use of org.apache.hadoop.hdfs.server.blockmanagement.BlockManager in project hadoop by apache.
the class TestReadStripedFileWithDecoding method testReportBadBlock.
/**
* After reading a corrupted block, make sure the client can correctly report
* the corruption to the NameNode.
*/
@Test
public void testReportBadBlock() throws IOException {
// create file
final Path file = new Path("/corrupted");
// length of "corruption"
final int length = 10;
final byte[] bytes = StripedFileTestUtil.generateBytes(length);
DFSTestUtil.writeFile(fs, file, bytes);
// corrupt the first data block
int dnIndex = findFirstDataNode(file, cellSize * dataBlocks);
Assert.assertNotEquals(-1, dnIndex);
LocatedStripedBlock slb = (LocatedStripedBlock) fs.getClient().getLocatedBlocks(file.toString(), 0, cellSize * dataBlocks).get(0);
final LocatedBlock[] blks = StripedBlockUtil.parseStripedBlockGroup(slb, cellSize, dataBlocks, parityBlocks);
// find the first block file
File storageDir = cluster.getInstanceStorageDir(dnIndex, 0);
File blkFile = MiniDFSCluster.getBlockFile(storageDir, blks[0].getBlock());
Assert.assertTrue("Block file does not exist", blkFile.exists());
// corrupt the block file
LOG.info("Deliberately corrupting file " + blkFile.getName());
try (FileOutputStream out = new FileOutputStream(blkFile)) {
out.write("corruption".getBytes());
}
// in NameNode
for (DataNode dn : cluster.getDataNodes()) {
DataNodeTestUtils.setHeartbeatsDisabledForTests(dn, true);
}
try {
// do stateful read
StripedFileTestUtil.verifyStatefulRead(fs, file, length, bytes, ByteBuffer.allocate(1024));
// check whether the corruption has been reported to the NameNode
final FSNamesystem ns = cluster.getNamesystem();
final BlockManager bm = ns.getBlockManager();
BlockInfo blockInfo = (ns.getFSDirectory().getINode4Write(file.toString()).asFile().getBlocks())[0];
Assert.assertEquals(1, bm.getCorruptReplicas(blockInfo).size());
} finally {
for (DataNode dn : cluster.getDataNodes()) {
DataNodeTestUtils.setHeartbeatsDisabledForTests(dn, false);
}
}
}
use of org.apache.hadoop.hdfs.server.blockmanagement.BlockManager in project hadoop by apache.
the class TestDeadDatanode method testDeadDatanode.
/**
* Test to ensure namenode rejects request from dead datanode
* - Start a cluster
* - Shutdown the datanode and wait for it to be marked dead at the namenode
* - Send datanode requests to Namenode and make sure it is rejected
* appropriately.
*/
@Test
public void testDeadDatanode() throws Exception {
Configuration conf = new HdfsConfiguration();
conf.setInt(DFSConfigKeys.DFS_NAMENODE_HEARTBEAT_RECHECK_INTERVAL_KEY, 500);
conf.setLong(DFSConfigKeys.DFS_HEARTBEAT_INTERVAL_KEY, 1L);
cluster = new MiniDFSCluster.Builder(conf).build();
cluster.waitActive();
String poolId = cluster.getNamesystem().getBlockPoolId();
// wait for datanode to be marked live
DataNode dn = cluster.getDataNodes().get(0);
DatanodeRegistration reg = InternalDataNodeTestUtils.getDNRegistrationForBP(cluster.getDataNodes().get(0), poolId);
DFSTestUtil.waitForDatanodeState(cluster, reg.getDatanodeUuid(), true, 20000);
// Shutdown and wait for datanode to be marked dead
dn.shutdown();
DFSTestUtil.waitForDatanodeState(cluster, reg.getDatanodeUuid(), false, 20000);
DatanodeProtocol dnp = cluster.getNameNodeRpc();
ReceivedDeletedBlockInfo[] blocks = { new ReceivedDeletedBlockInfo(new Block(0), ReceivedDeletedBlockInfo.BlockStatus.RECEIVED_BLOCK, null) };
StorageReceivedDeletedBlocks[] storageBlocks = { new StorageReceivedDeletedBlocks(new DatanodeStorage(reg.getDatanodeUuid()), blocks) };
// Ensure blockReceived call from dead datanode is not rejected with
// IOException, since it's async, but the node remains unregistered.
