blob: d96b718a4338a6fd745f96f3b9e2448a9f4cd843 [file] [log] [blame]
/**
* 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.hdfs.server.namenode;
import junit.framework.TestCase;
import java.io.*;
import java.net.URI;
import java.util.Iterator;
import java.util.concurrent.Callable;
import java.util.concurrent.ExecutorService;
import java.util.concurrent.Executors;
import org.apache.commons.logging.Log;
import org.apache.commons.logging.LogFactory;
import org.apache.hadoop.conf.Configuration;
import org.apache.hadoop.fs.FileSystem;
import org.apache.hadoop.fs.permission.*;
import org.apache.hadoop.hdfs.HdfsConfiguration;
import org.apache.hadoop.hdfs.MiniDFSCluster;
import org.apache.hadoop.hdfs.server.namenode.EditLogFileInputStream;
import org.apache.hadoop.hdfs.server.common.Storage.StorageDirectory;
import org.apache.hadoop.hdfs.server.namenode.FSImage.NameNodeDirType;
import org.apache.hadoop.hdfs.server.namenode.FSImage.NameNodeFile;
import org.apache.hadoop.hdfs.server.namenode.metrics.NameNodeMetrics;
import org.apache.hadoop.metrics.util.MetricsTimeVaryingInt;
import org.mockito.Mockito;
import static org.mockito.Matchers.anyObject;
import static org.mockito.Mockito.doThrow;
import static org.mockito.Mockito.spy;
/**
* This class tests the creation and validation of a checkpoint.
*/
public class TestEditLog extends TestCase {
private static final Log LOG = LogFactory.getLog(TestEditLog.class);
static final int NUM_DATA_NODES = 0;
// This test creates NUM_THREADS threads and each thread does
// 2 * NUM_TRANSACTIONS Transactions concurrently.
static final int NUM_TRANSACTIONS = 100;
static final int NUM_THREADS = 100;
//
// an object that does a bunch of transactions
//
static class Transactions implements Runnable {
FSNamesystem namesystem;
int numTransactions;
short replication = 3;
long blockSize = 64;
Transactions(FSNamesystem ns, int num) {
namesystem = ns;
numTransactions = num;
}
// add a bunch of transactions.
public void run() {
PermissionStatus p = namesystem.createFsOwnerPermissions(
new FsPermission((short)0777));
FSEditLog editLog = namesystem.getEditLog();
for (int i = 0; i < numTransactions; i++) {
INodeFileUnderConstruction inode = new INodeFileUnderConstruction(
p, replication, blockSize, 0, "", "", null);
editLog.logOpenFile("/filename" + i, inode);
editLog.logCloseFile("/filename" + i, inode);
editLog.logSync();
}
}
}
/**
* Tests transaction logging in dfs.
*/
public void testEditLog() throws IOException {
testEditLog(2048);
// force edit buffer to automatically sync on each log of edit log entry
testEditLog(1);
}
/**
* Test edit log with different initial buffer size
*
* @param initialSize initial edit log buffer size
* @throws IOException
*/
private void testEditLog(int initialSize) throws IOException {
// start a cluster
Configuration conf = new HdfsConfiguration();
MiniDFSCluster cluster = null;
FileSystem fileSys = null;
try {
cluster = new MiniDFSCluster.Builder(conf).numDataNodes(NUM_DATA_NODES).build();
cluster.waitActive();
fileSys = cluster.getFileSystem();
final FSNamesystem namesystem = cluster.getNamesystem();
for (Iterator<URI> it = cluster.getNameDirs().iterator(); it.hasNext(); ) {
File dir = new File(it.next().getPath());
System.out.println(dir);
}
FSImage fsimage = namesystem.getFSImage();
FSEditLog editLog = fsimage.getEditLog();
// set small size of flush buffer
editLog.setBufferCapacity(initialSize);
editLog.close();
editLog.open();
// Create threads and make them run transactions concurrently.
