Examples with PSAttemptId com.tencent.angel.ps.PSAttemptId used on opensource projects

Example 11 with PSAttemptId

use of com.tencent.angel.ps.PSAttemptId in project angel by Tencent.

the class YarnContainerAllocator method handleFinishContainers.

@SuppressWarnings("unchecked")
private void handleFinishContainers(List<ContainerStatus> finishedContainers) {
    for (ContainerStatus cont : finishedContainers) {
        LOG.info("Received completed container:" + cont);
        Id id = assignedContainerToIDMap.get(cont.getContainerId());
        if (id == null) {
            LOG.error("Container complete event for unknown container id " + cont.getContainerId());
        } else {
            assignedContainerToIDMap.remove(cont.getContainerId());
            idToContainerMap.remove(id);
            // dispatch container exit message to corresponding components
            String diagnostics = StringInterner.weakIntern(cont.getDiagnostics());
            if (id instanceof PSAttemptId) {
                context.getEventHandler().handle(new PSAttemptDiagnosticsUpdateEvent(diagnostics, (PSAttemptId) id));
                context.getEventHandler().handle(createContainerFinishedEvent(cont, (PSAttemptId) id));
            } else if (id instanceof PSAgentAttemptId) {
                context.getEventHandler().handle(new PSAgentAttemptDiagnosticsUpdateEvent((PSAgentAttemptId) id, diagnostics));
                context.getEventHandler().handle(createContainerFinishedEvent(cont, (PSAgentAttemptId) id));
            } else if (id instanceof WorkerAttemptId) {
                context.getEventHandler().handle(new WorkerAttemptDiagnosticsUpdateEvent((WorkerAttemptId) id, diagnostics));
                context.getEventHandler().handle(createContainerFinishedEvent(cont, (WorkerAttemptId) id));
            }
        }
    }
}

Example 12 with PSAttemptId

use of com.tencent.angel.ps.PSAttemptId in project angel by Tencent.

the class AppStateStorage method writePSMeta.

/**
 * write ps meta to file
 * @param psManager ps meta storage
 * @throws IOException
 */
public void writePSMeta(ParameterServerManager psManager) throws IOException {
    try {
        psMetaLock.lock();
        // generate a temporary file
        String psMetaFile = getPsMetaFile();
        String tmpFile = getPSMetaTmpeFile(psMetaFile);
        Path tmpPath = new Path(writeDir, tmpFile);
        FSDataOutputStream outputStream = fs.create(tmpPath);
        // write ps meta to the temporary file first.
        Map<ParameterServerId, AMParameterServer> psMap = psManager.getParameterServerMap();
        outputStream.writeInt(psMap.size());
        PSAttemptId attemptId = null;
        int nextAttemptIndex = 0;
        for (Entry<ParameterServerId, AMParameterServer> entry : psMap.entrySet()) {
            outputStream.writeInt(entry.getKey().getIndex());
            attemptId = entry.getValue().getRunningAttemptId();
            nextAttemptIndex = entry.getValue().getNextAttemptNumber();
            if (attemptId != null) {
                nextAttemptIndex = attemptId.getIndex();
            }
            outputStream.writeInt(nextAttemptIndex);
        }
        outputStream.close();
        // rename the temporary file to the final file
        Path psMetaFilePath = new Path(writeDir, psMetaFile);
        HdfsUtil.rename(tmpPath, psMetaFilePath, fs);
        // if the old final file exist, just remove it
        if (lastPsMetaFilePath != null) {
            fs.delete(lastPsMetaFilePath, false);
        }
        lastPsMetaFilePath = psMetaFilePath;
    } finally {
        psMetaLock.unlock();
    }
}

Example 13 with PSAttemptId

use of com.tencent.angel.ps.PSAttemptId in project angel by Tencent.

the class MasterServiceTest method setup.

