/**
* 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();
}
}
}
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;
}
private int getTrueReplication(MiniDFSCluster cluster, ExtendedBlock block) throws IOException {
int count = 0;
for (DataNode dn : cluster.getDataNodes()) {
if (DataNodeTestUtils.getFSDataset(dn)
.getStoredBlock(block.getBlockPoolId(), block.getBlockId())
!= null) {
count++;
}
}
return count;
}
/**
* Corrupt a block on a data node. Replace the block file content with content of 1, 2,
* ...BLOCK_SIZE.
*
* @param block the ExtendedBlock to be corrupted
* @param dn the data node where the block needs to be corrupted
* @throws FileNotFoundException
* @throws IOException
*/
private static void corruptBlock(final ExtendedBlock block, final DataNode dn)
throws FileNotFoundException, IOException {
final File f =
DataNodeTestUtils.getBlockFile(dn, block.getBlockPoolId(), block.getLocalBlock());
final RandomAccessFile raFile = new RandomAccessFile(f, "rw");
final byte[] bytes = new byte[(int) BLOCK_SIZE];
for (int i = 0; i < BLOCK_SIZE; i++) {
bytes[i] = (byte) (i);
}
raFile.write(bytes);
raFile.close();
}
/**
* 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());
}
/**
* Another regression test for HDFS-2742. This tests the following sequence: - DN does a block
* report while file is open. This BR contains the block in RBW state. - The block report is
* delayed in reaching the standby. - The file is closed. - The standby processes the OP_ADD and
* OP_CLOSE operations before the RBW block report arrives. - The standby should not mark the
* block as corrupt.
*/
@Test
public void testRBWReportArrivesAfterEdits() throws Exception {
final CountDownLatch brFinished = new CountDownLatch(1);
DelayAnswer delayer =
new GenericTestUtils.DelayAnswer(LOG) {
@Override
protected Object passThrough(InvocationOnMock invocation) throws Throwable {
try {
return super.passThrough(invocation);
} finally {
// inform the test that our block report went through.
brFinished.countDown();
}
}
};
FSDataOutputStream out = fs.create(TEST_FILE_PATH);
try {
AppendTestUtil.write(out, 0, 10);
out.hflush();
DataNode dn = cluster.getDataNodes().get(0);
DatanodeProtocolClientSideTranslatorPB spy = DataNodeTestUtils.spyOnBposToNN(dn, nn2);
Mockito.doAnswer(delayer)
.when(spy)
.blockReport(
Mockito.<DatanodeRegistration>anyObject(),
Mockito.anyString(),
Mockito.<StorageBlockReport[]>anyObject());
dn.scheduleAllBlockReport(0);
delayer.waitForCall();
} finally {
IOUtils.closeStream(out);
}
cluster.transitionToStandby(0);
cluster.transitionToActive(1);
delayer.proceed();
brFinished.await();
// 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);
}
// test writeToRbw
@Test
public void testWriteToRbw() throws Exception {
MiniDFSCluster cluster = new MiniDFSCluster.Builder(new HdfsConfiguration()).build();
try {
cluster.waitActive();
DataNode dn = cluster.getDataNodes().get(0);
FsDatasetImpl dataSet = (FsDatasetImpl) DataNodeTestUtils.getFSDataset(dn);
// set up replicasMap
String bpid = cluster.getNamesystem().getBlockPoolId();
ExtendedBlock[] blocks = setup(bpid, dataSet);
// test writeToRbw
testWriteToRbw(dataSet, blocks);
} finally {
cluster.shutdown();
}
}
/**
* The following test first creates a file with a few blocks. It randomly truncates the replica of
* the last block stored in each datanode. Finally, it triggers block synchronization to
* synchronize all stored block.