Thread threadId[] = new Thread[NUM_THREADS];
for (int i = 0; i < NUM_THREADS; i++) {
Transactions trans = new Transactions(namesystem, NUM_TRANSACTIONS);
threadId[i] = new Thread(trans, "TransactionThread-" + i);
threadId[i].start();
}
// wait for all transactions to get over
for (int i = 0; i < NUM_THREADS; i++) {
try {
threadId[i].join();
} catch (InterruptedException e) {
i--; // retry
}
}
editLog.close();
editLog.open();
// Verify that we can read in all the transactions that we have written.
// If there were any corruptions, it is likely that the reading in
// of these transactions will throw an exception.
//
FSEditLogLoader loader = new FSEditLogLoader(namesystem);
for (Iterator<StorageDirectory> it =
fsimage.dirIterator(NameNodeDirType.EDITS); it.hasNext();) {
File editFile = FSImage.getImageFile(it.next(), NameNodeFile.EDITS);
System.out.println("Verifying file: " + editFile);
int numEdits = loader.loadFSEdits(
new EditLogFileInputStream(editFile));
int numLeases = namesystem.leaseManager.countLease();
System.out.println("Number of outstanding leases " + numLeases);
assertEquals(0, numLeases);
assertTrue("Verification for " + editFile + " failed. " +
"Expected " + (NUM_THREADS * 2 * NUM_TRANSACTIONS) + " transactions. "+
"Found " + numEdits + " transactions.",
numEdits == NUM_THREADS * 2 * NUM_TRANSACTIONS);
}
} finally {
if(fileSys != null) fileSys.close();
if(cluster != null) cluster.shutdown();
}
}
private void doLogEdit(ExecutorService exec, final FSEditLog log,
final String filename) throws Exception
{
exec.submit(new Callable<Void>() {
public Void call() {
log.logSetReplication(filename, (short)1);
return null;
}
}).get();
}
private void doCallLogSync(ExecutorService exec, final FSEditLog log)
throws Exception
{
exec.submit(new Callable<Void>() {
public Void call() {
log.logSync();
return null;
}
}).get();
}
private void doCallLogSyncAll(ExecutorService exec, final FSEditLog log)
throws Exception
{
exec.submit(new Callable<Void>() {
public Void call() throws Exception {
log.logSyncAll();
return null;
}
}).get();
}
public void testSyncBatching() throws Exception {
// start a cluster
Configuration conf = new HdfsConfiguration();
MiniDFSCluster cluster = null;
FileSystem fileSys = null;
ExecutorService threadA = Executors.newSingleThreadExecutor();
ExecutorService threadB = Executors.newSingleThreadExecutor();
try {
cluster = new MiniDFSCluster.Builder(conf).numDataNodes(NUM_DATA_NODES).build();
cluster.waitActive();
fileSys = cluster.getFileSystem();
final FSNamesystem namesystem = cluster.getNamesystem();
FSImage fsimage = namesystem.getFSImage();
final FSEditLog editLog = fsimage.getEditLog();
assertEquals("should start with no txids synced",
0, editLog.getSyncTxId());
// Log an edit from thread A
doLogEdit(threadA, editLog, "thread-a 1");
assertEquals("logging edit without syncing should do not affect txid",
0, editLog.getSyncTxId());
// Log an edit from thread B
doLogEdit(threadB, editLog, "thread-b 1");
assertEquals("logging edit without syncing should do not affect txid",
0, editLog.getSyncTxId());
// Now ask to sync edit from B, which should sync both edits.
