/**
* Make sure at least one non-transient volume has a saved copy of the replica. An infinite loop
* is used to ensure the async lazy persist tasks are completely done before verification. Caller
* of ensureLazyPersistBlocksAreSaved expects either a successful pass or timeout failure.
*/
protected final void ensureLazyPersistBlocksAreSaved(LocatedBlocks locatedBlocks)
throws IOException, InterruptedException {
final String bpid = cluster.getNamesystem().getBlockPoolId();
List<? extends FsVolumeSpi> volumes = cluster.getDataNodes().get(0).getFSDataset().getVolumes();
final Set<Long> persistedBlockIds = new HashSet<Long>();
while (persistedBlockIds.size() < locatedBlocks.getLocatedBlocks().size()) {
// Take 1 second sleep before each verification iteration
Thread.sleep(1000);
for (LocatedBlock lb : locatedBlocks.getLocatedBlocks()) {
for (FsVolumeSpi v : volumes) {
if (v.isTransientStorage()) {
continue;
}
FsVolumeImpl volume = (FsVolumeImpl) v;
File lazyPersistDir = volume.getBlockPoolSlice(bpid).getLazypersistDir();
long blockId = lb.getBlock().getBlockId();
File targetDir = DatanodeUtil.idToBlockDir(lazyPersistDir, blockId);
File blockFile = new File(targetDir, lb.getBlock().getBlockName());
if (blockFile.exists()) {
// Found a persisted copy for this block and added to the Set
persistedBlockIds.add(blockId);
}
}
}
}
// We should have found a persisted copy for each located block.
assertThat(persistedBlockIds.size(), is(locatedBlocks.getLocatedBlocks().size()));
}
/**
* Regression test for HDFS-2742. The issue in this bug was: - DN does a block report while file
* is open. This BR contains the block in RBW state. - Standby queues the RBW state in
* PendingDatanodeMessages - Standby processes edit logs during failover. Before fixing this bug,
* it was mistakenly applying the RBW reported state after the block had been completed, causing
* the block to get marked corrupt. Instead, we should now be applying the RBW message on OP_ADD,
* and then the FINALIZED message on OP_CLOSE.
*/
@Test
public void testBlockReportsWhileFileBeingWritten() throws Exception {
FSDataOutputStream out = fs.create(TEST_FILE_PATH);
try {
AppendTestUtil.write(out, 0, 10);
out.hflush();
// Block report will include the RBW replica, but will be
// queued on the StandbyNode.
cluster.triggerBlockReports();
} finally {
IOUtils.closeStream(out);
}
cluster.transitionToStandby(0);
cluster.transitionToActive(1);
// Verify that no replicas are marked corrupt, and that the
// file is readable from the failed-over standby.
BlockManagerTestUtil.updateState(nn1.getNamesystem().getBlockManager());
BlockManagerTestUtil.updateState(nn2.getNamesystem().getBlockManager());
assertEquals(0, nn1.getNamesystem().getCorruptReplicaBlocks());
assertEquals(0, nn2.getNamesystem().getCorruptReplicaBlocks());
DFSTestUtil.readFile(fs, TEST_FILE_PATH);
}
@BeforeClass
public static void setUp() throws Exception {
Configuration conf = new Configuration();
conf.set("hadoop.security.auth_to_local", "RULE:[2:$1]");
dfsCluster = new MiniDFSCluster(conf, numSlaves, true, null);
jConf = new JobConf(conf);
mrCluster =
new MiniMRCluster(
0,
0,
numSlaves,
dfsCluster.getFileSystem().getUri().toString(),
1,
null,
null,
null,
jConf);
createTokenFileJson();
verifySecretKeysInJSONFile();
NameNodeAdapter.getDtSecretManager(dfsCluster.getNamesystem()).startThreads();
FileSystem fs = dfsCluster.getFileSystem();
p1 = new Path("file1");
p2 = new Path("file2");
p1 = fs.makeQualified(p1);
}
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);
}
/** 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);
}
@Before
public void setUp() throws Exception {
cluster = new MiniDFSCluster.Builder(CONF).build();
cluster.waitActive();
cluster.getNameNode();
fs = (DistributedFileSystem) cluster.getFileSystem();
}
@BeforeClass
public static void setUpBeforeClass() throws Exception {
File minidfsDir = new File("target/minidfs-" + UUID.randomUUID()).getAbsoluteFile();
minidfsDir.mkdirs();
Assert.assertTrue(minidfsDir.exists());
System.setProperty(MiniDFSCluster.PROP_TEST_BUILD_DATA, minidfsDir.getPath());
Configuration conf = new HdfsConfiguration();
conf.set("dfs.namenode.fs-limits.min-block-size", String.valueOf(32));
EditLogFileOutputStream.setShouldSkipFsyncForTesting(true);
miniDFS = new MiniDFSCluster.Builder(conf).numDataNodes(3).build();
