Examples with LocatedBlocks org.apache.hadoop.hdfs.protocol.LocatedBlocks used on opensource projects

Example 86 with LocatedBlocks

use of org.apache.hadoop.hdfs.protocol.LocatedBlocks in project hadoop by apache.

the class TestHFlush method doTheJob.

/**
   * The method starts new cluster with defined Configuration; creates a file
   * with specified block_size and writes 10 equal sections in it; it also calls
   * hflush/hsync after each write and throws an IOException in case of an error.
   * 
   * @param conf cluster configuration
   * @param fileName of the file to be created and processed as required
   * @param block_size value to be used for the file's creation
   * @param replicas is the number of replicas
   * @param isSync hsync or hflush         
   * @param syncFlags specify the semantic of the sync/flush
   * @throws IOException in case of any errors
   */
public static void doTheJob(Configuration conf, final String fileName, long block_size, short replicas, boolean isSync, EnumSet<SyncFlag> syncFlags) throws IOException {
    byte[] fileContent;
    final int SECTIONS = 10;
    fileContent = AppendTestUtil.initBuffer(AppendTestUtil.FILE_SIZE);
    MiniDFSCluster cluster = new MiniDFSCluster.Builder(conf).numDataNodes(replicas).build();
    // Make sure we work with DFS in order to utilize all its functionality
    DistributedFileSystem fileSystem = cluster.getFileSystem();
    FSDataInputStream is;
    try {
        Path path = new Path(fileName);
        final String pathName = new Path(fileSystem.getWorkingDirectory(), path).toUri().getPath();
        FSDataOutputStream stm = fileSystem.create(path, false, 4096, replicas, block_size);
        System.out.println("Created file " + fileName);
        int tenth = AppendTestUtil.FILE_SIZE / SECTIONS;
        int rounding = AppendTestUtil.FILE_SIZE - tenth * SECTIONS;
        for (int i = 0; i < SECTIONS; i++) {
            System.out.println("Writing " + (tenth * i) + " to " + (tenth * (i + 1)) + " section to file " + fileName);
            // write to the file
            stm.write(fileContent, tenth * i, tenth);
            // Wait while hflush/hsync pushes all packets through built pipeline
            if (isSync) {
                ((DFSOutputStream) stm.getWrappedStream()).hsync(syncFlags);
            } else {
                ((DFSOutputStream) stm.getWrappedStream()).hflush();
            }
            // Check file length if updatelength is required
            if (isSync && syncFlags.contains(SyncFlag.UPDATE_LENGTH)) {
                long currentFileLength = fileSystem.getFileStatus(path).getLen();
                assertEquals("File size doesn't match for hsync/hflush with updating the length", tenth * (i + 1), currentFileLength);
            } else if (isSync && syncFlags.contains(SyncFlag.END_BLOCK)) {
                LocatedBlocks blocks = fileSystem.dfs.getLocatedBlocks(pathName, 0);
                assertEquals(i + 1, blocks.getLocatedBlocks().size());
            }
            byte[] toRead = new byte[tenth];
            byte[] expected = new byte[tenth];
            System.arraycopy(fileContent, tenth * i, expected, 0, tenth);
            // Open the same file for read. Need to create new reader after every write operation(!)
            is = fileSystem.open(path);
            is.seek(tenth * i);
            int readBytes = is.read(toRead, 0, tenth);
            System.out.println("Has read " + readBytes);
            assertTrue("Should've get more bytes", (readBytes > 0) && (readBytes <= tenth));
            is.close();
            checkData(toRead, 0, readBytes, expected, "Partial verification");
        }
        System.out.println("Writing " + (tenth * SECTIONS) + " to " + (tenth * SECTIONS + rounding) + " section to file " + fileName);
        stm.write(fileContent, tenth * SECTIONS, rounding);
        stm.close();
        assertEquals("File size doesn't match ", AppendTestUtil.FILE_SIZE, fileSystem.getFileStatus(path).getLen());
        AppendTestUtil.checkFullFile(fileSystem, path, fileContent.length, fileContent, "hflush()");
    } finally {
        fileSystem.close();
        cluster.shutdown();
    }
}

Example 87 with LocatedBlocks

use of org.apache.hadoop.hdfs.protocol.LocatedBlocks in project hadoop by apache.

the class TestGetBlocks method testReadSelectNonStaleDatanode.

