/** Test that the NN re-learns of volume failures after restart. */
@Test
public void testVolFailureStatsPreservedOnNNRestart() throws Exception {
assumeTrue(!System.getProperty("os.name").startsWith("Windows"));
// Bring up two more datanodes that can tolerate 1 failure
cluster.startDataNodes(conf, 2, true, null, null);
cluster.waitActive();
final DatanodeManager dm = cluster.getNamesystem().getBlockManager().getDatanodeManager();
long origCapacity = DFSTestUtil.getLiveDatanodeCapacity(dm);
long dnCapacity = DFSTestUtil.getDatanodeCapacity(dm, 0);
// Fail the first volume on both datanodes (we have to keep the
// third healthy so one node in the pipeline will not fail).
File dn1Vol1 = new File(dataDir, "data" + (2 * 0 + 1));
File dn2Vol1 = new File(dataDir, "data" + (2 * 1 + 1));
assertTrue("Couldn't chmod local vol", FileUtil.setExecutable(dn1Vol1, false));
assertTrue("Couldn't chmod local vol", FileUtil.setExecutable(dn2Vol1, false));
Path file1 = new Path("/test1");
DFSTestUtil.createFile(fs, file1, 1024, (short) 2, 1L);
DFSTestUtil.waitReplication(fs, file1, (short) 2);
// The NN reports two volumes failures
DFSTestUtil.waitForDatanodeStatus(
dm, 3, 0, 2, origCapacity - (1 * dnCapacity), WAIT_FOR_HEARTBEATS);
// After restarting the NN it still see the two failures
cluster.restartNameNode(0);
cluster.waitActive();
DFSTestUtil.waitForDatanodeStatus(
dm, 3, 0, 2, origCapacity - (1 * dnCapacity), WAIT_FOR_HEARTBEATS);
}
/** test savenamespace in the middle of a checkpoint */
@Test
public void testCheckpointWithSavenamespace() throws Exception {
Configuration conf = getConf();
MiniDFSCluster cluster = new MiniDFSCluster(conf, 1, true, null);
cluster.waitActive();
FSNamesystem fsn = FSNamesystem.getFSNamesystem();
// Replace the FSImage with a spy
final FSImage originalImage = fsn.dir.fsImage;
try {
doAnEdit(fsn, 1);
CheckpointSignature sig = fsn.rollEditLog();
LOG.warn("Checkpoint signature: " + sig);
// Do another edit
doAnEdit(fsn, 2);
// Save namespace
fsn.saveNamespace(true, false);
// try to do a rollFSImage, this should fail because the
// saveNamespace have already occured after the call to
// rollFSEdit
try {
fsn.rollFSImage(sig);
assertTrue(
"The rollFSImage immediately folloing the saveName " + " command should fail. ", false);
} catch (IOException e) {
LOG.info(
"Expected exception while invoking rollFSImage "
+ " after a successful call to saveNamespace."
+ e);
}
// Now shut down and restart the NN
originalImage.close();
fsn.close();
cluster.shutdown();
fsn = null;
// Start a new namesystem, which should be able to recover
// the namespace from the previous incarnation.
cluster = new MiniDFSCluster(conf, 1, false, null);
cluster.waitActive();
fsn = FSNamesystem.getFSNamesystem();
// Make sure the image loaded including our edits.
checkEditExists(cluster, 1);
checkEditExists(cluster, 2);
} finally {
if (fsn != null) {
fsn.close();
cluster.shutdown();
}
}
}
private void testSaveWhileEditsRolled(boolean dosafemode, boolean force, boolean uncompressed)
throws Exception {
Configuration conf = getConf();
MiniDFSCluster cluster = new MiniDFSCluster(conf, 1, true, null);
cluster.waitActive();
FSNamesystem fsn = FSNamesystem.getFSNamesystem();
// Replace the FSImage with a spy
FSImage originalImage = fsn.dir.fsImage;
FSImage spyImage = spy(originalImage);
spyImage.setStorageDirectories(
FSNamesystem.getNamespaceDirs(conf), FSNamesystem.getNamespaceEditsDirs(conf));
fsn.dir.fsImage = spyImage;
try {
doAnEdit(fsn, 1);
CheckpointSignature sig = fsn.rollEditLog();
LOG.warn("Checkpoint signature: " + sig);
// Do another edit
doAnEdit(fsn, 2);
// Save namespace
if (dosafemode) {
fsn.setSafeMode(SafeModeAction.SAFEMODE_ENTER);
}
fsn.saveNamespace(force, uncompressed);
// Now shut down and restart the NN
originalImage.close();
originalImage = null;
fsn.close();
fsn = null;
cluster.shutdown();
// Start a new namesystem, which should be able to recover
// the namespace from the previous incarnation.
