
@Parameterized.Parameters
public static Iterable<Object[]> data() {
return Arrays.asList(new Object[][] { { new IdReadWriteLock(ReferenceType.WEAK) },
{ new IdReadWriteLock(ReferenceType.SOFT) } });
return Arrays.asList(new Object[][] { { new IdReadWriteLock<Long>(ReferenceType.WEAK) },
{ new IdReadWriteLock<Long>(ReferenceType.SOFT) } });
}
private Map<Long, String> idOwner = new ConcurrentHashMap<>();