/**
   * Test if the datanodes returned by
   * {@link ClientProtocol#getBlockLocations(String, long, long)} is correct
   * when stale nodes checking is enabled. Also test during the scenario when 1)
   * stale nodes checking is enabled, 2) a writing is going on, 3) a datanode
   * becomes stale happen simultaneously
   * 
   * @throws Exception
   */
@Test
public void testReadSelectNonStaleDatanode() throws Exception {
    HdfsConfiguration conf = new HdfsConfiguration();
    conf.setBoolean(DFSConfigKeys.DFS_NAMENODE_AVOID_STALE_DATANODE_FOR_READ_KEY, true);
    long staleInterval = 30 * 1000 * 60;
    conf.setLong(DFSConfigKeys.DFS_NAMENODE_STALE_DATANODE_INTERVAL_KEY, staleInterval);
    MiniDFSCluster cluster = new MiniDFSCluster.Builder(conf).numDataNodes(numDatanodes).racks(racks).build();
    cluster.waitActive();
    InetSocketAddress addr = new InetSocketAddress("localhost", cluster.getNameNodePort());
    DFSClient client = new DFSClient(addr, conf);
    List<DatanodeDescriptor> nodeInfoList = cluster.getNameNode().getNamesystem().getBlockManager().getDatanodeManager().getDatanodeListForReport(DatanodeReportType.LIVE);
    assertEquals("Unexpected number of datanodes", numDatanodes, nodeInfoList.size());
    FileSystem fileSys = cluster.getFileSystem();
    FSDataOutputStream stm = null;
    try {
        // do the writing but do not close the FSDataOutputStream
        // in order to mimic the ongoing writing
        final Path fileName = new Path("/file1");
        stm = fileSys.create(fileName, true, fileSys.getConf().getInt(CommonConfigurationKeys.IO_FILE_BUFFER_SIZE_KEY, 4096), (short) 3, blockSize);
        stm.write(new byte[(blockSize * 3) / 2]);
        // We do not close the stream so that
        // the writing seems to be still ongoing
        stm.hflush();
        LocatedBlocks blocks = client.getNamenode().getBlockLocations(fileName.toString(), 0, blockSize);
        DatanodeInfo[] nodes = blocks.get(0).getLocations();
        assertEquals(nodes.length, 3);
        DataNode staleNode = null;
        DatanodeDescriptor staleNodeInfo = null;
        // stop the heartbeat of the first node
        staleNode = this.stopDataNodeHeartbeat(cluster, nodes[0].getHostName());
        assertNotNull(staleNode);
        // set the first node as stale
        staleNodeInfo = cluster.getNameNode().getNamesystem().getBlockManager().getDatanodeManager().getDatanode(staleNode.getDatanodeId());
        DFSTestUtil.resetLastUpdatesWithOffset(staleNodeInfo, -(staleInterval + 1));
        LocatedBlocks blocksAfterStale = client.getNamenode().getBlockLocations(fileName.toString(), 0, blockSize);
        DatanodeInfo[] nodesAfterStale = blocksAfterStale.get(0).getLocations();
        assertEquals(nodesAfterStale.length, 3);
        assertEquals(nodesAfterStale[2].getHostName(), nodes[0].getHostName());
        // restart the staleNode's heartbeat
        DataNodeTestUtils.setHeartbeatsDisabledForTests(staleNode, false);
        // reset the first node as non-stale, so as to avoid two stale nodes
        DFSTestUtil.resetLastUpdatesWithOffset(staleNodeInfo, 0);
        LocatedBlock lastBlock = client.getLocatedBlocks(fileName.toString(), 0, Long.MAX_VALUE).getLastLocatedBlock();
        nodes = lastBlock.getLocations();
        assertEquals(nodes.length, 3);
        // stop the heartbeat of the first node for the last block
        staleNode = this.stopDataNodeHeartbeat(cluster, nodes[0].getHostName());
        assertNotNull(staleNode);
        // set the node as stale
        DatanodeDescriptor dnDesc = cluster.getNameNode().getNamesystem().getBlockManager().getDatanodeManager().getDatanode(staleNode.getDatanodeId());
        DFSTestUtil.resetLastUpdatesWithOffset(dnDesc, -(staleInterval + 1));
        LocatedBlock lastBlockAfterStale = client.getLocatedBlocks(fileName.toString(), 0, Long.MAX_VALUE).getLastLocatedBlock();
        nodesAfterStale = lastBlockAfterStale.getLocations();
        assertEquals(nodesAfterStale.length, 3);
        assertEquals(nodesAfterStale[2].getHostName(), nodes[0].getHostName());
    } finally {
        if (stm != null) {
            stm.close();
        }
        client.close();
        cluster.shutdown();
    }
}

Example 88 with LocatedBlocks

use of org.apache.hadoop.hdfs.protocol.LocatedBlocks in project hadoop by apache.

the class TestReadStripedFileWithDecoding method corruptBlocks.