*/
public void testBlockSynchronization() throws Exception {
final int ORG_FILE_SIZE = 3000;
Configuration conf = new HdfsConfiguration();
conf.setLong(DFSConfigKeys.DFS_BLOCK_SIZE_KEY, BLOCK_SIZE);
MiniDFSCluster cluster = null;
try {
cluster = new MiniDFSCluster.Builder(conf).numDataNodes(5).build();
cluster.waitActive();
// create a file
DistributedFileSystem dfs = (DistributedFileSystem) cluster.getFileSystem();
String filestr = "/foo";
Path filepath = new Path(filestr);
DFSTestUtil.createFile(dfs, filepath, ORG_FILE_SIZE, REPLICATION_NUM, 0L);
assertTrue(dfs.exists(filepath));
DFSTestUtil.waitReplication(dfs, filepath, REPLICATION_NUM);
// get block info for the last block
LocatedBlock locatedblock =
TestInterDatanodeProtocol.getLastLocatedBlock(dfs.dfs.getNamenode(), filestr);
DatanodeInfo[] datanodeinfos = locatedblock.getLocations();
assertEquals(REPLICATION_NUM, datanodeinfos.length);
// connect to data nodes
DataNode[] datanodes = new DataNode[REPLICATION_NUM];
for (int i = 0; i < REPLICATION_NUM; i++) {
datanodes[i] = cluster.getDataNode(datanodeinfos[i].getIpcPort());
assertTrue(datanodes[i] != null);
}
// verify Block Info
ExtendedBlock lastblock = locatedblock.getBlock();
DataNode.LOG.info("newblocks=" + lastblock);
for (int i = 0; i < REPLICATION_NUM; i++) {
checkMetaInfo(lastblock, datanodes[i]);
}
DataNode.LOG.info("dfs.dfs.clientName=" + dfs.dfs.clientName);
cluster.getNameNodeRpc().append(filestr, dfs.dfs.clientName);
// expire lease to trigger block recovery.
waitLeaseRecovery(cluster);
Block[] updatedmetainfo = new Block[REPLICATION_NUM];
long oldSize = lastblock.getNumBytes();
lastblock =
TestInterDatanodeProtocol.getLastLocatedBlock(dfs.dfs.getNamenode(), filestr).getBlock();
long currentGS = lastblock.getGenerationStamp();
for (int i = 0; i < REPLICATION_NUM; i++) {
updatedmetainfo[i] =
DataNodeTestUtils.getFSDataset(datanodes[i])
.getStoredBlock(lastblock.getBlockPoolId(), lastblock.getBlockId());
assertEquals(lastblock.getBlockId(), updatedmetainfo[i].getBlockId());
assertEquals(oldSize, updatedmetainfo[i].getNumBytes());
assertEquals(currentGS, updatedmetainfo[i].getGenerationStamp());
}
// verify that lease recovery does not occur when namenode is in safemode
System.out.println("Testing that lease recovery cannot happen during safemode.");
filestr = "/foo.safemode";
filepath = new Path(filestr);
dfs.create(filepath, (short) 1);
cluster.getNameNodeRpc().setSafeMode(HdfsConstants.SafeModeAction.SAFEMODE_ENTER);
assertTrue(dfs.dfs.exists(filestr));
DFSTestUtil.waitReplication(dfs, filepath, (short) 1);
waitLeaseRecovery(cluster);
// verify that we still cannot recover the lease
LeaseManager lm = NameNodeAdapter.getLeaseManager(cluster.getNamesystem());
assertTrue("Found " + lm.countLease() + " lease, expected 1", lm.countLease() == 1);
cluster.getNameNodeRpc().setSafeMode(HdfsConstants.SafeModeAction.SAFEMODE_LEAVE);
} finally {
if (cluster != null) {
cluster.shutdown();
}
}
}
/**
* Test if {@link FSNamesystem#handleHeartbeat} can pick up replication and/or invalidate requests
* and observes the max limit
*/
@Test
public void testHeartbeat() throws Exception {