dir = new Path(miniDFS.getURI() + "/dir");
FileSystem fs = miniDFS.getFileSystem();
fs.mkdirs(dir);
writeFile(fs, new Path(dir + "/forAllTests/" + "path"), 1000);
dummyEtc = new File(minidfsDir, "dummy-etc");
dummyEtc.mkdirs();
Assert.assertTrue(dummyEtc.exists());
Configuration dummyConf = new Configuration(false);
for (String file : new String[] {"core", "hdfs", "mapred", "yarn"}) {
File siteXml = new File(dummyEtc, file + "-site.xml");
FileOutputStream out = new FileOutputStream(siteXml);
dummyConf.writeXml(out);
out.close();
}
resourcesDir = minidfsDir.getAbsolutePath();
hadoopConfDir = dummyEtc.getName();
System.setProperty("sdc.resources.dir", resourcesDir);
;
}
/** copy empty directory on dfs file system */
public void testEmptyDir() throws Exception {
String namenode = null;
MiniDFSCluster cluster = null;
try {
Configuration conf = new Configuration();
cluster = new MiniDFSCluster(conf, 2, true, null);
final FileSystem hdfs = cluster.getFileSystem();
namenode = FileSystem.getDefaultUri(conf).toString();
if (namenode.startsWith("hdfs://")) {
FileSystem fs = FileSystem.get(URI.create(namenode), new Configuration());
fs.mkdirs(new Path("/empty"));
ToolRunner.run(
new DistCpV1(conf),
new String[] {"-log", namenode + "/logs", namenode + "/empty", namenode + "/dest"});
fs = FileSystem.get(URI.create(namenode + "/destdat"), conf);
assertTrue(
"Destination directory does not exist.", fs.exists(new Path(namenode + "/dest")));
deldir(hdfs, "/dest");
deldir(hdfs, "/empty");
deldir(hdfs, "/logs");
}
} finally {
if (cluster != null) {
cluster.shutdown();
}
}
}
/** copy files from dfs file system to local file system */
public void testCopyFromDfsToLocal() throws Exception {
MiniDFSCluster cluster = null;
try {
Configuration conf = new Configuration();
final FileSystem localfs = FileSystem.get(LOCAL_FS, conf);
cluster = new MiniDFSCluster(conf, 1, true, null);
final FileSystem hdfs = cluster.getFileSystem();
final String namenode = FileSystem.getDefaultUri(conf).toString();
if (namenode.startsWith("hdfs://")) {
MyFile[] files = createFiles(URI.create(namenode), "/srcdat");
ToolRunner.run(
new DistCpV1(conf),
new String[] {
"-log", "/logs", namenode + "/srcdat", "file:///" + TEST_ROOT_DIR + "/destdat"
});
assertTrue(
"Source and destination directories do not match.",
checkFiles(localfs, TEST_ROOT_DIR + "/destdat", files));
assertTrue("Log directory does not exist.", hdfs.exists(new Path("/logs")));
deldir(localfs, TEST_ROOT_DIR + "/destdat");
deldir(hdfs, "/logs");
deldir(hdfs, "/srcdat");
}
} finally {
if (cluster != null) {
cluster.shutdown();
}
}
}
/** Test truncate over quota does not mark file as UC or create a lease */
@Test(timeout = 60000)
public void testTruncateOverQuota() throws Exception {
final Path dir = new Path("/TestTruncateOverquota");
final Path file = new Path(dir, "file");
// create partial block file
dfs.mkdirs(dir);
DFSTestUtil.createFile(dfs, file, BLOCKSIZE / 2, REPLICATION, seed);
// lower quota to cause exception when appending to partial block
dfs.setQuota(dir, Long.MAX_VALUE - 1, 1);
final INodeDirectory dirNode = fsdir.getINode4Write(dir.toString()).asDirectory();
final long spaceUsed =
dirNode.getDirectoryWithQuotaFeature().getSpaceConsumed().getStorageSpace();
try {
dfs.truncate(file, BLOCKSIZE / 2 - 1);
Assert.fail("truncate didn't fail");
} catch (RemoteException e) {
assertTrue(e.getClassName().contains("DSQuotaExceededException"));
}
// check that the file exists, isn't UC, and has no dangling lease
LeaseManager lm = cluster.getNamesystem().getLeaseManager();
INodeFile inode = fsdir.getINode(file.toString()).asFile();
Assert.assertNotNull(inode);
Assert.assertFalse("should not be UC", inode.isUnderConstruction());
Assert.assertNull("should not have a lease", lm.getLease(inode));
// make sure the quota usage is unchanged
final long newSpaceUsed =
dirNode.getDirectoryWithQuotaFeature().getSpaceConsumed().getStorageSpace();
assertEquals(spaceUsed, newSpaceUsed);
// make sure edits aren't corrupted
dfs.recoverLease(file);
cluster.restartNameNodes();
}
private void corruptBlock(
MiniDFSCluster cluster, FileSystem fs, final Path fileName, int dnIndex, Block block)
throws IOException {
// corrupt the block on datanode dnIndex
// the indexes change once the nodes are restarted.