cluster = new MiniDFSCluster(conf, 1, false, null);
cluster.waitActive();
fsn = FSNamesystem.getFSNamesystem();
// Make sure the image loaded including our edits.
checkEditExists(cluster, 1);
checkEditExists(cluster, 2);
} finally {
if (originalImage != null) {
originalImage.close();
}
if (fsn != null) {
fsn.close();
cluster.shutdown();
}
}
}
/** Test that there are under replication blocks after vol failures */
@Test
public void testUnderReplicationAfterVolFailure() throws Exception {
// The test uses DataNodeTestUtils#injectDataDirFailure() to simulate
// volume failures which is currently not supported on Windows.
assumeTrue(!Path.WINDOWS);
// Bring up one more datanode
cluster.startDataNodes(conf, 1, true, null, null);
cluster.waitActive();
final BlockManager bm = cluster.getNamesystem().getBlockManager();
Path file1 = new Path("/test1");
DFSTestUtil.createFile(fs, file1, 1024, (short) 3, 1L);
DFSTestUtil.waitReplication(fs, file1, (short) 3);
// Fail the first volume on both datanodes
File dn1Vol1 = new File(dataDir, "data" + (2 * 0 + 1));
File dn2Vol1 = new File(dataDir, "data" + (2 * 1 + 1));
DataNodeTestUtils.injectDataDirFailure(dn1Vol1, dn2Vol1);
Path file2 = new Path("/test2");
DFSTestUtil.createFile(fs, file2, 1024, (short) 3, 1L);
DFSTestUtil.waitReplication(fs, file2, (short) 3);
// underReplicatedBlocks are due to failed volumes
int underReplicatedBlocks =
BlockManagerTestUtil.checkHeartbeatAndGetUnderReplicatedBlocksCount(
cluster.getNamesystem(), bm);
assertTrue(
"There is no under replicated block after volume failure", underReplicatedBlocks > 0);
}
@Before
public void setupCluster() throws Exception {
conf = new Configuration();
conf.setInt(DFSConfigKeys.DFS_BLOCK_SIZE_KEY, SMALL_BLOCK);
// Bump up replication interval so that we only run replication
// checks explicitly.
conf.setInt(DFSConfigKeys.DFS_NAMENODE_REPLICATION_INTERVAL_KEY, 600);
// Increase max streams so that we re-replicate quickly.
conf.setInt(DFSConfigKeys.DFS_NAMENODE_REPLICATION_MAX_STREAMS_KEY, 1000);
// See RandomDeleterPolicy javadoc.
conf.setClass(
DFSConfigKeys.DFS_BLOCK_REPLICATOR_CLASSNAME_KEY,
RandomDeleterPolicy.class,
BlockPlacementPolicy.class);
conf.setInt(DFSConfigKeys.DFS_HA_TAILEDITS_PERIOD_KEY, 1);
cluster =
new MiniDFSCluster.Builder(conf)
.nnTopology(MiniDFSNNTopology.simpleHATopology())
.numDataNodes(3)
.build();
nn1 = cluster.getNameNode(0);
nn2 = cluster.getNameNode(1);
cluster.waitActive();
cluster.transitionToActive(0);
// Trigger block reports so that the first NN trusts all
// of the DNs, and will issue deletions
cluster.triggerBlockReports();
fs = HATestUtil.configureFailoverFs(cluster, conf);
}
@Test(timeout = 120000)
public void testSeekAfterSetDropBehind() throws Exception {
// start a cluster
LOG.info("testSeekAfterSetDropBehind");
Configuration conf = new HdfsConfiguration();
MiniDFSCluster cluster = null;
String TEST_PATH = "/test";
int TEST_PATH_LEN = MAX_TEST_FILE_LEN;
try {
cluster = new MiniDFSCluster.Builder(conf).numDataNodes(1).build();
cluster.waitActive();
FileSystem fs = cluster.getFileSystem();
createHdfsFile(fs, new Path(TEST_PATH), TEST_PATH_LEN, false);
// verify that we can seek after setDropBehind
FSDataInputStream fis = fs.open(new Path(TEST_PATH));
try {
Assert.assertTrue(fis.read() != -1); // create BlockReader
fis.setDropBehind(false); // clear BlockReader
fis.seek(2); // seek
} finally {
fis.close();
}
} finally {
if (cluster != null) {
cluster.shutdown();
}
}
}
/**
* Initialize the cluster, wait for it to become active, and get FileSystem instances for our test
* users.