final Configuration conf = new HdfsConfiguration();
final MiniDFSCluster cluster = new MiniDFSCluster.Builder(conf).build();
try {
cluster.waitActive();
final FSNamesystem namesystem = cluster.getNamesystem();
final HeartbeatManager hm =
namesystem.getBlockManager().getDatanodeManager().getHeartbeatManager();
final String poolId = namesystem.getBlockPoolId();
final DatanodeRegistration nodeReg =
DataNodeTestUtils.getDNRegistrationForBP(cluster.getDataNodes().get(0), poolId);
final DatanodeDescriptor dd = NameNodeAdapter.getDatanode(namesystem, nodeReg);
final String storageID = DatanodeStorage.generateUuid();
dd.updateStorage(new DatanodeStorage(storageID));
final int REMAINING_BLOCKS = 1;
final int MAX_REPLICATE_LIMIT =
conf.getInt(DFSConfigKeys.DFS_NAMENODE_REPLICATION_MAX_STREAMS_KEY, 2);
final int MAX_INVALIDATE_LIMIT = DFSConfigKeys.DFS_BLOCK_INVALIDATE_LIMIT_DEFAULT;
final int MAX_INVALIDATE_BLOCKS = 2 * MAX_INVALIDATE_LIMIT + REMAINING_BLOCKS;
final int MAX_REPLICATE_BLOCKS = 2 * MAX_REPLICATE_LIMIT + REMAINING_BLOCKS;
final DatanodeStorageInfo[] ONE_TARGET = {dd.getStorageInfo(storageID)};
try {
namesystem.writeLock();
synchronized (hm) {
for (int i = 0; i < MAX_REPLICATE_BLOCKS; i++) {
dd.addBlockToBeReplicated(
new Block(i, 0, GenerationStamp.LAST_RESERVED_STAMP), ONE_TARGET);
}
DatanodeCommand[] cmds =
NameNodeAdapter.sendHeartBeat(nodeReg, dd, namesystem).getCommands();
assertEquals(1, cmds.length);
assertEquals(DatanodeProtocol.DNA_TRANSFER, cmds[0].getAction());
assertEquals(MAX_REPLICATE_LIMIT, ((BlockCommand) cmds[0]).getBlocks().length);
ArrayList<Block> blockList = new ArrayList<Block>(MAX_INVALIDATE_BLOCKS);
for (int i = 0; i < MAX_INVALIDATE_BLOCKS; i++) {
blockList.add(new Block(i, 0, GenerationStamp.LAST_RESERVED_STAMP));
}
dd.addBlocksToBeInvalidated(blockList);
cmds = NameNodeAdapter.sendHeartBeat(nodeReg, dd, namesystem).getCommands();
assertEquals(2, cmds.length);
assertEquals(DatanodeProtocol.DNA_TRANSFER, cmds[0].getAction());
assertEquals(MAX_REPLICATE_LIMIT, ((BlockCommand) cmds[0]).getBlocks().length);
assertEquals(DatanodeProtocol.DNA_INVALIDATE, cmds[1].getAction());
assertEquals(MAX_INVALIDATE_LIMIT, ((BlockCommand) cmds[1]).getBlocks().length);
cmds = NameNodeAdapter.sendHeartBeat(nodeReg, dd, namesystem).getCommands();
assertEquals(2, cmds.length);
assertEquals(DatanodeProtocol.DNA_TRANSFER, cmds[0].getAction());
assertEquals(REMAINING_BLOCKS, ((BlockCommand) cmds[0]).getBlocks().length);
assertEquals(DatanodeProtocol.DNA_INVALIDATE, cmds[1].getAction());
assertEquals(MAX_INVALIDATE_LIMIT, ((BlockCommand) cmds[1]).getBlocks().length);
cmds = NameNodeAdapter.sendHeartBeat(nodeReg, dd, namesystem).getCommands();
assertEquals(1, cmds.length);
assertEquals(DatanodeProtocol.DNA_INVALIDATE, cmds[0].getAction());
assertEquals(REMAINING_BLOCKS, ((BlockCommand) cmds[0]).getBlocks().length);
cmds = NameNodeAdapter.sendHeartBeat(nodeReg, dd, namesystem).getCommands();
assertEquals(0, cmds.length);
}
} finally {
namesystem.writeUnlock();
}
} finally {
cluster.shutdown();
}
}
/**
* Test if {@link FSNamesystem#handleHeartbeat} correctly selects data node targets for block
* recovery.