@Before
public void setup() throws Exception {
    try {
        // set basic configuration keys
        Configuration conf = new Configuration();
        conf.setBoolean("mapred.mapper.new-api", true);
        conf.setBoolean(AngelConf.ANGEL_JOB_OUTPUT_PATH_DELETEONEXIST, true);
        conf.set(AngelConf.ANGEL_TASK_USER_TASKCLASS, DummyTask.class.getName());
        // use local deploy mode and dummy dataspliter
        conf.set(AngelConf.ANGEL_DEPLOY_MODE, "LOCAL");
        conf.setBoolean(AngelConf.ANGEL_AM_USE_DUMMY_DATASPLITER, true);
        conf.set(AngelConf.ANGEL_INPUTFORMAT_CLASS, CombineTextInputFormat.class.getName());
        conf.set(AngelConf.ANGEL_SAVE_MODEL_PATH, LOCAL_FS + TMP_PATH + "/out");
        conf.set(AngelConf.ANGEL_TRAIN_DATA_PATH, LOCAL_FS + TMP_PATH + "/in");
        conf.set(AngelConf.ANGEL_LOG_PATH, LOCAL_FS + TMP_PATH + "/log");
        conf.setInt(AngelConf.ANGEL_WORKERGROUP_NUMBER, 1);
        conf.setInt(AngelConf.ANGEL_PS_NUMBER, 1);
        conf.setInt(AngelConf.ANGEL_WORKER_TASK_NUMBER, 2);
        // get a angel client
        angelClient = AngelClientFactory.get(conf);
        // add matrix
        MatrixContext mMatrix = new MatrixContext();
        mMatrix.setName("w1");
        mMatrix.setRowNum(1);
        mMatrix.setColNum(100000);
        mMatrix.setMaxRowNumInBlock(1);
        mMatrix.setMaxColNumInBlock(50000);
        mMatrix.setRowType(RowType.T_INT_DENSE);
        mMatrix.set(MatrixConf.MATRIX_OPLOG_ENABLEFILTER, "false");
        mMatrix.set(MatrixConf.MATRIX_HOGWILD, "true");
        mMatrix.set(MatrixConf.MATRIX_AVERAGE, "false");
        mMatrix.set(MatrixConf.MATRIX_OPLOG_TYPE, "DENSE_INT");
        angelClient.addMatrix(mMatrix);
        MatrixContext mMatrix2 = new MatrixContext();
        mMatrix2.setName("w2");
        mMatrix2.setRowNum(1);
        mMatrix2.setColNum(100000);
        mMatrix2.setMaxRowNumInBlock(1);
        mMatrix2.setMaxColNumInBlock(50000);
        mMatrix2.setRowType(RowType.T_DOUBLE_DENSE);
        mMatrix2.set(MatrixConf.MATRIX_OPLOG_ENABLEFILTER, "false");
        mMatrix2.set(MatrixConf.MATRIX_HOGWILD, "false");
        mMatrix2.set(MatrixConf.MATRIX_AVERAGE, "false");
        mMatrix2.set(MatrixConf.MATRIX_OPLOG_TYPE, "DENSE_DOUBLE");
        angelClient.addMatrix(mMatrix2);
        angelClient.startPSServer();
        angelClient.run();
        Thread.sleep(5000);
        group0Id = new WorkerGroupId(0);
        worker0Id = new WorkerId(group0Id, 0);
        worker0Attempt0Id = new WorkerAttemptId(worker0Id, 0);
        task0Id = new TaskId(0);
        task1Id = new TaskId(1);
        psId = new ParameterServerId(0);
        psAttempt0Id = new PSAttemptId(psId, 0);
    } catch (Exception x) {
        LOG.error("setup failed ", x);
        throw x;
    }
}

Example 14 with PSAttemptId

use of com.tencent.angel.ps.PSAttemptId in project angel by Tencent.

the class PSFailedReportTest method testPSFailedReport.