*
* @param format if true, format the NameNode and DataNodes before starting up
* @throws Exception if any step fails
*/
private static void initCluster(boolean format) throws Exception {
cluster = new MiniDFSCluster.Builder(conf).numDataNodes(1).format(format).build();
cluster.waitActive();
hdfs = cluster.getFileSystem();
fsAsBruce = DFSTestUtil.getFileSystemAs(BRUCE, conf);
fsAsDiana = DFSTestUtil.getFileSystemAs(DIANA, conf);
}
/**
* test 1. create DFS cluster with 3 storage directories - 2 EDITS_IMAGE, 1 EDITS 2. create a
* cluster and write a file 3. corrupt/disable one storage (or two) by removing 4. run
* doCheckpoint - it will fail on removed dirs (which will invalidate the storages) 5. write
* another file 6. check that edits and fsimage differ 7. run doCheckpoint 8. verify that all the
* image and edits files are the same.
*/
@Test
public void testStorageRestore() throws Exception {
int numDatanodes = 2;
cluster =
new MiniDFSCluster(0, config, numDatanodes, true, false, true, null, null, null, null);
cluster.waitActive();
SecondaryNameNode secondary = new SecondaryNameNode(config);
FileSystem fs = cluster.getFileSystem();
Path path = new Path("/", "test");
writeFile(fs, path, 2);
invalidateStorage(cluster.getNameNode().getFSImage());
path = new Path("/", "test1");
writeFile(fs, path, 2);
checkFiles(false);
secondary.doCheckpoint();
checkFiles(true);
secondary.shutdown();
cluster.shutdown();
}
protected void setupCluster(boolean simulated, long minFileSize, String[] racks, String[] hosts)
throws IOException {
conf = new Configuration();
localFileSys = FileSystem.getLocal(conf);
conf.setLong("dfs.blockreport.intervalMsec", 1000L);
conf.set("dfs.replication.pending.timeout.sec", "2");
conf.setLong("dfs.block.size", 1L);
conf.set(
"dfs.block.replicator.classname",
"org.apache.hadoop.hdfs.server.namenode.BlockPlacementPolicyRaid");
conf.setLong("hdfs.raid.min.filesize", minFileSize);
Utils.loadTestCodecs(conf, 5, 5, 1, 3, "/raid", "/raidrs", false, true);
conf.setInt("io.bytes.per.checksum", 1);
excludeFile = new Path(TEST_DIR, "exclude" + System.currentTimeMillis());
cleanFile(excludeFile);
conf.set("dfs.hosts.exclude", excludeFile.toUri().getPath());
writeConfigFile(excludeFile, null);
if (!simulated) {
cluster = new MiniDFSCluster(conf, hosts.length, true, racks, hosts);
} else {
long[] capacities = new long[] {CAPACITY, CAPACITY, CAPACITY};
cluster = new MiniDFSCluster(0, conf, hosts.length, true, true, null, racks, capacities);
}
cluster.waitActive();
namesystem = cluster.getNameNode().getNamesystem();
Assert.assertTrue(
"BlockPlacementPolicy type is not correct.",
namesystem.replicator instanceof BlockPlacementPolicyRaid);
policy = (BlockPlacementPolicyRaid) namesystem.replicator;
fs = cluster.getFileSystem();
dfs = (DistributedFileSystem) fs;
TestDirectoryRaidDfs.setupStripeStore(conf, fs);
}
/* This test start a one-node cluster, fill the node to be 30% full;
* It then adds an empty node and start balancing.
