Search in sources :

Example 6 with WorkerAttemptId

use of com.tencent.angel.worker.WorkerAttemptId in project angel by Tencent.

the class LocalContainerLauncher method launch.

@SuppressWarnings("unchecked")
private void launch(ContainerLauncherEvent event) {
    Id id = event.getId();
    if (id instanceof PSAttemptId) {
        LocalPS ps = new LocalPS((PSAttemptId) id, context.getMasterService().getLocation(), context.getConf());
        context.getEventHandler().handle(new PSAttemptEvent(PSAttemptEventType.PA_CONTAINER_LAUNCHED, (PSAttemptId) id));
        try {
            ps.start();
            LocalClusterContext.get().addPS((PSAttemptId) id, ps);
        } catch (Exception e) {
            LOG.error("launch ps failed.", e);
            context.getEventHandler().handle(new PSAttemptEvent(PSAttemptEventType.PA_CONTAINER_LAUNCH_FAILED, (PSAttemptId) id));
        }
    } else if (id instanceof WorkerAttemptId) {
        LocalWorker worker = new LocalWorker(context.getConf(), context.getApplicationId(), context.getUser(), (WorkerAttemptId) id, context.getMasterService().getLocation(), 0, false);
        context.getEventHandler().handle(new WorkerAttemptEvent(WorkerAttemptEventType.CONTAINER_LAUNCHED, (WorkerAttemptId) id));
        try {
            worker.start();
            LocalClusterContext.get().addWorker((WorkerAttemptId) id, worker);
        } catch (Exception e) {
            LOG.error("launch worker failed.", e);
            context.getEventHandler().handle(new WorkerAttemptEvent(WorkerAttemptEventType.CONTAINER_LAUNCH_FAILED, (WorkerAttemptId) id));
        }
    }
}
Also used : PSAttemptId(com.tencent.angel.ps.PSAttemptId) WorkerAttemptId(com.tencent.angel.worker.WorkerAttemptId) LocalWorker(com.tencent.angel.localcluster.LocalWorker) LocalPS(com.tencent.angel.localcluster.LocalPS) WorkerAttemptEvent(com.tencent.angel.master.worker.attempt.WorkerAttemptEvent) PSAttemptId(com.tencent.angel.ps.PSAttemptId) WorkerAttemptId(com.tencent.angel.worker.WorkerAttemptId) Id(com.tencent.angel.common.Id) PSAttemptEvent(com.tencent.angel.master.ps.attempt.PSAttemptEvent)

Example 7 with WorkerAttemptId

use of com.tencent.angel.worker.WorkerAttemptId in project angel by Tencent.

the class WorkerManager method checkHBTimeOut.

public void checkHBTimeOut() {
    // check whether worker heartbeat timeout
    Iterator<Map.Entry<WorkerAttemptId, Long>> workerIt = workerLastHeartbeatTS.entrySet().iterator();
    long currentTs = System.currentTimeMillis();
    while (workerIt.hasNext()) {
        Entry<WorkerAttemptId, Long> workerEntry = workerIt.next();
        if (currentTs - workerEntry.getValue() > workerTimeOutMS) {
            LOG.error(workerEntry.getKey() + " heartbeat timeout!!!");
            context.getEventHandler().handle(new WorkerAttemptDiagnosticsUpdateEvent(workerEntry.getKey(), "heartbeat timeout"));
            context.getEventHandler().handle(new WorkerAttemptEvent(WorkerAttemptEventType.ERROR, workerEntry.getKey()));
            workerIt.remove();
        }
    }
}
Also used : Entry(java.util.Map.Entry) WorkerAttemptId(com.tencent.angel.worker.WorkerAttemptId) WorkerAttemptDiagnosticsUpdateEvent(com.tencent.angel.master.worker.attempt.WorkerAttemptDiagnosticsUpdateEvent) WorkerAttemptEvent(com.tencent.angel.master.worker.attempt.WorkerAttemptEvent)

Example 8 with WorkerAttemptId

use of com.tencent.angel.worker.WorkerAttemptId in project angel by Tencent.

the class IndexGetRowsHashTest method setup.