*/
@Test
public void testHeartbeatBlockRecovery() throws Exception {
final Configuration conf = new HdfsConfiguration();
final MiniDFSCluster cluster = new MiniDFSCluster.Builder(conf).numDataNodes(3).build();
try {
cluster.waitActive();
final FSNamesystem namesystem = cluster.getNamesystem();
final HeartbeatManager hm =
namesystem.getBlockManager().getDatanodeManager().getHeartbeatManager();
final String poolId = namesystem.getBlockPoolId();
final DatanodeRegistration nodeReg1 =
DataNodeTestUtils.getDNRegistrationForBP(cluster.getDataNodes().get(0), poolId);
final DatanodeDescriptor dd1 = NameNodeAdapter.getDatanode(namesystem, nodeReg1);
dd1.updateStorage(new DatanodeStorage(DatanodeStorage.generateUuid()));
final DatanodeRegistration nodeReg2 =
DataNodeTestUtils.getDNRegistrationForBP(cluster.getDataNodes().get(1), poolId);
final DatanodeDescriptor dd2 = NameNodeAdapter.getDatanode(namesystem, nodeReg2);
dd2.updateStorage(new DatanodeStorage(DatanodeStorage.generateUuid()));
final DatanodeRegistration nodeReg3 =
DataNodeTestUtils.getDNRegistrationForBP(cluster.getDataNodes().get(2), poolId);
final DatanodeDescriptor dd3 = NameNodeAdapter.getDatanode(namesystem, nodeReg3);
dd3.updateStorage(new DatanodeStorage(DatanodeStorage.generateUuid()));
try {
namesystem.writeLock();
synchronized (hm) {
NameNodeAdapter.sendHeartBeat(nodeReg1, dd1, namesystem);
NameNodeAdapter.sendHeartBeat(nodeReg2, dd2, namesystem);
NameNodeAdapter.sendHeartBeat(nodeReg3, dd3, namesystem);
// Test with all alive nodes.
DFSTestUtil.resetLastUpdatesWithOffset(dd1, 0);
DFSTestUtil.resetLastUpdatesWithOffset(dd2, 0);
DFSTestUtil.resetLastUpdatesWithOffset(dd3, 0);
final DatanodeStorageInfo[] storages = {
dd1.getStorageInfos()[0], dd2.getStorageInfos()[0], dd3.getStorageInfos()[0]
};
BlockInfo blockInfo =
new BlockInfoContiguous(
new Block(0, 0, GenerationStamp.LAST_RESERVED_STAMP), (short) 3);
blockInfo.convertToBlockUnderConstruction(BlockUCState.UNDER_RECOVERY, storages);
dd1.addBlockToBeRecovered(blockInfo);
DatanodeCommand[] cmds =
NameNodeAdapter.sendHeartBeat(nodeReg1, dd1, namesystem).getCommands();
assertEquals(1, cmds.length);
assertEquals(DatanodeProtocol.DNA_RECOVERBLOCK, cmds[0].getAction());
BlockRecoveryCommand recoveryCommand = (BlockRecoveryCommand) cmds[0];
assertEquals(1, recoveryCommand.getRecoveringBlocks().size());
DatanodeInfo[] recoveringNodes =
recoveryCommand
.getRecoveringBlocks()
.toArray(new BlockRecoveryCommand.RecoveringBlock[0])[0]
.getLocations();
assertEquals(3, recoveringNodes.length);
assertEquals(recoveringNodes[0], dd1);
assertEquals(recoveringNodes[1], dd2);
assertEquals(recoveringNodes[2], dd3);
// Test with one stale node.
DFSTestUtil.resetLastUpdatesWithOffset(dd1, 0);
// More than the default stale interval of 30 seconds.
DFSTestUtil.resetLastUpdatesWithOffset(dd2, -40 * 1000);
DFSTestUtil.resetLastUpdatesWithOffset(dd3, 0);
blockInfo =
new BlockInfoContiguous(
new Block(0, 0, GenerationStamp.LAST_RESERVED_STAMP), (short) 3);
blockInfo.convertToBlockUnderConstruction(BlockUCState.UNDER_RECOVERY, storages);
dd1.addBlockToBeRecovered(blockInfo);
cmds = NameNodeAdapter.sendHeartBeat(nodeReg1, dd1, namesystem).getCommands();
assertEquals(1, cmds.length);
assertEquals(DatanodeProtocol.DNA_RECOVERBLOCK, cmds[0].getAction());
recoveryCommand = (BlockRecoveryCommand) cmds[0];
assertEquals(1, recoveryCommand.getRecoveringBlocks().size());
recoveringNodes =
recoveryCommand
.getRecoveringBlocks()
.toArray(new BlockRecoveryCommand.RecoveringBlock[0])[0]
.getLocations();
assertEquals(2, recoveringNodes.length);
// dd2 is skipped.
assertEquals(recoveringNodes[0], dd1);
assertEquals(recoveringNodes[1], dd3);
// Test with all stale node.