* @param newCapacity new node's capacity
* @param new
*/
private void test(long[] capacities, String[] racks, long newCapacity, String newRack)
throws Exception {
int numOfDatanodes = capacities.length;
assertEquals(numOfDatanodes, racks.length);
cluster = new MiniDFSCluster(0, CONF, capacities.length, true, true, null, racks, capacities);
try {
cluster.waitActive();
client = DFSClient.createNamenode(CONF);
long totalCapacity = 0L;
for (long capacity : capacities) {
totalCapacity += capacity;
}
// fill up the cluster to be 30% full
long totalUsedSpace = totalCapacity * 3 / 10;
createFile(totalUsedSpace / numOfDatanodes, (short) numOfDatanodes);
// start up an empty node with the same capacity and on the same rack
cluster.startDataNodes(CONF, 1, true, null, new String[] {newRack}, new long[] {newCapacity});
totalCapacity += newCapacity;
// run balancer and validate results
runBalancer(totalUsedSpace, totalCapacity);
} finally {
cluster.shutdown();
}
}
/* fill up a cluster with <code>numNodes</code> datanodes
* whose used space to be <code>size</code>
*/
private Block[] generateBlocks(long size, short numNodes) throws IOException {
cluster = new MiniDFSCluster(CONF, numNodes, true, null);
try {
cluster.waitActive();
client = DFSClient.createNamenode(CONF);
short replicationFactor = (short) (numNodes - 1);
long fileLen = size / replicationFactor;
createFile(fileLen, replicationFactor);
List<LocatedBlock> locatedBlocks =
client.getBlockLocations(fileName, 0, fileLen).getLocatedBlocks();
int numOfBlocks = locatedBlocks.size();
Block[] blocks = new Block[numOfBlocks];
for (int i = 0; i < numOfBlocks; i++) {
Block b = locatedBlocks.get(i).getBlock();
blocks[i] = new Block(b.getBlockId(), b.getNumBytes(), b.getGenerationStamp());
}
return blocks;
} finally {
cluster.shutdown();
}
}
/* we first start a cluster and fill the cluster up to a certain size.
* then redistribute blocks according the required distribution.
* Afterwards a balancer is running to balance the cluster.
*/
private void testUnevenDistribution(long distribution[], long capacities[], String[] racks)
throws Exception {
int numDatanodes = distribution.length;
if (capacities.length != numDatanodes || racks.length != numDatanodes) {
throw new IllegalArgumentException("Array length is not the same");
}
// calculate total space that need to be filled
long totalUsedSpace = 0L;
for (int i = 0; i < distribution.length; i++) {
totalUsedSpace += distribution[i];
}
// fill the cluster
Block[] blocks = generateBlocks(totalUsedSpace, (short) numDatanodes);
// redistribute blocks
Block[][] blocksDN = distributeBlocks(blocks, (short) (numDatanodes - 1), distribution);
// restart the cluster: do NOT format the cluster
CONF.set("dfs.safemode.threshold.pct", "0.0f");
cluster = new MiniDFSCluster(0, CONF, numDatanodes, false, true, null, racks, capacities);
cluster.waitActive();
client = DFSClient.createNamenode(CONF);
cluster.injectBlocks(blocksDN);
long totalCapacity = 0L;
for (long capacity : capacities) {
totalCapacity += capacity;
}
runBalancer(totalUsedSpace, totalCapacity);
}
/** Test DFS Raid */
public void testBlockMissingException() throws Exception {
LOG.info("Test testBlockMissingException started.");
long blockSize = 1024L;
int numBlocks = 4;
conf = new Configuration();
try {
dfs = new MiniDFSCluster(conf, NUM_DATANODES, true, null);
dfs.waitActive();
fileSys = (DistributedFileSystem) dfs.getFileSystem();
Path file1 = new Path("/user/dhruba/raidtest/file1");
createOldFile(fileSys, file1, 1, numBlocks, blockSize);
// extract block locations from File system. Wait till file is closed.