// But the datadirectory will not change
assertTrue(cluster.corruptReplica(block.getBlockName(), dnIndex));
DataNodeProperties dnProps = cluster.stopDataNode(0);
// Each datanode has multiple data dirs, check each
for (int dn = dnIndex * 2; dn < dnIndex * 2 + 2; dn++) {
File dataDir = new File(MiniDFSCluster.getBaseDirectory() + "data");
File scanLogFile =
new File(
dataDir,
"data"
+ (dn + 1)
+ MiniDFSCluster.FINALIZED_DIR_NAME
+ "dncp_block_verification.log.curr");
if (scanLogFile.exists()) {
// wait for one minute for deletion to succeed;
for (int i = 0; !scanLogFile.delete(); i++) {
assertTrue("Could not delete log file in one minute", i < 60);
try {
Thread.sleep(1000);
} catch (InterruptedException ignored) {
}
}
}
}
// restart the detained so the corrupt replica will be detected
cluster.restartDataNode(dnProps);
}
@AfterClass
public static void afterClass() throws Exception {
if (cluster == null) return;
FileSystem fs = cluster.getFileSystem();
bench.cleanup(fs);
cluster.shutdown();
}
@After
public void shutDownCluster() throws Exception {
// Dump all RamDisk JMX metrics before shutdown the cluster
printRamDiskJMXMetrics();
if (fs != null) {
fs.close();
fs = null;
client = null;
}
if (cluster != null) {
cluster.shutdownDataNodes();
cluster.shutdown();
cluster = null;
}
if (jmx != null) {
jmx = null;
}
IOUtils.closeQuietly(sockDir);
sockDir = null;
}
protected final boolean verifyDeletedBlocks(LocatedBlocks locatedBlocks)
throws IOException, InterruptedException {
LOG.info("Verifying replica has no saved copy after deletion.");
triggerBlockReport();
while (DataNodeTestUtils.getPendingAsyncDeletions(cluster.getDataNodes().get(0)) > 0L) {
Thread.sleep(1000);
}
final String bpid = cluster.getNamesystem().getBlockPoolId();
List<? extends FsVolumeSpi> volumes = cluster.getDataNodes().get(0).getFSDataset().getVolumes();
// Make sure deleted replica does not have a copy on either finalized dir of
// transient volume or finalized dir of non-transient volume
for (FsVolumeSpi v : volumes) {
FsVolumeImpl volume = (FsVolumeImpl) v;
File targetDir =
(v.isTransientStorage())
? volume.getBlockPoolSlice(bpid).getFinalizedDir()
: volume.getBlockPoolSlice(bpid).getLazypersistDir();
if (verifyBlockDeletedFromDir(targetDir, locatedBlocks) == false) {
return false;
}
}
return true;
}
@BeforeClass
public static void clusterSetupAtBegining()
throws IOException, LoginException, URISyntaxException {
SupportsBlocks = true;
CONF.setBoolean(DFSConfigKeys.DFS_NAMENODE_DELEGATION_TOKEN_ALWAYS_USE_KEY, true);
cluster =
new MiniDFSCluster.Builder(CONF)
.nnTopology(MiniDFSNNTopology.simpleFederatedTopology(2))
.numDataNodes(2)
.build();
cluster.waitClusterUp();
fHdfs = cluster.getFileSystem(0);
fHdfs2 = cluster.getFileSystem(1);
fHdfs
.getConf()
.set(CommonConfigurationKeys.FS_DEFAULT_NAME_KEY, FsConstants.VIEWFS_URI.toString());
fHdfs2
.getConf()
.set(CommonConfigurationKeys.FS_DEFAULT_NAME_KEY, FsConstants.VIEWFS_URI.toString());
defaultWorkingDirectory =
fHdfs.makeQualified(
new Path("/user/" + UserGroupInformation.getCurrentUser().getShortUserName()));
defaultWorkingDirectory2 =
fHdfs2.makeQualified(
new Path("/user/" + UserGroupInformation.getCurrentUser().getShortUserName()));
fHdfs.mkdirs(defaultWorkingDirectory);
fHdfs2.mkdirs(defaultWorkingDirectory2);
}
/** tests basedir option copying files from dfs file system to dfs file system */
public void testBasedir() throws Exception {
String namenode = null;
MiniDFSCluster cluster = null;
try {
Configuration conf = new Configuration();
cluster = new MiniDFSCluster(conf, 2, true, null);
final FileSystem hdfs = cluster.getFileSystem();
namenode = FileSystem.getDefaultUri(conf).toString();
