HDFS-14731. [FGL] Remove redundant locking on NameNode. Contributed by Konstantin V Shvachko.

(cherry picked from commit ecbcb058b8)
This commit is contained in:
Konstantin V Shvachko 2020-02-21 17:53:37 -08:00
parent 7bfa123d58
commit 4af188587f
7 changed files with 46 additions and 32 deletions

View File

@ -218,7 +218,12 @@ public class BackupImage extends FSImage {
}
lastAppliedTxId = logLoader.getLastAppliedTxId();
getNamesystem().dir.updateCountForQuota();
getNamesystem().writeLock();
try {
getNamesystem().dir.updateCountForQuota();
} finally {
getNamesystem().writeUnlock();
}
} finally {
backupInputStream.clear();
}

View File

@ -187,11 +187,11 @@ public class EncryptionZoneManager {
final int count) throws IOException {
INodesInPath iip;
final FSPermissionChecker pc = dir.getPermissionChecker();
dir.readLock();
dir.getFSNamesystem().readLock();
try {
iip = dir.resolvePath(pc, zone, DirOp.READ);
} finally {
dir.readUnlock();
dir.getFSNamesystem().readUnlock();
}
reencryptionHandler
.pauseForTestingAfterNthCheckpoint(iip.getLastINode().getId(), count);
@ -280,11 +280,11 @@ public class EncryptionZoneManager {
if (getProvider() == null || reencryptionHandler == null) {
return;
}
dir.writeLock();
dir.getFSNamesystem().writeLock();
try {
reencryptionHandler.stopThreads();
} finally {
dir.writeUnlock();
dir.getFSNamesystem().writeUnlock();
}
if (reencryptHandlerExecutor != null) {
reencryptHandlerExecutor.shutdownNow();

View File

@ -382,7 +382,6 @@ final class FSDirEncryptionZoneOp {
static void saveFileXAttrsForBatch(FSDirectory fsd,
List<FileEdekInfo> batch) {
assert fsd.getFSNamesystem().hasWriteLock();
assert !fsd.hasWriteLock();
if (batch != null && !batch.isEmpty()) {
for (FileEdekInfo entry : batch) {
final INode inode = fsd.getInode(entry.getInodeId());
@ -727,13 +726,13 @@ final class FSDirEncryptionZoneOp {
final FSPermissionChecker pc, final String zone) throws IOException {
assert dir.getProvider() != null;
final INodesInPath iip;
dir.readLock();
dir.getFSNamesystem().readLock();
try {
iip = dir.resolvePath(pc, zone, DirOp.READ);
dir.ezManager.checkEncryptionZoneRoot(iip.getLastINode(), zone);
return dir.ezManager.getKeyName(iip);
} finally {
dir.readUnlock();
dir.getFSNamesystem().readUnlock();
}
}
}