LocatedBlocks locations = null;
locations =
fileSys.dfs.namenode.getBlockLocations(file1.toString(), 0, numBlocks * blockSize);
// remove block of file
LOG.info("Remove first block of file");
corruptBlock(file1, locations.get(0).getBlock());
// validate that the system throws BlockMissingException
validateFile(fileSys, file1);
} finally {
if (fileSys != null) fileSys.close();
if (dfs != null) dfs.shutdown();
}
LOG.info("Test testBlockMissingException completed.");
}
@Before
public void setUp() throws Exception {
cluster = new MiniDFSCluster.Builder(CONF).build();
cluster.waitActive();
cluster.getNameNode();
fs = (DistributedFileSystem) cluster.getFileSystem();
}
@Override
protected void setUp() throws Exception {
cluster = new MiniDFSCluster(CONF, 1, true, null);
cluster.waitActive();
fs = (DistributedFileSystem) cluster.getFileSystem();
metrics = fs.getClient().getDFSClientMetrics();
}
@Before
public void setUp() throws Exception {
conf = new HdfsConfiguration();
conf.setLong(DFSConfigKeys.DFS_BLOCK_SIZE_KEY, 512L);
cluster = new MiniDFSCluster.Builder(conf).numDataNodes(1).build();
cluster.waitActive();
fs = cluster.getFileSystem();
}
@Override
protected void setUp() throws Exception {
cluster = new MiniDFSCluster(CONF, 1, true, null);
cluster.waitActive();
cluster.getNameNode();
nnMetrics = NameNode.getNameNodeMetrics();
fs = (DistributedFileSystem) cluster.getFileSystem();
}
@Before
public void setUp() throws Exception {
conf = new Configuration();
conf.setLong(DFSConfigKeys.DFS_BLOCK_SIZE_KEY, BLOCKSIZE);
cluster = new MiniDFSCluster.Builder(conf).numDataNodes(REPLICATION).build();
cluster.waitActive();
fsdir = cluster.getNamesystem().getFSDirectory();
dfs = cluster.getFileSystem();
}
private static MiniDFSCluster startMini(String testName) throws IOException {
File baseDir = new File("./target/hdfs/" + testName).getAbsoluteFile();
FileUtil.fullyDelete(baseDir);
Configuration conf = new Configuration();
conf.set(MiniDFSCluster.HDFS_MINIDFS_BASEDIR, baseDir.getAbsolutePath());
MiniDFSCluster.Builder builder = new MiniDFSCluster.Builder(conf);
MiniDFSCluster hdfsCluster = builder.clusterId(testName).build();
hdfsCluster.waitActive();
return hdfsCluster;
}
public void testRPCInterrupted() throws IOException, InterruptedException {
final MiniDFSCluster cluster;
Configuration conf = new Configuration();
cluster = new MiniDFSCluster(conf, 3, true, null);
final AtomicBoolean passed = new AtomicBoolean(false);
try {
cluster.waitActive();
Thread rpcThread =
new Thread(
new Runnable() {
@Override
public void run() {
FileSystem fs = null;
try {
fs = cluster.getUniqueFileSystem();
int i = 0;
while (true) {
fs.create(new Path(String.format("/file-%09d", i++)));
fs.listStatus(new Path("/"));
}
} catch (IOException e) {
if (e.getCause() instanceof InterruptedException) {
// the underlying InterruptedException should be wrapped to an
// IOException and end up here
passed.set(true);
} else {
passed.set(false);
fail(e.getMessage());
}
} finally {
if (fs != null) {
try {
fs.close();
} catch (IOException e) {
passed.set(false);
LOG.error(e);
fail(e.toString());
}
}
}
}
});
FileSystem fs2 = cluster.getUniqueFileSystem();
rpcThread.start();
Thread.sleep(1000);
rpcThread.interrupt();
rpcThread.join();
FileStatus[] statuses = fs2.listStatus(new Path("/"));
assertTrue("expect at least 1 file created", statuses.length > 0);
assertTrue("error in writing thread, see above", passed.get());
} finally {
cluster.shutdown();
}
}
/**
* testing that APPEND operation can handle token expiration when re-establishing pipeline is
* needed
*/
@Test
public void testAppend() throws Exception {
MiniDFSCluster cluster = null;
int numDataNodes = 2;
Configuration conf = getConf(numDataNodes);
try {
cluster = new MiniDFSCluster.Builder(conf).numDataNodes(numDataNodes).build();
cluster.waitActive();
assertEquals(numDataNodes, cluster.getDataNodes().size());
final NameNode nn = cluster.getNameNode();
final BlockManager bm = nn.getNamesystem().getBlockManager();
final BlockTokenSecretManager sm = bm.getBlockTokenSecretManager();
// set a short token lifetime (1 second)
SecurityTestUtil.setBlockTokenLifetime(sm, 1000L);
Path fileToAppend = new Path(FILE_TO_APPEND);
FileSystem fs = cluster.getFileSystem();
byte[] expected = generateBytes(FILE_SIZE);
// write a one-byte file
FSDataOutputStream stm = writeFile(fs, fileToAppend, (short) numDataNodes, BLOCK_SIZE);
stm.write(expected, 0, 1);
stm.close();
// open the file again for append
stm = fs.append(fileToAppend);
int mid = expected.length - 1;
stm.write(expected, 1, mid - 1);
stm.hflush();
/*
* wait till token used in stm expires
*/
Token<BlockTokenIdentifier> token = DFSTestUtil.getBlockToken(stm);
while (!SecurityTestUtil.isBlockTokenExpired(token)) {
try {
Thread.sleep(10);
} catch (InterruptedException ignored) {
}
}
// remove a datanode to force re-establishing pipeline
cluster.stopDataNode(0);
// append the rest of the file
stm.write(expected, mid, expected.length - mid);
stm.close();
// check if append is successful
FSDataInputStream in5 = fs.open(fileToAppend);
assertTrue(checkFile1(in5, expected));
} finally {
if (cluster != null) {
cluster.shutdown();
}
}
}
@BeforeClass
public static void setUp() throws Exception {
Configuration conf = new Configuration();
cluster = new MiniDFSCluster.Builder(conf).numDataNodes(REPLICATION).build();
cluster.waitActive();
FSNamesystem fsn = cluster.getNamesystem();
fsdir = fsn.getFSDirectory();
hdfs = cluster.getFileSystem();
}
/**
* The following test first creates a file. It verifies the block information from a datanode.