@Before
public void setup() throws Exception {
    // 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, 2);
    conf.setInt(AngelConf.ANGEL_WORKER_TASK_NUMBER, 1);
    conf.setInt(AngelConf.ANGEL_MODEL_PARTITIONER_PARTITION_SIZE, 1000);
    conf.setBoolean("use.new.split", true);
    conf.setInt(AngelConf.ANGEL_PSAGENT_CACHE_SYNC_TIMEINTERVAL_MS, 10);
    conf.setInt(AngelConf.ANGEL_WORKER_HEARTBEAT_INTERVAL_MS, 1000);
    conf.setInt(AngelConf.ANGEL_PS_HEARTBEAT_INTERVAL_MS, 1000);
    conf.setInt(AngelConf.ANGEL_WORKER_MAX_ATTEMPTS, 1);
    conf.setInt(AngelConf.ANGEL_PS_MAX_ATTEMPTS, 1);
    // get a angel client
    angelClient = AngelClientFactory.get(conf);
    // add sparse double matrix
    MatrixContext sMat = new MatrixContext();
    sMat.setName(SPARSE_DOUBLE_MAT);
    sMat.setRowNum(rowNum);
    sMat.setRowType(RowType.T_DOUBLE_SPARSE);
    sMat.setPartitionNum(partNum);
    sMat.setPartitionerClass(HashPartitioner.class);
    angelClient.addMatrix(sMat);
    // add sparse float matrix
    MatrixContext sfMat = new MatrixContext();
    sfMat.setName(SPARSE_FLOAT_MAT);
    sfMat.setRowNum(rowNum);
    sfMat.setPartitionNum(partNum);
    sfMat.setPartitionerClass(HashPartitioner.class);
    sfMat.setRowType(RowType.T_FLOAT_SPARSE);
    angelClient.addMatrix(sfMat);
    // add sparse float matrix
    MatrixContext siMat = new MatrixContext();
    siMat.setName(SPARSE_INT_MAT);
    siMat.setRowNum(rowNum);
    siMat.setPartitionNum(partNum);
    siMat.setPartitionerClass(HashPartitioner.class);
    siMat.setRowType(RowType.T_INT_SPARSE);
    angelClient.addMatrix(siMat);
    // add sparse long matrix
    MatrixContext slMat = new MatrixContext();
    slMat.setName(SPARSE_LONG_MAT);
    slMat.setRowNum(rowNum);
    slMat.setPartitionNum(partNum);
    slMat.setPartitionerClass(HashPartitioner.class);
    slMat.setRowType(RowType.T_LONG_SPARSE);
    angelClient.addMatrix(slMat);
    // add sparse long-key double matrix
    MatrixContext dLongKeysMatrix = new MatrixContext();
    dLongKeysMatrix.setName(SPARSE_DOUBLE_LONG_MAT);
    dLongKeysMatrix.setRowNum(rowNum);
    dLongKeysMatrix.setPartitionNum(partNum);
    dLongKeysMatrix.setPartitionerClass(HashPartitioner.class);
    dLongKeysMatrix.setRowType(RowType.T_DOUBLE_SPARSE_LONGKEY);
    angelClient.addMatrix(dLongKeysMatrix);
    // add sparse long-key float matrix
    MatrixContext slfMatrix = new MatrixContext();
    slfMatrix.setName(SPARSE_FLOAT_LONG_MAT);
    slfMatrix.setRowNum(rowNum);
    slfMatrix.setPartitionNum(partNum);
    slfMatrix.setPartitionerClass(HashPartitioner.class);
    slfMatrix.setRowType(RowType.T_FLOAT_SPARSE_LONGKEY);
    angelClient.addMatrix(slfMatrix);
    // add sparse long-key int matrix
    MatrixContext sliMatrix = new MatrixContext();
    sliMatrix.setName(SPARSE_INT_LONG_MAT);
    sliMatrix.setRowNum(rowNum);
    sliMatrix.setPartitionNum(partNum);
    sliMatrix.setPartitionerClass(HashPartitioner.class);
    sliMatrix.setRowType(RowType.T_INT_SPARSE_LONGKEY);
    angelClient.addMatrix(sliMatrix);
    // add sparse long-key long matrix
    MatrixContext sllMatrix = new MatrixContext();
    sllMatrix.setName(SPARSE_LONG_LONG_MAT);
    sllMatrix.setRowNum(rowNum);
    sllMatrix.setPartitionNum(partNum);
    sllMatrix.setPartitionerClass(HashPartitioner.class);
    sllMatrix.setRowType(RowType.T_LONG_SPARSE_LONGKEY);
    angelClient.addMatrix(sllMatrix);
    // Start PS
    angelClient.startPSServer();
    // Start to run application
    angelClient.run();
    Thread.sleep(5000);
    psId = new ParameterServerId(0);
    psAttempt0Id = new PSAttemptId(psId, 0);
    WorkerGroupId workerGroupId = new WorkerGroupId(0);
    workerId = new WorkerId(workerGroupId, 0);
    workerAttempt0Id = new WorkerAttemptId(workerId, 0);
}
Also used : CombineTextInputFormat(org.apache.hadoop.mapreduce.lib.input.CombineTextInputFormat) MatrixContext(com.tencent.angel.ml.matrix.MatrixContext) Configuration(org.apache.hadoop.conf.Configuration) PSAttemptId(com.tencent.angel.ps.PSAttemptId) WorkerAttemptId(com.tencent.angel.worker.WorkerAttemptId) ParameterServerId(com.tencent.angel.ps.ParameterServerId) WorkerId(com.tencent.angel.worker.WorkerId) WorkerGroupId(com.tencent.angel.worker.WorkerGroupId) Before(org.junit.Before)