doCallLogSync(threadB, editLog);
assertEquals("logSync from second thread should bump txid up to 2",
2, editLog.getSyncTxId());
// Now ask to sync edit from A, which was already batched in - thus
// it should increment the batch count metric
NameNodeMetrics metrics = NameNode.getNameNodeMetrics();
metrics.transactionsBatchedInSync = Mockito.mock(MetricsTimeVaryingInt.class);
doCallLogSync(threadA, editLog);
assertEquals("logSync from first thread shouldn't change txid",
2, editLog.getSyncTxId());
//Should have incremented the batch count exactly once
Mockito.verify(metrics.transactionsBatchedInSync,
Mockito.times(1)).inc();
} finally {
threadA.shutdown();
threadB.shutdown();
if(fileSys != null) fileSys.close();
if(cluster != null) cluster.shutdown();
}
}
/**
* Test what happens with the following sequence:
*
* Thread A writes edit
* Thread B calls logSyncAll
* calls close() on stream
* Thread A calls logSync
*
* This sequence is legal and can occur if enterSafeMode() is closely
* followed by saveNamespace.
*/
public void testBatchedSyncWithClosedLogs() throws Exception {
// start a cluster
Configuration conf = new HdfsConfiguration();
MiniDFSCluster cluster = null;
FileSystem fileSys = null;
ExecutorService threadA = Executors.newSingleThreadExecutor();
ExecutorService threadB = Executors.newSingleThreadExecutor();
try {
cluster = new MiniDFSCluster.Builder(conf).numDataNodes(NUM_DATA_NODES).build();
cluster.waitActive();
fileSys = cluster.getFileSystem();
final FSNamesystem namesystem = cluster.getNamesystem();
FSImage fsimage = namesystem.getFSImage();
final FSEditLog editLog = fsimage.getEditLog();
// Log an edit from thread A
doLogEdit(threadA, editLog, "thread-a 1");
assertEquals("logging edit without syncing should do not affect txid",
0, editLog.getSyncTxId());
// logSyncAll in Thread B
doCallLogSyncAll(threadB, editLog);
assertEquals("logSyncAll should sync thread A's transaction",
1, editLog.getSyncTxId());
// Close edit log
editLog.close();
// Ask thread A to finish sync (which should be a no-op)
doCallLogSync(threadA, editLog);
} finally {
threadA.shutdown();
threadB.shutdown();
if(fileSys != null) fileSys.close();
if(cluster != null) cluster.shutdown();
}
}
public void testFailedOpen() throws Exception {
Configuration conf = new HdfsConfiguration();
MiniDFSCluster cluster = null;
cluster = new MiniDFSCluster.Builder(conf).numDataNodes(NUM_DATA_NODES).build();
cluster.waitActive();
FSNamesystem fsn = cluster.getNamesystem();
// Set up spys
FSImage originalImage = fsn.getFSImage();
originalImage.unlockAll();
FSImage spyImage = spy(originalImage);
FSEditLog editLog = originalImage.getEditLog();
FSEditLog spyLog = spy(editLog);
fsn.dir.fsImage = spyImage;
spyImage.setStorageDirectories(
FSNamesystem.getNamespaceDirs(conf),
FSNamesystem.getNamespaceEditsDirs(conf));
// Fail every attempt to open a new edit file
doThrow(new IOException("Injected fault: open")).
when(spyLog).addNewEditLogStream((File)anyObject());
try {
spyLog.close();
spyLog.open();
fail("open did not fail even when all directories failed!");
} catch(IOException ioe) {
LOG.info("Got expected exception", ioe);
} finally {
spyLog.close();
}
// Reset and try it with a working open
editLog.close();
originalImage.close();
fsn.close();
cluster = new MiniDFSCluster.Builder(conf).numDataNodes(NUM_DATA_NODES).build();
cluster.waitActive();
fsn = cluster.getNamesystem();
originalImage = fsn.getFSImage();
originalImage.unlockAll();
editLog = originalImage.getEditLog();
originalImage.setStorageDirectories(
FSNamesystem.getNamespaceDirs(conf),
FSNamesystem.getNamespaceEditsDirs(conf));
editLog.close();
editLog.open();
// Close everything off
editLog.close();
originalImage.close();
fsn.close();
}
}