if (namenode.startsWith("hdfs://")) {
MyFile[] files = createFiles(URI.create(namenode), "/basedir/middle/srcdat");
ToolRunner.run(
new DistCpV1(conf),
new String[] {
"-basedir", "/basedir", namenode + "/basedir/middle/srcdat", namenode + "/destdat"
});
assertTrue(
"Source and destination directories do not match.",
checkFiles(hdfs, "/destdat/middle/srcdat", files));
deldir(hdfs, "/destdat");
deldir(hdfs, "/basedir");
deldir(hdfs, "/logs");
}
} finally {
if (cluster != null) {
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);
}
/* 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(timeout = 60000)
public void testSymlinkHdfsDisable() throws Exception {
Configuration conf = new HdfsConfiguration();
// disable symlink resolution
conf.setBoolean(CommonConfigurationKeys.FS_CLIENT_RESOLVE_REMOTE_SYMLINKS_KEY, false);
// spin up minicluster, get dfs and filecontext
MiniDFSCluster cluster = new MiniDFSCluster.Builder(conf).build();
DistributedFileSystem dfs = cluster.getFileSystem();
FileContext fc = FileContext.getFileContext(cluster.getURI(0), conf);
// Create test files/links
FileContextTestHelper helper = new FileContextTestHelper("/tmp/TestSymlinkHdfsDisable");
Path root = helper.getTestRootPath(fc);
Path target = new Path(root, "target");
Path link = new Path(root, "link");
DFSTestUtil.createFile(dfs, target, 4096, (short) 1, 0xDEADDEAD);
fc.createSymlink(target, link, false);
// Try to resolve links with FileSystem and FileContext
try {
fc.open(link);
fail("Expected error when attempting to resolve link");
} catch (IOException e) {
GenericTestUtils.assertExceptionContains("resolution is disabled", e);
}
try {
dfs.open(link);
fail("Expected error when attempting to resolve link");
} catch (IOException e) {
GenericTestUtils.assertExceptionContains("resolution is disabled", e);
}
}
/* 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();
}
}
@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();
}
}
}
/* 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();
}
}
/** 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 createHDFS() {
try {
Configuration hdConf = new Configuration();
File baseDir = new File("./target/hdfs/hdfsTest").getAbsoluteFile();
FileUtil.fullyDelete(baseDir);
hdConf.set(MiniDFSCluster.HDFS_MINIDFS_BASEDIR, baseDir.getAbsolutePath());
MiniDFSCluster.Builder builder = new MiniDFSCluster.Builder(hdConf);
hdfsCluster = builder.build();
hdfsURI =
"hdfs://" + hdfsCluster.getURI().getHost() + ":" + hdfsCluster.getNameNodePort() + "/";
hdPath = new org.apache.hadoop.fs.Path("/test");
hdfs = hdPath.getFileSystem(hdConf);
FSDataOutputStream stream = hdfs.create(hdPath);
for (int i = 0; i < 10; i++) {
stream.write("Hello HDFS\n".getBytes());
}
stream.close();
} catch (Throwable e) {
e.printStackTrace();
Assert.fail("Test failed " + e.getMessage());
}
}
/*
* Since NameNode will not persist any locations of the block, addBlock()
* retry call after restart NN should re-select the locations and return to
* client. refer HDFS-5257
*/
@Test
public void testAddBlockRetryShouldReturnBlockWithLocations() throws Exception {
final String src = "/testAddBlockRetryShouldReturnBlockWithLocations";
NamenodeProtocols nameNodeRpc = cluster.getNameNodeRpc();
// create file
nameNodeRpc.create(
src,
FsPermission.getFileDefault(),
"clientName",
new EnumSetWritable<CreateFlag>(EnumSet.of(CreateFlag.CREATE)),
true,
(short) 3,
1024,
null);
// start first addBlock()
LOG.info("Starting first addBlock for " + src);
LocatedBlock lb1 =
nameNodeRpc.addBlock(src, "clientName", null, null, INodeId.GRANDFATHER_INODE_ID, null);
assertTrue("Block locations should be present", lb1.getLocations().length > 0);
cluster.restartNameNode();