dnp.blockReceivedAndDeleted(reg, poolId, storageBlocks);
BlockManager bm = cluster.getNamesystem().getBlockManager();
// IBRs are async, make sure the NN processes all of them.
bm.flushBlockOps();
assertFalse(bm.getDatanodeManager().getDatanode(reg).isRegistered());
// Ensure blockReport from dead datanode is rejected with IOException
StorageBlockReport[] report = { new StorageBlockReport(new DatanodeStorage(reg.getDatanodeUuid()), BlockListAsLongs.EMPTY) };
try {
dnp.blockReport(reg, poolId, report, new BlockReportContext(1, 0, System.nanoTime(), 0L, true));
fail("Expected IOException is not thrown");
} catch (IOException ex) {
// Expected
}
// Ensure heartbeat from dead datanode is rejected with a command
// that asks datanode to register again
StorageReport[] rep = { new StorageReport(new DatanodeStorage(reg.getDatanodeUuid()), false, 0, 0, 0, 0, 0) };
DatanodeCommand[] cmd = dnp.sendHeartbeat(reg, rep, 0L, 0L, 0, 0, 0, null, true, SlowPeerReports.EMPTY_REPORT).getCommands();
assertEquals(1, cmd.length);
assertEquals(cmd[0].getAction(), RegisterCommand.REGISTER.getAction());
}
use of org.apache.hadoop.hdfs.server.blockmanagement.BlockManager in project hadoop by apache.
the class TestDeadDatanode method testDeadNodeAsBlockTarget.
@Test
public void testDeadNodeAsBlockTarget() throws Exception {
Configuration conf = new HdfsConfiguration();
conf.setInt(DFSConfigKeys.DFS_NAMENODE_HEARTBEAT_RECHECK_INTERVAL_KEY, 500);
conf.setLong(DFSConfigKeys.DFS_HEARTBEAT_INTERVAL_KEY, 1L);
cluster = new MiniDFSCluster.Builder(conf).numDataNodes(3).build();
cluster.waitActive();
String poolId = cluster.getNamesystem().getBlockPoolId();
// wait for datanode to be marked live
DataNode dn = cluster.getDataNodes().get(0);
DatanodeRegistration reg = InternalDataNodeTestUtils.getDNRegistrationForBP(cluster.getDataNodes().get(0), poolId);
// Get the updated datanode descriptor
BlockManager bm = cluster.getNamesystem().getBlockManager();
DatanodeManager dm = bm.getDatanodeManager();
Node clientNode = dm.getDatanode(reg);
DFSTestUtil.waitForDatanodeState(cluster, reg.getDatanodeUuid(), true, 20000);
// Shutdown and wait for datanode to be marked dead
dn.shutdown();
DFSTestUtil.waitForDatanodeState(cluster, reg.getDatanodeUuid(), false, 20000);
// Get the updated datanode descriptor available in DNM
// choose the targets, but local node should not get selected as this is not
// part of the cluster anymore
DatanodeStorageInfo[] results = bm.chooseTarget4NewBlock("/hello", 3, clientNode, new HashSet<Node>(), 256 * 1024 * 1024L, null, (byte) 7, BlockType.CONTIGUOUS, null);
for (DatanodeStorageInfo datanodeStorageInfo : results) {
assertFalse("Dead node should not be choosen", datanodeStorageInfo.getDatanodeDescriptor().equals(clientNode));
}
}
use of org.apache.hadoop.hdfs.server.blockmanagement.BlockManager in project hadoop by apache.
the class TestDecommission method testRecommission.
/**
* Test that over-replicated blocks are deleted on recommission.