Example 9 with WorkerAttemptId

use of com.tencent.angel.worker.WorkerAttemptId in project angel by Tencent.

the class PSAgentTest method setup.

@BeforeClass
public static 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());
        conf.set(AngelConf.ANGEL_PS_ROUTER_TYPE, "range");
        // use local deploy mode and dummy dataspliter
        conf.set(AngelConf.ANGEL_DEPLOY_MODE, "LOCAL");
        conf.setBoolean(AngelConf.ANGEL_AM_USE_DUMMY_DATASPLITER, true);
        // conf.setInt(AngelConf.ANGEL_PREPROCESS_VECTOR_MAXDIM, 10000);
        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_PSAGENT_CACHE_SYNC_TIMEINTERVAL_MS, 10);
        conf.setInt(AngelConf.ANGEL_WORKER_HEARTBEAT_INTERVAL_MS, 1000);
        conf.setInt(AngelConf.ANGEL_PS_HEARTBEAT_INTERVAL_MS, 1000);
        conf.setInt(AngelConf.ANGEL_WORKER_MAX_ATTEMPTS, 1);
        conf.setInt(AngelConf.ANGEL_PS_MAX_ATTEMPTS, 1);
        // 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_DOUBLE_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_DOUBLE");
        mMatrix.setPartitionerClass(ColumnRangePartitioner.class);
        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, "true");
        mMatrix2.set(MatrixConf.MATRIX_AVERAGE, "false");
        mMatrix2.set(MatrixConf.MATRIX_OPLOG_TYPE, "DENSE_DOUBLE");
        mMatrix2.setPartitionerClass(ColumnRangePartitioner.class);
        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;
    }
}
Also used : CombineTextInputFormat(org.apache.hadoop.mapreduce.lib.input.CombineTextInputFormat) MatrixContext(com.tencent.angel.ml.matrix.MatrixContext) TaskId(com.tencent.angel.worker.task.TaskId) Configuration(org.apache.hadoop.conf.Configuration) PSAttemptId(com.tencent.angel.ps.PSAttemptId) WorkerAttemptId(com.tencent.angel.worker.WorkerAttemptId) WorkerId(com.tencent.angel.worker.WorkerId) ParameterServerId(com.tencent.angel.ps.ParameterServerId) WorkerGroupId(com.tencent.angel.worker.WorkerGroupId) BeforeClass(org.junit.BeforeClass)