DFSTestUtil.resetLastUpdatesWithOffset(dd1, -60 * 1000);
// More than the default stale interval of 30 seconds.
DFSTestUtil.resetLastUpdatesWithOffset(dd2, -40 * 1000);
DFSTestUtil.resetLastUpdatesWithOffset(dd3, -80 * 1000);
blockInfo =
new BlockInfoContiguous(
new Block(0, 0, GenerationStamp.LAST_RESERVED_STAMP), (short) 3);
blockInfo.convertToBlockUnderConstruction(BlockUCState.UNDER_RECOVERY, storages);
dd1.addBlockToBeRecovered(blockInfo);
cmds = NameNodeAdapter.sendHeartBeat(nodeReg1, dd1, namesystem).getCommands();
assertEquals(1, cmds.length);
assertEquals(DatanodeProtocol.DNA_RECOVERBLOCK, cmds[0].getAction());
recoveryCommand = (BlockRecoveryCommand) cmds[0];
assertEquals(1, recoveryCommand.getRecoveringBlocks().size());
recoveringNodes =
recoveryCommand
.getRecoveringBlocks()
.toArray(new BlockRecoveryCommand.RecoveringBlock[0])[0]
.getLocations();
// Only dd1 is included since it heart beated and hence its not stale
// when the list of recovery blocks is constructed.
assertEquals(3, recoveringNodes.length);
assertEquals(recoveringNodes[0], dd1);
assertEquals(recoveringNodes[1], dd2);
assertEquals(recoveringNodes[2], dd3);
}
} finally {
namesystem.writeUnlock();
}
} finally {
cluster.shutdown();
}
}
/** Test for {@link FsDatasetImpl#updateReplicaUnderRecovery(ExtendedBlock, long, long)} */
@Test
public void testUpdateReplicaUnderRecovery() throws IOException {
MiniDFSCluster cluster = null;
try {
cluster = new MiniDFSCluster.Builder(conf).numDataNodes(3).build();
cluster.waitActive();
String bpid = cluster.getNamesystem().getBlockPoolId();
// create a file
DistributedFileSystem dfs = (DistributedFileSystem) cluster.getFileSystem();
String filestr = "/foo";
Path filepath = new Path(filestr);
DFSTestUtil.createFile(dfs, filepath, 1024L, (short) 3, 0L);
// get block info
final LocatedBlock locatedblock =
getLastLocatedBlock(DFSClientAdapter.getDFSClient(dfs).getNamenode(), filestr);
final DatanodeInfo[] datanodeinfo = locatedblock.getLocations();
Assert.assertTrue(datanodeinfo.length > 0);
// get DataNode and FSDataset objects
final DataNode datanode = cluster.getDataNode(datanodeinfo[0].getIpcPort());
Assert.assertTrue(datanode != null);
// initReplicaRecovery
final ExtendedBlock b = locatedblock.getBlock();
final long recoveryid = b.getGenerationStamp() + 1;
final long newlength = b.getNumBytes() - 1;
final FsDatasetSpi<?> fsdataset = DataNodeTestUtils.getFSDataset(datanode);
final ReplicaRecoveryInfo rri =
fsdataset.initReplicaRecovery(new RecoveringBlock(b, null, recoveryid));
// check replica
final ReplicaInfo replica =
FsDatasetTestUtil.fetchReplicaInfo(fsdataset, bpid, b.getBlockId());
Assert.assertEquals(ReplicaState.RUR, replica.getState());
// check meta data before update
FsDatasetImpl.checkReplicaFiles(replica);
// case "THIS IS NOT SUPPOSED TO HAPPEN"
// with (block length) != (stored replica's on disk length).