@Test
public void testPSFailedReport() throws Exception {
    ParameterServerId ps1Id = new ParameterServerId(0);
    final ParameterServerId ps2Id = new ParameterServerId(1);
    PSAttemptId ps1Attempt0Id = new PSAttemptId(ps1Id, 0);
    PSAttemptId ps2Attempt0Id = new PSAttemptId(ps2Id, 0);
    PSAttemptId ps2Attempt1Id = new PSAttemptId(ps2Id, 1);
    ParameterServer ps1Attempt0 = LocalClusterContext.get().getPS(ps1Attempt0Id).getPS();
    ParameterServer ps2Attempt0 = LocalClusterContext.get().getPS(ps2Attempt0Id).getPS();
    WorkerId worker0Id = new WorkerId(new WorkerGroupId(0), 0);
    WorkerAttemptId worker0Attempt0Id = new WorkerAttemptId(worker0Id, 0);
    Worker worker0 = LocalClusterContext.get().getWorker(worker0Attempt0Id).getWorker();
    TaskContext task0Context = worker0.getTaskManager().getRunningTask().get(task0Id).getTaskContext();
    MatrixClient matrixClient = task0Context.getMatrix("w1");
    int iterNum = 20;
    for (int i = 0; i < iterNum; i++) {
        DenseIntVector update = new DenseIntVector(dim);
        for (int j = 0; j < dim; j++) {
            update.set(j, 1);
        }
        update.setMatrixId(matrixClient.getMatrixId());
        update.setRowId(0);
        matrixClient.increment(update);
        matrixClient.clock().get();
        Thread.sleep(1000);
        MatrixStorageManager ps1Storage = ps1Attempt0.getMatrixStorageManager();
        ServerMatrix ps1w1 = ps1Storage.getMatrix(matrixClient.getMatrixId());
        assertNotNull(ps1w1.getPartition(0));
        assertNotNull(ps1w1.getPartition(1));
        IntBuffer row0Part0 = ((ServerDenseIntRow) ps1w1.getRow(0, 0)).getData();
        int part0Size = ps1w1.getRow(0, 0).size();
        IntBuffer row0Part1 = ((ServerDenseIntRow) ps1w1.getRow(1, 0)).getData();
        int part1Size = ps1w1.getRow(1, 0).size();
        assertEquals(sum(row0Part0, part0Size), (i + 1) * dim / 2);
        assertEquals(sum(row0Part1, part1Size), (i + 1) * dim / 2);
        MatrixStorageManager ps2Storage = ps2Attempt0.getMatrixStorageManager();
        ServerMatrix ps2w1 = ps2Storage.getMatrix(matrixClient.getMatrixId());
        assertNotNull(ps2w1.getPartition(0));
        assertNotNull(ps2w1.getPartition(1));
        row0Part0 = ((ServerDenseIntRow) ps2w1.getRow(0, 0)).getData();
        part0Size = ps2w1.getRow(0, 0).size();
        row0Part1 = ((ServerDenseIntRow) ps2w1.getRow(1, 0)).getData();
        part1Size = ps2w1.getRow(1, 0).size();
        assertEquals(sum(row0Part0, part0Size), (i + 1) * dim / 2);
        assertEquals(sum(row0Part1, part1Size), (i + 1) * dim / 2);
    }
    LOG.info("===================================================================ps2 failed");
    HashMap<PSLocation, Integer> failedCounters = new HashMap<>();
    PSLocation psLoc = new PSLocation(ps2Id, ps2Attempt0.getLocationManager().getPsLocation(ps2Id));
    failedCounters.put(psLoc, 10000);
    worker0.getPSAgent().getMasterClient().psFailedReport(failedCounters);
    Thread.sleep(20000);
    for (int i = iterNum; i < 2 * iterNum; i++) {
        DenseIntVector update = new DenseIntVector(dim);
        for (int j = 0; j < dim; j++) {
            update.set(j, 1);
        }
        update.setMatrixId(matrixClient.getMatrixId());
        update.setRowId(0);
        matrixClient.increment(update);
        matrixClient.clock().get();
        Thread.sleep(1000);
        MatrixStorageManager ps1Storage = ps1Attempt0.getMatrixStorageManager();
        ServerMatrix ps1w1 = ps1Storage.getMatrix(matrixClient.getMatrixId());
        assertNotNull(ps1w1.getPartition(0));
        assertNotNull(ps1w1.getPartition(1));
        IntBuffer row0Part0 = ((ServerDenseIntRow) ps1w1.getRow(0, 0)).getData();
        int part0Size = ps1w1.getRow(0, 0).size();
        IntBuffer row0Part1 = ((ServerDenseIntRow) ps1w1.getRow(1, 0)).getData();
        int part1Size = ps1w1.getRow(1, 0).size();
        assertEquals(sum(row0Part0, part0Size), (i + 1) * dim / 2);
        assertEquals(sum(row0Part1, part1Size), (i + 1) * dim / 2);
    }
    ParameterServer ps2Attempt = LocalClusterContext.get().getPS(ps2Attempt1Id).getPS();
    for (int i = iterNum * 2; i < 3 * iterNum; i++) {
        DenseIntVector update = new DenseIntVector(dim);
        for (int j = 0; j < dim; j++) {
            update.set(j, 1);
        }
        update.setMatrixId(matrixClient.getMatrixId());
        update.setRowId(0);
        matrixClient.increment(update);
        matrixClient.clock().get();
        Thread.sleep(1000);
        MatrixStorageManager ps1Storage = ps1Attempt0.getMatrixStorageManager();
        ServerMatrix ps1w1 = ps1Storage.getMatrix(matrixClient.getMatrixId());
        assertNotNull(ps1w1.getPartition(0));
        assertNotNull(ps1w1.getPartition(1));
        IntBuffer row0Part0 = ((ServerDenseIntRow) ps1w1.getRow(0, 0)).getData();
        int part0Size = ps1w1.getRow(0, 0).size();
        IntBuffer row0Part1 = ((ServerDenseIntRow) ps1w1.getRow(1, 0)).getData();
        int part1Size = ps1w1.getRow(1, 0).size();
        assertEquals(sum(row0Part0, part0Size), (i + 1) * dim / 2);
        assertEquals(sum(row0Part1, part1Size), (i + 1) * dim / 2);
        MatrixStorageManager ps2Storage = ps2Attempt.getMatrixStorageManager();
        ServerMatrix ps2w1 = ps2Storage.getMatrix(matrixClient.getMatrixId());
        assertNotNull(ps2w1.getPartition(0));
        assertNotNull(ps2w1.getPartition(1));
        row0Part0 = ((ServerDenseIntRow) ps2w1.getRow(0, 0)).getData();
        part0Size = ps2w1.getRow(0, 0).size();
        row0Part1 = ((ServerDenseIntRow) ps2w1.getRow(1, 0)).getData();
        part1Size = ps2w1.getRow(1, 0).size();
        assertEquals(sum(row0Part0, part0Size), (i + 1) * dim / 2);
        assertEquals(sum(row0Part1, part1Size), (i + 1) * dim / 2);
    }
}