private void corruptBlocks(Path srcPath, int dataBlkDelNum, int parityBlkDelNum, boolean deleteBlockFile) throws IOException {
    int recoverBlkNum = dataBlkDelNum + parityBlkDelNum;
    LocatedBlocks locatedBlocks = getLocatedBlocks(srcPath);
    LocatedStripedBlock lastBlock = (LocatedStripedBlock) locatedBlocks.getLastLocatedBlock();
    int[] delDataBlkIndices = StripedFileTestUtil.randomArray(0, dataBlocks, dataBlkDelNum);
    Assert.assertNotNull(delDataBlkIndices);
    int[] delParityBlkIndices = StripedFileTestUtil.randomArray(dataBlocks, dataBlocks + parityBlocks, parityBlkDelNum);
    Assert.assertNotNull(delParityBlkIndices);
    int[] delBlkIndices = new int[recoverBlkNum];
    System.arraycopy(delDataBlkIndices, 0, delBlkIndices, 0, delDataBlkIndices.length);
    System.arraycopy(delParityBlkIndices, 0, delBlkIndices, delDataBlkIndices.length, delParityBlkIndices.length);
    ExtendedBlock[] delBlocks = new ExtendedBlock[recoverBlkNum];
    for (int i = 0; i < recoverBlkNum; i++) {
        delBlocks[i] = StripedBlockUtil.constructInternalBlock(lastBlock.getBlock(), cellSize, dataBlocks, delBlkIndices[i]);
        if (deleteBlockFile) {
            // delete the block file
            cluster.corruptBlockOnDataNodesByDeletingBlockFile(delBlocks[i]);
        } else {
            // corrupt the block file
            cluster.corruptBlockOnDataNodes(delBlocks[i]);
        }
    }
}

Example 89 with LocatedBlocks

use of org.apache.hadoop.hdfs.protocol.LocatedBlocks in project hadoop by apache.

the class TestReconstructStripedFile method assertFileBlocksReconstruction.

/**
   * Test the file blocks reconstruction.
   * 1. Check the replica is reconstructed in the target datanode,
   *    and verify the block replica length, generationStamp and content.
   * 2. Read the file and verify content.
   */
private void assertFileBlocksReconstruction(String fileName, int fileLen, ReconstructionType type, int toRecoverBlockNum) throws Exception {
    if (toRecoverBlockNum < 1 || toRecoverBlockNum > parityBlkNum) {
        Assert.fail("toRecoverBlockNum should be between 1 ~ " + parityBlkNum);
    }
    assertTrue("File length must be positive.", fileLen > 0);
    Path file = new Path(fileName);
    final byte[] data = new byte[fileLen];
    Arrays.fill(data, (byte) 1);
    DFSTestUtil.writeFile(fs, file, data);
    StripedFileTestUtil.waitBlockGroupsReported(fs, fileName);
    LocatedBlocks locatedBlocks = StripedFileTestUtil.getLocatedBlocks(file, fs);
    assertEquals(locatedBlocks.getFileLength(), fileLen);
    LocatedStripedBlock lastBlock = (LocatedStripedBlock) locatedBlocks.getLastLocatedBlock();
    DatanodeInfo[] storageInfos = lastBlock.getLocations();
    byte[] indices = lastBlock.getBlockIndices();
    BitSet bitset = new BitSet(dnNum);
    for (DatanodeInfo storageInfo : storageInfos) {
        bitset.set(dnMap.get(storageInfo));
    }
    int[] dead = generateDeadDnIndices(type, toRecoverBlockNum, indices);
    LOG.info("Note: indices == " + Arrays.toString(indices) + ". Generate errors on datanodes: " + Arrays.toString(dead));
    DatanodeInfo[] dataDNs = new DatanodeInfo[toRecoverBlockNum];
    int[] deadDnIndices = new int[toRecoverBlockNum];
    ExtendedBlock[] blocks = new ExtendedBlock[toRecoverBlockNum];
    File[] replicas = new File[toRecoverBlockNum];
    long[] replicaLengths = new long[toRecoverBlockNum];
    File[] metadatas = new File[toRecoverBlockNum];
    byte[][] replicaContents = new byte[toRecoverBlockNum][];
    Map<ExtendedBlock, DataNode> errorMap = new HashMap<>(dead.length);
    for (int i = 0; i < toRecoverBlockNum; i++) {
        dataDNs[i] = storageInfos[dead[i]];
        deadDnIndices[i] = dnMap.get(dataDNs[i]);
        // Check the block replica file on deadDn before it dead.
        blocks[i] = StripedBlockUtil.constructInternalBlock(lastBlock.getBlock(), cellSize, dataBlkNum, indices[dead[i]]);
        errorMap.put(blocks[i], cluster.getDataNodes().get(deadDnIndices[i]));
        replicas[i] = cluster.getBlockFile(deadDnIndices[i], blocks[i]);
        replicaLengths[i] = replicas[i].length();
        metadatas[i] = cluster.getBlockMetadataFile(deadDnIndices[i], blocks[i]);
        // the block replica on the datanode should be the same as expected
        assertEquals(replicaLengths[i], StripedBlockUtil.getInternalBlockLength(lastBlock.getBlockSize(), cellSize, dataBlkNum, indices[dead[i]]));
        assertTrue(metadatas[i].getName().endsWith(blocks[i].getGenerationStamp() + ".meta"));
        LOG.info("replica " + i + " locates in file: " + replicas[i]);
        replicaContents[i] = DFSTestUtil.readFileAsBytes(replicas[i]);
    }
    int lastGroupDataLen = fileLen % (dataBlkNum * blockSize);
    int lastGroupNumBlk = lastGroupDataLen == 0 ? dataBlkNum : Math.min(dataBlkNum, ((lastGroupDataLen - 1) / cellSize + 1));
    int groupSize = lastGroupNumBlk + parityBlkNum;
    // shutdown datanodes or generate corruption
    int stoppedDN = generateErrors(errorMap, type);
    // Check the locatedBlocks of the file again
    locatedBlocks = StripedFileTestUtil.getLocatedBlocks(file, fs);
    lastBlock = (LocatedStripedBlock) locatedBlocks.getLastLocatedBlock();
    storageInfos = lastBlock.getLocations();
    assertEquals(storageInfos.length, groupSize - stoppedDN);
    int[] targetDNs = new int[dnNum - groupSize];
    int n = 0;
    for (int i = 0; i < dnNum; i++) {
        if (!bitset.get(i)) {
            // not contain replica of the block.
            targetDNs[n++] = i;
        }
    }
    StripedFileTestUtil.waitForReconstructionFinished(file, fs, groupSize);
    targetDNs = sortTargetsByReplicas(blocks, targetDNs);
    // Check the replica on the new target node.
    for (int i = 0; i < toRecoverBlockNum; i++) {
        File replicaAfterReconstruction = cluster.getBlockFile(targetDNs[i], blocks[i]);
        LOG.info("replica after reconstruction " + replicaAfterReconstruction);
        File metadataAfterReconstruction = cluster.getBlockMetadataFile(targetDNs[i], blocks[i]);
        assertEquals(replicaLengths[i], replicaAfterReconstruction.length());
        LOG.info("replica before " + replicas[i]);
        assertTrue(metadataAfterReconstruction.getName().endsWith(blocks[i].getGenerationStamp() + ".meta"));
        byte[] replicaContentAfterReconstruction = DFSTestUtil.readFileAsBytes(replicaAfterReconstruction);
        Assert.assertArrayEquals(replicaContents[i], replicaContentAfterReconstruction);
    }
}