{
// create a block with same id and gs but different length.
final ExtendedBlock tmp =
new ExtendedBlock(
b.getBlockPoolId(),
rri.getBlockId(),
rri.getNumBytes() - 1,
rri.getGenerationStamp());
try {
// update should fail
fsdataset.updateReplicaUnderRecovery(tmp, recoveryid, newlength);
Assert.fail();
} catch (IOException ioe) {
System.out.println("GOOD: getting " + ioe);
}
}
// update
final String storageID =
fsdataset.updateReplicaUnderRecovery(
new ExtendedBlock(b.getBlockPoolId(), rri), recoveryid, newlength);
assertTrue(storageID != null);
} finally {
if (cluster != null) cluster.shutdown();
}
}
public static void checkMetaInfo(ExtendedBlock b, DataNode dn) throws IOException {
Block metainfo =
DataNodeTestUtils.getFSDataset(dn).getStoredBlock(b.getBlockPoolId(), b.getBlockId());
Assert.assertEquals(b.getBlockId(), metainfo.getBlockId());
Assert.assertEquals(b.getNumBytes(), metainfo.getNumBytes());
}
protected final void triggerBlockReport() throws IOException, InterruptedException {
// Trigger block report to NN
DataNodeTestUtils.triggerBlockReport(cluster.getDataNodes().get(0));
Thread.sleep(10 * 1000);
}
/**
* Test the case that a replica is reported corrupt while it is not in blocksMap. Make sure that
* ArrayIndexOutOfBounds does not thrown. See Hadoop-4351.
*
* <p>TODO HOPS This test fails as it tries to remove a non-existing replica. Calling
* findAndMarkBlockAsCorrupt from a DataNode that does not store any replica for this specific
* block will lead to a tuple did not exist exception. The reason for this is that
* BlockManager.removeStoredBlock is called with a node that does not store a replica and hence
* the delete will not be able to succeed during commit.
*/
@Test
public void testArrayOutOfBoundsException() throws Exception {
MiniDFSCluster cluster = null;
try {
Configuration conf = new HdfsConfiguration();
cluster = new MiniDFSCluster.Builder(conf).numDataNodes(2).build();
cluster.waitActive();
FileSystem fs = cluster.getFileSystem();
final Path FILE_PATH = new Path("/tmp.txt");
final long FILE_LEN = 1L;
DFSTestUtil.createFile(fs, FILE_PATH, FILE_LEN, (short) 2, 1L);
// get the block
final String bpid = cluster.getNamesystem().getBlockPoolId();
File storageDir = cluster.getInstanceStorageDir(0, 0);
File dataDir = MiniDFSCluster.getFinalizedDir(storageDir, bpid);
assertTrue("Data directory does not exist", dataDir.exists());
ExtendedBlock blk = getBlock(bpid, dataDir);
if (blk == null) {
storageDir = cluster.getInstanceStorageDir(0, 1);
dataDir = MiniDFSCluster.getFinalizedDir(storageDir, bpid);
blk = getBlock(bpid, dataDir);
}
assertFalse(
"Data directory does not contain any blocks or there was an " + "IO error", blk == null);
// start a third datanode
cluster.startDataNodes(conf, 1, true, null, null);
ArrayList<DataNode> datanodes = cluster.getDataNodes();
assertEquals(datanodes.size(), 3);
DataNode dataNode = datanodes.get(2);
// report corrupted block by the third datanode
DatanodeRegistration dnR =
DataNodeTestUtils.getDNRegistrationForBP(dataNode, blk.getBlockPoolId());
// Get the storage id of one of the storages on the datanode
String storageId =
cluster
.getNamesystem()
.getBlockManager()
.getDatanodeManager()
.getDatanode(dataNode.getDatanodeId())
.getStorageInfos()[0]
.getStorageID();
cluster
.getNamesystem()
.getBlockManager()
.findAndMarkBlockAsCorrupt(blk, new DatanodeInfo(dnR), storageId, "some test reason");
// open the file
fs.open(FILE_PATH);
// clean up
fs.delete(FILE_PATH, false);
} finally {
if (cluster != null) {
cluster.shutdown();
}
}
}