Examples with TaskContext com.tencent.angel.worker.task.TaskContext used on opensource projects

Example 1 with TaskContext

use of com.tencent.angel.worker.task.TaskContext 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 2 with TaskContext

use of com.tencent.angel.worker.task.TaskContext in project angel by Tencent.

the class PeriodHATest method testHA.

@Test
public void testHA() 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");
    ps2Attempt0.failed("exit");
    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 3 with TaskContext

use of com.tencent.angel.worker.task.TaskContext in project angel by Tencent.

the class MasterRecoverTest method testMasterRecover.

@SuppressWarnings("unchecked")
@Test
public void testMasterRecover() throws Exception {
    try {
        ApplicationAttemptId appAttempt1Id = ApplicationAttemptId.newInstance(LocalClusterContext.get().getAppId(), 1);
        ApplicationAttemptId appAttempt2Id = ApplicationAttemptId.newInstance(LocalClusterContext.get().getAppId(), 2);
        AngelApplicationMaster angelAppMaster = LocalClusterContext.get().getMaster().getAppMaster();
        LOG.info("angelAppMaster.getAppContext().getApplicationAttemptId()=" + angelAppMaster.getAppContext().getApplicationAttemptId());
        assertEquals(angelAppMaster.getAppContext().getApplicationAttemptId(), appAttempt1Id);
        ParameterServerManager psManager = angelAppMaster.getAppContext().getParameterServerManager();
        ParameterServer ps = LocalClusterContext.get().getPS(psAttempt0Id).getPS();
        int w1Id = angelAppMaster.getAppContext().getMatrixMetaManager().getMatrix("w1").getId();
        int w2Id = angelAppMaster.getAppContext().getMatrixMetaManager().getMatrix("w2").getId();
        Location masterLoc = LocalClusterContext.get().getMaster().getAppMaster().getAppContext().getMasterService().getLocation();
        TConnection connection = TConnectionManager.getConnection(ps.getConf());
        MasterProtocol master = connection.getMasterService(masterLoc.getIp(), masterLoc.getPort());
        int task0Iteration = 2;
        int task1Iteration = 1;
        int task0w1Clock = 10;
        int task0w2Clock = 20;
        int task1w1Clock = 9;
        int task1w2Clock = 19;
        int w1Clock = (task0w1Clock < task1w1Clock) ? task0w1Clock : task1w1Clock;
        int w2Clock = (task0w2Clock < task1w2Clock) ? task0w2Clock : task1w2Clock;
        master.taskIteration(null, TaskIterationRequest.newBuilder().setIteration(task0Iteration).setTaskId(ProtobufUtil.convertToIdProto(task0Id)).build());
        master.taskIteration(null, TaskIterationRequest.newBuilder().setIteration(task1Iteration).setTaskId(ProtobufUtil.convertToIdProto(task1Id)).build());
        master.taskClock(null, TaskClockRequest.newBuilder().setTaskId(ProtobufUtil.convertToIdProto(task0Id)).setMatrixClock(MatrixClock.newBuilder().setMatrixId(w1Id).setClock(task0w1Clock).build()).build());
        master.taskClock(null, TaskClockRequest.newBuilder().setTaskId(ProtobufUtil.convertToIdProto(task0Id)).setMatrixClock(MatrixClock.newBuilder().setMatrixId(w2Id).setClock(task0w2Clock).build()).build());
        master.taskClock(null, TaskClockRequest.newBuilder().setTaskId(ProtobufUtil.convertToIdProto(task1Id)).setMatrixClock(MatrixClock.newBuilder().setMatrixId(w1Id).setClock(task1w1Clock).build()).build());
        master.taskClock(null, TaskClockRequest.newBuilder().setTaskId(ProtobufUtil.convertToIdProto(task1Id)).setMatrixClock(MatrixClock.newBuilder().setMatrixId(w2Id).setClock(task1w2Clock).build()).build());
        int writeIntervalMS = LocalClusterContext.get().getConf().getInt(AngelConf.ANGEL_AM_WRITE_STATE_INTERVAL_MS, AngelConf.DEFAULT_ANGEL_AM_WRITE_STATE_INTERVAL_MS);
        Thread.sleep(writeIntervalMS * 2);
        angelAppMaster.getAppContext().getEventHandler().handle(new InternalErrorEvent(angelAppMaster.getAppContext().getApplicationId(), "failed", true));
        Thread.sleep(10000);
        angelAppMaster = LocalClusterContext.get().getMaster().getAppMaster();
        assertEquals(angelAppMaster.getAppContext().getApp().getExternAppState(), AppState.RUNNING);
        LOG.info("angelAppMaster.getAppContext().getApplicationAttemptId()=" + angelAppMaster.getAppContext().getApplicationAttemptId());
        assertEquals(angelAppMaster.getAppContext().getApplicationAttemptId(), appAttempt2Id);
        PartitionKey w1Part0Key = new PartitionKey(0, w1Id, 0, 0, 1, 50000);
        PartitionKey w1Part1Key = new PartitionKey(1, w1Id, 0, 50000, 1, 100000);
        PartitionKey w2Part0Key = new PartitionKey(0, w2Id, 0, 0, 1, 50000);
        PartitionKey w2Part1Key = new PartitionKey(1, w2Id, 0, 50000, 1, 100000);
        Worker worker = LocalClusterContext.get().getWorker(worker0Attempt0Id).getWorker();
        LOG.info("worker=" + worker);
        LOG.info("worker.getTaskManager()=" + worker.getTaskManager());
        LOG.info("worker.getTaskManager().getRunningTask()=" + worker.getTaskManager().getRunningTask().size());
        TaskContext task0Context = worker.getTaskManager().getRunningTask().get(task0Id).getTaskContext();
        TaskContext task1Context = worker.getTaskManager().getRunningTask().get(task1Id).getTaskContext();
        assertEquals(task0Context.getEpoch(), task0Iteration);
        assertEquals(task1Context.getEpoch(), task1Iteration);
        LOG.info("===============worker.getPSAgent().getMatrixMetaManager().getMatrixMetas().size()=" + worker.getPSAgent().getMatrixMetaManager().getMatrixMetas().size());
        assertTrue(worker.getPSAgent().getMatrixMetaManager().exist(w1Id));
        assertTrue(worker.getPSAgent().getMatrixMetaManager().exist(w2Id));
        assertEquals(worker.getPSAgent().getMatrixMetaManager().getPss(w1Part0Key).get(0), psId);
        assertEquals(worker.getPSAgent().getMatrixMetaManager().getPss(w1Part1Key).get(0), psId);
        assertEquals(worker.getPSAgent().getMatrixMetaManager().getPss(w2Part0Key).get(0), psId);
        assertEquals(worker.getPSAgent().getMatrixMetaManager().getPss(w2Part1Key).get(0), psId);
        ps = LocalClusterContext.get().getPS(psAttempt0Id).getPS();
        checkMatrixInfo(ps, w1Id, w2Id, w1Clock, w2Clock);
        angelAppMaster.getAppContext().getEventHandler().handle(new InternalErrorEvent(angelAppMaster.getAppContext().getApplicationId(), "failed", true));
        Thread.sleep(10000);
        angelAppMaster = LocalClusterContext.get().getMaster().getAppMaster();
        assertEquals(angelAppMaster.getAppContext().getApplicationAttemptId(), appAttempt2Id);
        assertEquals(angelAppMaster.getAppContext().getApp().getExternAppState(), AppState.FAILED);
    } catch (Exception x) {
        LOG.error("run testMasterRecover failed ", x);
        throw x;
    }
}