Example 10 with WorkerAttemptId

use of com.tencent.angel.worker.WorkerAttemptId in project angel by Tencent.

the class UpdateRowsHashTest method setup.

@Before
public void setup() throws Exception {
    // 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, 2);
    conf.setInt(AngelConf.ANGEL_WORKER_TASK_NUMBER, 1);
    conf.setInt(AngelConf.ANGEL_MODEL_PARTITIONER_PARTITION_SIZE, 1000);
    conf.setBoolean("use.new.split", false);
    conf.setInt(AngelConf.ANGEL_PSAGENT_CACHE_SYNC_TIMEINTERVAL_MS, 10);
    conf.setInt(AngelConf.ANGEL_WORKER_HEARTBEAT_INTERVAL_MS, 1000);
    conf.setInt(AngelConf.ANGEL_PS_HEARTBEAT_INTERVAL_MS, 1000);
    conf.setInt(AngelConf.ANGEL_WORKER_MAX_ATTEMPTS, 1);
    conf.setInt(AngelConf.ANGEL_PS_MAX_ATTEMPTS, 1);
    // get a angel client
    angelClient = AngelClientFactory.get(conf);
    // add sparse double matrix
    MatrixContext sMat = new MatrixContext();
    sMat.setName(SPARSE_DOUBLE_MAT);
    sMat.setRowNum(rowNum);
    sMat.setPartitionerClass(HashPartitioner.class);
    sMat.setPartitionNum(partNum);
    sMat.setRowType(RowType.T_DOUBLE_SPARSE);
    angelClient.addMatrix(sMat);
    // add sparse float matrix
    MatrixContext sfMat = new MatrixContext();
    sfMat.setName(SPARSE_FLOAT_MAT);
    sfMat.setRowNum(rowNum);
    sfMat.setPartitionerClass(HashPartitioner.class);
    sfMat.setPartitionNum(partNum);
    sfMat.setRowType(RowType.T_FLOAT_SPARSE);
    angelClient.addMatrix(sfMat);
    // add sparse float matrix
    MatrixContext siMat = new MatrixContext();
    siMat.setName(SPARSE_INT_MAT);
    siMat.setRowNum(rowNum);
    siMat.setPartitionerClass(HashPartitioner.class);
    siMat.setPartitionNum(partNum);
    siMat.setRowType(RowType.T_INT_SPARSE);
    angelClient.addMatrix(siMat);
    // add sparse long matrix
    MatrixContext slMat = new MatrixContext();
    slMat.setName(SPARSE_LONG_MAT);
    slMat.setRowNum(rowNum);
    slMat.setPartitionerClass(HashPartitioner.class);
    slMat.setPartitionNum(partNum);
    slMat.setRowType(RowType.T_LONG_SPARSE);
    angelClient.addMatrix(slMat);
    // add sparse long-key double matrix
    MatrixContext dLongKeysMatrix = new MatrixContext();
    dLongKeysMatrix.setName(SPARSE_DOUBLE_LONG_MAT);
    dLongKeysMatrix.setRowNum(rowNum);
    dLongKeysMatrix.setPartitionerClass(HashPartitioner.class);
    dLongKeysMatrix.setPartitionNum(partNum);
    dLongKeysMatrix.setRowType(RowType.T_DOUBLE_SPARSE_LONGKEY);
    angelClient.addMatrix(dLongKeysMatrix);
    // add sparse long-key float matrix
    MatrixContext slfMatrix = new MatrixContext();
    slfMatrix.setName(SPARSE_FLOAT_LONG_MAT);
    slfMatrix.setRowNum(rowNum);
    slfMatrix.setPartitionerClass(HashPartitioner.class);
    slfMatrix.setPartitionNum(partNum);
    slfMatrix.setRowType(RowType.T_FLOAT_SPARSE_LONGKEY);
    angelClient.addMatrix(slfMatrix);
    // add sparse long-key int matrix
    MatrixContext sliMatrix = new MatrixContext();
    sliMatrix.setName(SPARSE_INT_LONG_MAT);
    sliMatrix.setRowNum(rowNum);
    sliMatrix.setPartitionerClass(HashPartitioner.class);
    sliMatrix.setPartitionNum(partNum);
    sliMatrix.setRowType(RowType.T_INT_SPARSE_LONGKEY);
    angelClient.addMatrix(sliMatrix);
    // add sparse long-key long matrix
    MatrixContext sllMatrix = new MatrixContext();
    sllMatrix.setName(SPARSE_LONG_LONG_MAT);
    sllMatrix.setRowNum(rowNum);
    sllMatrix.setPartitionerClass(HashPartitioner.class);
    sllMatrix.setPartitionNum(partNum);
    sllMatrix.setRowType(RowType.T_LONG_SPARSE_LONGKEY);
    angelClient.addMatrix(sllMatrix);
    // Start PS
    angelClient.startPSServer();
    // Start to run application
    angelClient.run();
    Thread.sleep(5000);
    psId = new ParameterServerId(0);
    psAttempt0Id = new PSAttemptId(psId, 0);
    WorkerGroupId workerGroupId = new WorkerGroupId(0);
    workerId = new WorkerId(workerGroupId, 0);
    workerAttempt0Id = new WorkerAttemptId(workerId, 0);
}
Also used : CombineTextInputFormat(org.apache.hadoop.mapreduce.lib.input.CombineTextInputFormat) MatrixContext(com.tencent.angel.ml.matrix.MatrixContext) Configuration(org.apache.hadoop.conf.Configuration) PSAttemptId(com.tencent.angel.ps.PSAttemptId) WorkerAttemptId(com.tencent.angel.worker.WorkerAttemptId) ParameterServerId(com.tencent.angel.ps.ParameterServerId) WorkerId(com.tencent.angel.worker.WorkerId) WorkerGroupId(com.tencent.angel.worker.WorkerGroupId) Before(org.junit.Before)