* Then, it updates the block with new information and verifies again.
*/
@Test
public void testBlockMetaDataInfo() throws Exception {
MiniDFSCluster cluster = null;
try {
cluster = new MiniDFSCluster.Builder(conf).numDataNodes(3).build();
cluster.waitActive();
// create a file
DistributedFileSystem dfs = (DistributedFileSystem) cluster.getFileSystem();
String filestr = "/foo";
Path filepath = new Path(filestr);
DFSTestUtil.createFile(dfs, filepath, 1024L, (short) 3, 0L);
assertTrue(dfs.exists(filepath));
// get block info
LocatedBlock locatedblock =
getLastLocatedBlock(DFSClientAdapter.getDFSClient(dfs).getNamenode(), filestr);
DatanodeInfo[] datanodeinfo = locatedblock.getLocations();
assertTrue(datanodeinfo.length > 0);
// connect to a data node
DataNode datanode = cluster.getDataNode(datanodeinfo[0].getIpcPort());
InterDatanodeProtocol idp =
DataNodeTestUtils.createInterDatanodeProtocolProxy(datanode, datanodeinfo[0], conf);
// stop block scanner, so we could compare lastScanTime
DataNodeTestUtils.shutdownBlockScanner(datanode);
// verify BlockMetaDataInfo
ExtendedBlock b = locatedblock.getBlock();
InterDatanodeProtocol.LOG.info("b=" + b + ", " + b.getClass());
checkMetaInfo(b, datanode);
long recoveryId = b.getGenerationStamp() + 1;
idp.initReplicaRecovery(new RecoveringBlock(b, locatedblock.getLocations(), recoveryId));
// verify updateBlock
ExtendedBlock newblock =
new ExtendedBlock(
b.getBlockPoolId(), b.getBlockId(), b.getNumBytes() / 2, b.getGenerationStamp() + 1);
idp.updateReplicaUnderRecovery(b, recoveryId, newblock.getNumBytes());
checkMetaInfo(newblock, datanode);
// Verify correct null response trying to init recovery for a missing block
ExtendedBlock badBlock = new ExtendedBlock("fake-pool", b.getBlockId(), 0, 0);
assertNull(
idp.initReplicaRecovery(
new RecoveringBlock(badBlock, locatedblock.getLocations(), recoveryId)));
} finally {
if (cluster != null) {
cluster.shutdown();
}
}
}
private void setupCluster() throws IOException {
setupConf();
// start the cluster with one datanode
cluster = new MiniDFSCluster(conf, 6, true, racks, hosts);
cluster.waitActive();
fs = cluster.getFileSystem();
placementMonitor = new PlacementMonitor(conf);
placementMonitor.start();
blockMover = placementMonitor.blockMover;
namenode = cluster.getNameNode();
datanodes = namenode.getDatanodeReport(DatanodeReportType.LIVE);
}
@Before
public void setUp() throws Exception {
// bring up a cluster of 2
conf = new HdfsConfiguration();
conf.setLong(DFSConfigKeys.DFS_BLOCK_SIZE_KEY, block_size);
// Allow a single volume failure (there are two volumes)
conf.setInt(DFSConfigKeys.DFS_DATANODE_FAILED_VOLUMES_TOLERATED_KEY, 1);
cluster = new MiniDFSCluster.Builder(conf).numDataNodes(dn_num).build();
cluster.waitActive();
fs = cluster.getFileSystem();
dataDir = new File(cluster.getDataDirectory());
}
/**
* Test whether we can delay the deletion of unknown blocks in DataNode's first several block
* reports.