View File

@ -81,7 +81,6 @@ import java.util.SortedSet;
import java.util.TreeSet;
import java.util.concurrent.ForkJoinPool;
import java.util.concurrent.RecursiveAction;
import java.util.concurrent.locks.ReentrantReadWriteLock;
import static org.apache.hadoop.fs.CommonConfigurationKeys.FS_PROTECTED_DIRECTORIES;
import static org.apache.hadoop.hdfs.DFSConfigKeys.DFS_NAMENODE_ACCESSTIME_PRECISION_DEFAULT;
@ -170,9 +169,6 @@ public class FSDirectory implements Closeable {
// Each entry in this set must be a normalized path.
private volatile SortedSet<String> protectedDirectories;
// lock to protect the directory and BlockMap
private final ReentrantReadWriteLock dirLock;
private final boolean isPermissionEnabled;
/**
* Support for ACLs is controlled by a configuration flag. If the
@ -212,37 +208,44 @@ public class FSDirectory implements Closeable {
attributeProvider = provider;
}
// utility methods to acquire and release read lock and write lock
/**
* The directory lock dirLock provided redundant locking.
* It has been used whenever namesystem.fsLock was used.
* dirLock is now removed and utility methods to acquire and release dirLock
* remain as placeholders only
*/
void readLock() {
this.dirLock.readLock().lock();
assert namesystem.hasReadLock() : "Should hold namesystem read lock";
}
void readUnlock() {
this.dirLock.readLock().unlock();
assert namesystem.hasReadLock() : "Should hold namesystem read lock";
}
void writeLock() {
this.dirLock.writeLock().lock();
assert namesystem.hasWriteLock() : "Should hold namesystem write lock";
}
void writeUnlock() {
this.dirLock.writeLock().unlock();
assert namesystem.hasWriteLock() : "Should hold namesystem write lock";
}
boolean hasWriteLock() {
return this.dirLock.isWriteLockedByCurrentThread();
return namesystem.hasWriteLock();
}
boolean hasReadLock() {
return this.dirLock.getReadHoldCount() > 0 || hasWriteLock();
return namesystem.hasReadLock();
}
@Deprecated // dirLock is obsolete, use namesystem.fsLock instead
public int getReadHoldCount() {
return this.dirLock.getReadHoldCount();
return namesystem.getReadHoldCount();
}
@Deprecated // dirLock is obsolete, use namesystem.fsLock instead
public int getWriteHoldCount() {
return this.dirLock.getWriteHoldCount();
return namesystem.getWriteHoldCount();
}
@VisibleForTesting
@ -268,7 +271,6 @@ public class FSDirectory implements Closeable {
FSDirectory(FSNamesystem ns, Configuration conf) throws IOException {
// used to enable/disable the use of expanded string tables.
SerialNumberManager.initialize(conf);
this.dirLock = new ReentrantReadWriteLock(true); // fair
this.inodeId = new INodeId();
rootDir = createRoot(ns);
inodeMap = INodeMap.newInstance(rootDir);
@ -1466,12 +1468,7 @@ public class FSDirectory implements Closeable {
* @return The inode associated with the given id
*/
public INode getInode(long id) {
readLock();
try {
return inodeMap.get(id);
} finally {
readUnlock();
}
return inodeMap.get(id);
}
@VisibleForTesting

View File

@ -3604,6 +3604,7 @@ public class FSNamesystem implements Namesystem, FSNamesystemMBean,
@Override
public INodeFile getBlockCollection(long id) {
assert hasReadLock() : "Accessing INode id = " + id + " without read lock";
INode inode = getFSDirectory().getInode(id);
return inode == null ? null : inode.asFile();
}

View File

@ -337,7 +337,7 @@ public class ReencryptionHandler implements Runnable {
}
final Long zoneId;
dir.readLock();
dir.getFSNamesystem().readLock();
try {
zoneId = getReencryptionStatus().getNextUnprocessedZone();
if (zoneId == null) {
@ -349,7 +349,7 @@ public class ReencryptionHandler implements Runnable {
getReencryptionStatus().markZoneStarted(zoneId);
resetSubmissionTracker(zoneId);
} finally {
dir.readUnlock();
dir.getFSNamesystem().readUnlock();
}
try {

View File

@ -503,7 +503,8 @@ public class TestBlocksWithNotEnoughRacks {
BlockInfo storedBlock = bm.getStoredBlock(b.getLocalBlock());
// The block should be replicated OK - so Reconstruction Work will be null
BlockReconstructionWork work = bm.scheduleReconstruction(storedBlock, 2);
BlockReconstructionWork work = scheduleReconstruction(
cluster.getNamesystem(), storedBlock, 2);
assertNull(work);
// Set the upgradeDomain to "3" for the 3 nodes hosting the block.
// Then alternately set the remaining 3 nodes to have an upgradeDomain
@ -519,7 +520,8 @@ public class TestBlocksWithNotEnoughRacks {
}
}
// Now reconWork is non-null and 2 extra targets are needed
work = bm.scheduleReconstruction(storedBlock, 2);
work = scheduleReconstruction(
cluster.getNamesystem(), storedBlock, 2);
assertEquals(2, work.getAdditionalReplRequired());
// Add the block to the replication queue and ensure it is replicated
@ -531,6 +533,16 @@ public class TestBlocksWithNotEnoughRacks {
}
}
static BlockReconstructionWork scheduleReconstruction(
FSNamesystem fsn, BlockInfo block, int priority) {
fsn.writeLock();
try {
return fsn.getBlockManager().scheduleReconstruction(block, priority);
} finally {
fsn.writeUnlock();
}
}
@Test
public void testUnderReplicatedRespectsRacksAndUpgradeDomain()
throws Exception {