*/
@Test(timeout = 120000)
public void testRecommission() throws Exception {
final int numDatanodes = 6;
try {
LOG.info("Starting test testRecommission");
startCluster(1, numDatanodes);
final Path file1 = new Path("testDecommission.dat");
final int replicas = numDatanodes - 1;
ArrayList<DatanodeInfo> decommissionedNodes = Lists.newArrayList();
final FileSystem fileSys = getCluster().getFileSystem();
// Write a file to n-1 datanodes
writeFile(fileSys, file1, replicas);
// Decommission one of the datanodes with a replica
BlockLocation loc = fileSys.getFileBlockLocations(file1, 0, 1)[0];
assertEquals("Unexpected number of replicas from getFileBlockLocations", replicas, loc.getHosts().length);
final String toDecomHost = loc.getNames()[0];
String toDecomUuid = null;
for (DataNode d : getCluster().getDataNodes()) {
if (d.getDatanodeId().getXferAddr().equals(toDecomHost)) {
toDecomUuid = d.getDatanodeId().getDatanodeUuid();
break;
}
}
assertNotNull("Could not find a dn with the block!", toDecomUuid);
final DatanodeInfo decomNode = takeNodeOutofService(0, toDecomUuid, 0, decommissionedNodes, AdminStates.DECOMMISSIONED);
decommissionedNodes.add(decomNode);
final BlockManager blockManager = getCluster().getNamesystem().getBlockManager();
final DatanodeManager datanodeManager = blockManager.getDatanodeManager();
BlockManagerTestUtil.recheckDecommissionState(datanodeManager);
// Ensure decommissioned datanode is not automatically shutdown
DFSClient client = getDfsClient(0);
assertEquals("All datanodes must be alive", numDatanodes, client.datanodeReport(DatanodeReportType.LIVE).length);
// wait for the block to be replicated
final ExtendedBlock b = DFSTestUtil.getFirstBlock(fileSys, file1);
final String uuid = toDecomUuid;
GenericTestUtils.waitFor(new Supplier<Boolean>() {
@Override
public Boolean get() {
BlockInfo info = blockManager.getStoredBlock(b.getLocalBlock());
int count = 0;
StringBuilder sb = new StringBuilder("Replica locations: ");
for (int i = 0; i < info.numNodes(); i++) {
DatanodeDescriptor dn = info.getDatanode(i);
sb.append(dn + ", ");
if (!dn.getDatanodeUuid().equals(uuid)) {
count++;
}
}
LOG.info(sb.toString());
LOG.info("Count: " + count);
return count == replicas;
}
}, 500, 30000);
// redecommission and wait for over-replication to be fixed
putNodeInService(0, decomNode);
BlockManagerTestUtil.recheckDecommissionState(datanodeManager);
DFSTestUtil.waitForReplication(getCluster(), b, 1, replicas, 0);
cleanupFile(fileSys, file1);
} finally {
shutdownCluster();
}
}
use of org.apache.hadoop.hdfs.server.blockmanagement.BlockManager in project hadoop by apache.
the class TestDecommission method testUsedCapacity.
@Test
public void testUsedCapacity() throws Exception {
int numNamenodes = 1;
int numDatanodes = 2;
startCluster(numNamenodes, numDatanodes);
FSNamesystem ns = getCluster().getNamesystem(0);
BlockManager blockManager = ns.getBlockManager();
DatanodeStatistics datanodeStatistics = blockManager.getDatanodeManager().getDatanodeStatistics();
long initialUsedCapacity = datanodeStatistics.getCapacityUsed();
long initialTotalCapacity = datanodeStatistics.getCapacityTotal();
long initialBlockPoolUsed = datanodeStatistics.getBlockPoolUsed();
ArrayList<ArrayList<DatanodeInfo>> namenodeDecomList = new ArrayList<ArrayList<DatanodeInfo>>(numNamenodes);
namenodeDecomList.add(0, new ArrayList<>(numDatanodes));
ArrayList<DatanodeInfo> decommissionedNodes = namenodeDecomList.get(0);
//decommission one node
DatanodeInfo decomNode = takeNodeOutofService(0, null, 0, decommissionedNodes, AdminStates.DECOMMISSIONED);
decommissionedNodes.add(decomNode);
long newUsedCapacity = datanodeStatistics.getCapacityUsed();
long newTotalCapacity = datanodeStatistics.getCapacityTotal();
long newBlockPoolUsed = datanodeStatistics.getBlockPoolUsed();
assertTrue("DfsUsedCapacity should not be the same after a node has " + "been decommissioned!", initialUsedCapacity != newUsedCapacity);
assertTrue("TotalCapacity should not be the same after a node has " + "been decommissioned!", initialTotalCapacity != newTotalCapacity);
assertTrue("BlockPoolUsed should not be the same after a node has " + "been decommissioned!", initialBlockPoolUsed != newBlockPoolUsed);
}
Aggregations