nameNodeRpc = cluster.getNameNodeRpc();
LocatedBlock lb2 =
nameNodeRpc.addBlock(src, "clientName", null, null, INodeId.GRANDFATHER_INODE_ID, null);
assertEquals("Blocks are not equal", lb1.getBlock(), lb2.getBlock());
assertTrue("Wrong locations with retry", lb2.getLocations().length > 0);
}
/**
* 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);
}
@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);
}
@Before
@Override
public void setUp() throws Exception {
super.setUp();
conf.setClass(
PolicyProvider.POLICY_PROVIDER_CONFIG, HDFSPolicyProvider.class, PolicyProvider.class);
// Many of the tests expect a replication value of 1 in the output
conf.setInt(DFSConfigKeys.DFS_REPLICATION_KEY, 1);
// Build racks and hosts configuration to test dfsAdmin -printTopology
String[] racks = {
"/rack1", "/rack1", "/rack2", "/rack2",
"/rack2", "/rack3", "/rack4", "/rack4"
};
String[] hosts = {
"host1", "host2", "host3", "host4",
"host5", "host6", "host7", "host8"
};
dfsCluster = new MiniDFSCluster.Builder(conf).numDataNodes(8).racks(racks).hosts(hosts).build();
dfsCluster.waitClusterUp();
namenode = conf.get(DFSConfigKeys.FS_DEFAULT_NAME_KEY, "file:///");
username = System.getProperty("user.name");
fs = dfsCluster.getFileSystem();
assertTrue("Not a HDFS: " + fs.getUri(), fs instanceof DistributedFileSystem);
}
/**
* Verify that the NameNode is able to still use <tt>READ_ONLY_SHARED</tt> replicas even when the
* single NORMAL replica is offline (and the effective replication count is 0).
*/
@Test
public void testNormalReplicaOffline() throws Exception {
// Stop the datanode hosting the NORMAL replica
cluster.stopDataNode(normalDataNode.getXferAddr());
// Force NameNode to detect that the datanode is down
BlockManagerTestUtil.noticeDeadDatanode(cluster.getNameNode(), normalDataNode.getXferAddr());
// The live replica count should now be zero (since the NORMAL replica is offline)
NumberReplicas numberReplicas = blockManager.countNodes(block);
assertThat(numberReplicas.liveReplicas(), is(0));
// The block should be reported as under-replicated
BlockManagerTestUtil.updateState(blockManager);
assertThat(blockManager.getUnderReplicatedBlocksCount(), is(1L));
// The BlockManager should be able to heal the replication count back to 1
// by triggering an inter-datanode replication from one of the READ_ONLY_SHARED replicas
BlockManagerTestUtil.computeAllPendingWork(blockManager);
DFSTestUtil.waitForReplication(cluster, extendedBlock, 1, 1, 0);
// There should now be 2 *locations* for the block, and 1 *replica*
assertThat(getLocatedBlock().getLocations().length, is(2));
validateNumberReplicas(1);
}
/**
* Test if fsck can return -1 in case of failure
*
* @throws Exception
*/
public void testFsckError() throws Exception {
MiniDFSCluster cluster = null;
try {
// bring up a one-node cluster
Configuration conf = new Configuration();
cluster = new MiniDFSCluster(conf, 1, true, null);
String fileName = "/test.txt";
Path filePath = new Path(fileName);
FileSystem fs = cluster.getFileSystem();
// create a one-block file
DFSTestUtil.createFile(fs, filePath, 1L, (short) 1, 1L);
DFSTestUtil.waitReplication(fs, filePath, (short) 1);
// intentionally corrupt NN data structure
INodeFile node = (INodeFile) cluster.getNameNode().namesystem.dir.rootDir.getNode(fileName);
assertEquals(node.blocks.length, 1);
node.blocks[0].setNumBytes(-1L); // set the block length to be negative
// run fsck and expect a failure with -1 as the error code
String outStr = runFsck(conf, -1, true, fileName);
System.out.println(outStr);
assertTrue(outStr.contains(NamenodeFsck.FAILURE_STATUS));
// clean up file system
fs.delete(filePath, true);
} finally {
if (cluster != null) {
cluster.shutdown();
}
}
}