Example 90 with LocatedBlocks

use of org.apache.hadoop.hdfs.protocol.LocatedBlocks in project hadoop by apache.

the class TestReplication method testBadBlockReportOnTransfer.

private void testBadBlockReportOnTransfer(boolean corruptBlockByDeletingBlockFile) throws Exception {
    Configuration conf = new HdfsConfiguration();
    FileSystem fs = null;
    DFSClient dfsClient = null;
    LocatedBlocks blocks = null;
    int replicaCount = 0;
    short replFactor = 1;
    MiniDFSCluster cluster = new MiniDFSCluster.Builder(conf).numDataNodes(2).build();
    cluster.waitActive();
    fs = cluster.getFileSystem();
    dfsClient = new DFSClient(new InetSocketAddress("localhost", cluster.getNameNodePort()), conf);
    // Create file with replication factor of 1
    Path file1 = new Path("/tmp/testBadBlockReportOnTransfer/file1");
    DFSTestUtil.createFile(fs, file1, 1024, replFactor, 0);
    DFSTestUtil.waitReplication(fs, file1, replFactor);
    // Corrupt the block belonging to the created file
    ExtendedBlock block = DFSTestUtil.getFirstBlock(fs, file1);
    int blockFilesCorrupted = corruptBlockByDeletingBlockFile ? cluster.corruptBlockOnDataNodesByDeletingBlockFile(block) : cluster.corruptBlockOnDataNodes(block);
    assertEquals("Corrupted too few blocks", replFactor, blockFilesCorrupted);
    // Increase replication factor, this should invoke transfer request
    // Receiving datanode fails on checksum and reports it to namenode
    replFactor = 2;
    fs.setReplication(file1, replFactor);
    // Now get block details and check if the block is corrupt
    blocks = dfsClient.getNamenode().getBlockLocations(file1.toString(), 0, Long.MAX_VALUE);
    while (blocks.get(0).isCorrupt() != true) {
        try {
            LOG.info("Waiting until block is marked as corrupt...");
            Thread.sleep(1000);
        } catch (InterruptedException ie) {
        }
        blocks = dfsClient.getNamenode().getBlockLocations(file1.toString(), 0, Long.MAX_VALUE);
    }
    replicaCount = blocks.get(0).getLocations().length;
    assertTrue(replicaCount == 1);
    cluster.shutdown();
}