*/
@Test
public void testPendingDeleteUnknownBlocks() throws Exception {
final int fileNum = 5; // 5 files
final Path[] files = new Path[fileNum];
final DataNodeProperties[] dnprops = new DataNodeProperties[REPLICATION];
// create a group of files, each file contains 1 block
for (int i = 0; i < fileNum; i++) {
files[i] = new Path("/file" + i);
DFSTestUtil.createFile(dfs, files[i], BLOCKSIZE, REPLICATION, i);
}
// wait until all DataNodes have replicas
waitForReplication();
for (int i = REPLICATION - 1; i >= 0; i--) {
dnprops[i] = cluster.stopDataNode(i);
}
Thread.sleep(2000);
// delete 2 files, we still have 3 files remaining so that we can cover
// every DN storage
for (int i = 0; i < 2; i++) {
dfs.delete(files[i], true);
}
// restart NameNode
cluster.restartNameNode(false);
InvalidateBlocks invalidateBlocks =
(InvalidateBlocks)
Whitebox.getInternalState(
cluster.getNamesystem().getBlockManager(), "invalidateBlocks");
InvalidateBlocks mockIb = Mockito.spy(invalidateBlocks);
Mockito.doReturn(1L).when(mockIb).getInvalidationDelay();
Whitebox.setInternalState(
cluster.getNamesystem().getBlockManager(), "invalidateBlocks", mockIb);
Assert.assertEquals(0L, cluster.getNamesystem().getPendingDeletionBlocks());
// restart DataNodes
for (int i = 0; i < REPLICATION; i++) {
cluster.restartDataNode(dnprops[i], true);
}
cluster.waitActive();
for (int i = 0; i < REPLICATION; i++) {
DataNodeTestUtils.triggerBlockReport(cluster.getDataNodes().get(i));
}
Thread.sleep(2000);
// make sure we have received block reports by checking the total block #
Assert.assertEquals(3, cluster.getNamesystem().getBlocksTotal());
Assert.assertEquals(4, cluster.getNamesystem().getPendingDeletionBlocks());
cluster.restartNameNode(true);
Thread.sleep(6000);
Assert.assertEquals(3, cluster.getNamesystem().getBlocksTotal());
Assert.assertEquals(0, cluster.getNamesystem().getPendingDeletionBlocks());
}
@Before
public void setUp() throws IOException {
conf = new HdfsConfiguration();
conf.setInt(DFSConfigKeys.DFS_BLOCK_SIZE_KEY, 100);
conf.setInt(DFSConfigKeys.DFS_BYTES_PER_CHECKSUM_KEY, 100);
cluster =
new MiniDFSCluster.Builder(conf)
.nnTopology(MiniDFSNNTopology.simpleFederatedTopology(3))
.build();
for (int i = 0; i < 3; i++) {
cluster.waitActive(i);
}
}
@Before
public void setUp() throws Exception {
config = new Configuration();
config.setClass(
"hadoop.security.group.mapping",
TestRefreshUserMappings.MockUnixGroupsMapping.class,
GroupMappingServiceProvider.class);
config.setLong("hadoop.security.groups.cache.secs", groupRefreshTimeoutSec);
Groups.getUserToGroupsMappingService(config);
FileSystem.setDefaultUri(config, "hdfs://localhost:" + "0");
cluster = new MiniDFSCluster(0, config, 1, true, true, true, null, null, null, null);
cluster.waitActive();
}
private void mySetup(int stripeLength) throws Exception {
if (System.getProperty("hadoop.log.dir") == null) {
String base = new File(".").getAbsolutePath();
System.setProperty("hadoop.log.dir", new Path(base).toString() + "/logs");
}
new File(TEST_DIR).mkdirs(); // Make sure data directory exists
conf = new Configuration();
conf.set("raid.config.file", CONFIG_FILE);
conf.setBoolean("raid.config.reload", true);
conf.setLong("raid.config.reload.interval", RELOAD_INTERVAL);