Aggregations

WorkerAttemptId (com.tencent.angel.worker.WorkerAttemptId)45 PSAttemptId (com.tencent.angel.ps.PSAttemptId)32 WorkerGroupId (com.tencent.angel.worker.WorkerGroupId)31 WorkerId (com.tencent.angel.worker.WorkerId)30 ParameterServerId (com.tencent.angel.ps.ParameterServerId)28 Configuration (org.apache.hadoop.conf.Configuration)27 CombineTextInputFormat (org.apache.hadoop.mapreduce.lib.input.CombineTextInputFormat)27 MatrixContext (com.tencent.angel.ml.matrix.MatrixContext)26 Before (org.junit.Before)22 TaskId (com.tencent.angel.worker.task.TaskId)9 UnvalidIdStrException (com.tencent.angel.exception.UnvalidIdStrException)4 WorkerAttemptEvent (com.tencent.angel.master.worker.attempt.WorkerAttemptEvent)4 Test (org.junit.Test)4 ServiceException (com.google.protobuf.ServiceException)3 Id (com.tencent.angel.common.Id)3 AngelException (com.tencent.angel.exception.AngelException)3 WorkerAttempt (com.tencent.angel.master.worker.attempt.WorkerAttempt)3 WorkerAttemptDiagnosticsUpdateEvent (com.tencent.angel.master.worker.attempt.WorkerAttemptDiagnosticsUpdateEvent)3 Worker (com.tencent.angel.worker.Worker)3 BeforeClass (org.junit.BeforeClass)3