Example 15 with PSAttemptId

use of com.tencent.angel.ps.PSAttemptId in project angel by Tencent.

the class PSManagerTest method setup.

@Before
public void setup() throws Exception {
    try {
        // set basic configuration keys
        Configuration conf = new Configuration();
        conf.setBoolean("mapred.mapper.new-api", true);
        conf.setBoolean(AngelConf.ANGEL_JOB_OUTPUT_PATH_DELETEONEXIST, true);
        conf.set(AngelConf.ANGEL_TASK_USER_TASKCLASS, DummyTask.class.getName());
        // use local deploy mode and dummy dataspliter
        conf.set(AngelConf.ANGEL_DEPLOY_MODE, "LOCAL");
        conf.setBoolean(AngelConf.ANGEL_AM_USE_DUMMY_DATASPLITER, true);
        conf.set(AngelConf.ANGEL_INPUTFORMAT_CLASS, CombineTextInputFormat.class.getName());
        conf.set(AngelConf.ANGEL_SAVE_MODEL_PATH, LOCAL_FS + TMP_PATH + "/out");
        conf.set(AngelConf.ANGEL_TRAIN_DATA_PATH, LOCAL_FS + TMP_PATH + "/in");
        conf.set(AngelConf.ANGEL_LOG_PATH, LOCAL_FS + TMP_PATH + "/log");
        conf.setInt(AngelConf.ANGEL_WORKERGROUP_NUMBER, 1);
        conf.setInt(AngelConf.ANGEL_PS_NUMBER, 1);
        conf.setInt(AngelConf.ANGEL_WORKER_TASK_NUMBER, 2);
        conf.setInt(AngelConf.ANGEL_PS_BACKUP_INTERVAL_MS, 5000);
        // get a angel client
        angelClient = AngelClientFactory.get(conf);
        // add matrix
        MatrixContext mMatrix = new MatrixContext();
        mMatrix.setName("w1");
        mMatrix.setRowNum(1);
        mMatrix.setColNum(100000);
        mMatrix.setMaxRowNumInBlock(1);
        mMatrix.setMaxColNumInBlock(50000);
        mMatrix.setRowType(RowType.T_INT_DENSE);
        mMatrix.set(MatrixConf.MATRIX_OPLOG_ENABLEFILTER, "false");
        mMatrix.set(MatrixConf.MATRIX_HOGWILD, "true");
        mMatrix.set(MatrixConf.MATRIX_AVERAGE, "false");
        mMatrix.set(MatrixConf.MATRIX_OPLOG_TYPE, "DENSE_INT");
        angelClient.addMatrix(mMatrix);
        MatrixContext mMatrix2 = new MatrixContext();
        mMatrix2.setName("w2");
        mMatrix2.setRowNum(1);
        mMatrix2.setColNum(100000);
        mMatrix2.setMaxRowNumInBlock(1);
        mMatrix2.setMaxColNumInBlock(50000);
        mMatrix2.setRowType(RowType.T_DOUBLE_DENSE);
        mMatrix2.set(MatrixConf.MATRIX_OPLOG_ENABLEFILTER, "false");
        mMatrix2.set(MatrixConf.MATRIX_HOGWILD, "false");
        mMatrix2.set(MatrixConf.MATRIX_AVERAGE, "false");
        mMatrix2.set(MatrixConf.MATRIX_OPLOG_TYPE, "DENSE_DOUBLE");
        angelClient.addMatrix(mMatrix2);
        angelClient.startPSServer();
        angelClient.run();
        Thread.sleep(5000);
        group0Id = new WorkerGroupId(0);
        worker0Id = new WorkerId(group0Id, 0);
        worker0Attempt0Id = new WorkerAttemptId(worker0Id, 0);
        task0Id = new TaskId(0);
        task1Id = new TaskId(1);
        psId = new ParameterServerId(0);
        psAttempt0Id = new PSAttemptId(psId, 0);
    } catch (Exception x) {
        LOG.error("setup failed ", x);
        throw x;
    }
}