// scan all policies once every 5 second
conf.setLong("raid.policy.rescan.interval", 5000);
// do not use map-reduce cluster for Raiding
conf.set("raid.classname", "org.apache.hadoop.raid.LocalRaidNode");
conf.set("raid.server.address", "localhost:" + MiniDFSCluster.getFreePort());
conf.set("mapred.raid.http.address", "localhost:0");
Utils.loadTestCodecs(
conf, stripeLength, stripeLength, 1, 3, "/destraid", "/destraidrs", false, true);
conf.setBoolean("dfs.permissions", false);
// Make sure initial repl is smaller than NUM_DATANODES
conf.setInt(RaidNode.RAID_PARITY_INITIAL_REPL_KEY, 1);
dfsCluster = new MiniDFSCluster(conf, NUM_DATANODES, true, null);
dfsCluster.waitActive();
fileSys = dfsCluster.getFileSystem();
namenode = fileSys.getUri().toString();
FileSystem.setDefaultUri(conf, namenode);
mr = new MiniMRCluster(4, namenode, 3);
jobTrackerName = "localhost:" + mr.getJobTrackerPort();
hftp = "hftp://localhost.localdomain:" + dfsCluster.getNameNodePort();
FileSystem.setDefaultUri(conf, namenode);
conf.set("mapred.job.tracker", jobTrackerName);
conf.set(RaidNode.RAID_CHECKSUM_STORE_CLASS_KEY, "org.apache.hadoop.raid.LocalChecksumStore");
conf.setBoolean(RaidNode.RAID_CHECKSUM_STORE_REQUIRED_KEY, true);
conf.set(LocalChecksumStore.LOCAL_CHECK_STORE_DIR_KEY, CHECKSUM_STORE_DIR);
conf.set(RaidNode.RAID_STRIPE_STORE_CLASS_KEY, "org.apache.hadoop.raid.LocalStripeStore");
conf.set(LocalStripeStore.LOCAL_STRIPE_STORE_DIR_KEY, STRIPE_STORE_DIR);
ConfigBuilder cb = new ConfigBuilder(CONFIG_FILE);
cb.addPolicy("RaidTest1", "/user/dhruba/raidtest", 1, 1);
cb.addPolicy("RaidTest2", "/user/dhruba/raidtestrs", 1, 1, "rs");
cb.persist();
}
@Test(timeout = 120000)
public void testFadviseSkippedForSmallReads() throws Exception {
// start a cluster
LOG.info("testFadviseSkippedForSmallReads");
tracker.clear();
Configuration conf = new HdfsConfiguration();
conf.setBoolean(DFSConfigKeys.DFS_DATANODE_DROP_CACHE_BEHIND_READS_KEY, true);
conf.setBoolean(DFSConfigKeys.DFS_DATANODE_DROP_CACHE_BEHIND_WRITES_KEY, true);
MiniDFSCluster cluster = null;
String TEST_PATH = "/test";
int TEST_PATH_LEN = MAX_TEST_FILE_LEN;
FSDataInputStream fis = null;
try {
cluster = new MiniDFSCluster.Builder(conf).numDataNodes(1).build();
cluster.waitActive();
FileSystem fs = cluster.getFileSystem();
// create new file
createHdfsFile(fs, new Path(TEST_PATH), TEST_PATH_LEN, null);
// Since the DataNode was configured with drop-behind, and we didn't
// specify any policy, we should have done drop-behind.
ExtendedBlock block =
cluster
.getNameNode()
.getRpcServer()
.getBlockLocations(TEST_PATH, 0, Long.MAX_VALUE)
.get(0)
.getBlock();
String fadvisedFileName = cluster.getBlockFile(0, block).getName();
Stats stats = tracker.getStats(fadvisedFileName);
stats.assertDroppedInRange(0, TEST_PATH_LEN - WRITE_PACKET_SIZE);
stats.clear();
stats.assertNotDroppedInRange(0, TEST_PATH_LEN);
// read file
fis = fs.open(new Path(TEST_PATH));
byte buf[] = new byte[17];
fis.readFully(4096, buf, 0, buf.length);
// we should not have dropped anything because of the small read.
stats = tracker.getStats(fadvisedFileName);
stats.assertNotDroppedInRange(0, TEST_PATH_LEN - WRITE_PACKET_SIZE);
} finally {
IOUtils.cleanup(null, fis);
if (cluster != null) {
cluster.shutdown();
}
}
}
