use of com.tencent.angel.worker.WorkerAttemptId in project angel by Tencent.
the class MasterService method getWorkerGroupMetaInfo.
/**
* get worker group information: tasks, workers, data splits
*
* @param controller rpc controller of protobuf
* @param request contains worker attempt id
*/
@Override
public GetWorkerGroupMetaInfoResponse getWorkerGroupMetaInfo(RpcController controller, GetWorkerGroupMetaInfoRequest request) throws ServiceException {
if (LOG.isDebugEnabled()) {
LOG.debug("receive get workergroup info, request=" + request);
}
WorkerAttemptId workerAttemptId = ProtobufUtil.convertToId(request.getWorkerAttemptId());
// find workergroup in worker manager
AMWorkerGroup group = context.getWorkerManager().getWorkerGroup(workerAttemptId.getWorkerId().getWorkerGroupId());
if (group == null || group.getState() == AMWorkerGroupState.NEW || group.getState() == AMWorkerGroupState.INITED) {
// if this worker group does not initialized, just return WORKERGROUP_NOTREADY
return GetWorkerGroupMetaInfoResponse.newBuilder().setWorkerGroupStatus(GetWorkerGroupMetaInfoResponse.WorkerGroupStatus.WORKERGROUP_NOTREADY).build();
} else if (group.getState() == AMWorkerGroupState.FAILED || group.getState() == AMWorkerGroupState.KILLED || group.getState() == AMWorkerGroupState.SUCCESS) {
// if this worker group run over, just return WORKERGROUP_EXITED
return GetWorkerGroupMetaInfoResponse.newBuilder().setWorkerGroupStatus(GetWorkerGroupMetaInfoResponse.WorkerGroupStatus.WORKERGROUP_EXITED).build();
} else {
// if this worker group is running now, return tasks, workers, data splits for it
try {
return ProtobufUtil.buildGetWorkerGroupMetaResponse(group, context.getDataSpliter().getSplits(group.getSplitIndex()), context.getConf());
} catch (Exception e) {
LOG.error("build workergroup information error", e);
throw new ServiceException(e);
}
}
}
use of com.tencent.angel.worker.WorkerAttemptId in project angel by Tencent.
the class YarnContainerAllocator method containerAssigned.
@SuppressWarnings("unchecked")
private void containerAssigned(Container allocated, ContainerRequest assigned) {
// Update resource requests
decContainerReq(assigned);
// send the container-assigned event to task attempt
if (assigned.id instanceof PSAttemptId) {
context.getEventHandler().handle(new PSAttemptContainerAssignedEvent((PSAttemptId) assigned.id, allocated));
} else if (assigned.id instanceof WorkerAttemptId) {
context.getEventHandler().handle(new WorkerAttemptContainerAssignedEvent((WorkerAttemptId) assigned.id, allocated));
}
assignedContainerToIDMap.put(allocated.getId(), assigned.id);
idToContainerMap.put(assigned.id, allocated);
LOG.info("Assigned container (" + allocated + ") " + " to task " + assigned.id + " on node " + allocated.getNodeId().toString());
}
use of com.tencent.angel.worker.WorkerAttemptId in project angel by Tencent.
the class GetRowsTest 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 dense double matrix
MatrixContext dMat = new MatrixContext();
dMat.setName(DENSE_DOUBLE_MAT);
dMat.setRowNum(rowNum);
dMat.setColNum(feaNum);
dMat.setMaxRowNumInBlock(blockRowNum);
dMat.setMaxColNumInBlock(blockColNum);
dMat.setRowType(RowType.T_DOUBLE_DENSE);
dMat.setPartitionerClass(ColumnRangePartitioner.class);
angelClient.addMatrix(dMat);
// add sparse double matrix
MatrixContext sMat = new MatrixContext();
sMat.setName(SPARSE_DOUBLE_MAT);
sMat.setRowNum(rowNum);
sMat.setColNum(feaNum);
sMat.setMaxRowNumInBlock(blockRowNum);
sMat.setMaxColNumInBlock(blockColNum);
sMat.setRowType(RowType.T_DOUBLE_SPARSE);
sMat.setPartitionNum(partNum);
sMat.setPartitionerClass(HashPartitioner.class);
angelClient.addMatrix(sMat);
// add dense float matrix
MatrixContext dfMat = new MatrixContext();
dfMat.setName(DENSE_FLOAT_MAT);
dfMat.setRowNum(rowNum);
dfMat.setColNum(feaNum);
dfMat.setMaxRowNumInBlock(blockRowNum);
dfMat.setMaxColNumInBlock(blockColNum);
dfMat.setRowType(RowType.T_FLOAT_DENSE);
dfMat.setPartitionerClass(ColumnRangePartitioner.class);
angelClient.addMatrix(dfMat);
// add sparse float matrix
MatrixContext sfMat = new MatrixContext();
sfMat.setName(SPARSE_FLOAT_MAT);
sfMat.setRowNum(rowNum);
sfMat.setColNum(feaNum);
sfMat.setMaxRowNumInBlock(blockRowNum);
sfMat.setMaxColNumInBlock(blockColNum);
sfMat.setRowType(RowType.T_FLOAT_SPARSE);
sfMat.setPartitionNum(partNum);
sfMat.setPartitionerClass(HashPartitioner.class);
angelClient.addMatrix(sfMat);
// add dense float matrix
MatrixContext diMat = new MatrixContext();
diMat.setName(DENSE_INT_MAT);
diMat.setRowNum(rowNum);
diMat.setColNum(feaNum);
diMat.setMaxRowNumInBlock(blockRowNum);
diMat.setMaxColNumInBlock(blockColNum);
diMat.setRowType(RowType.T_INT_DENSE);
diMat.setPartitionerClass(ColumnRangePartitioner.class);
angelClient.addMatrix(diMat);
// add sparse float matrix
MatrixContext siMat = new MatrixContext();
siMat.setName(SPARSE_INT_MAT);
siMat.setRowNum(rowNum);
siMat.setColNum(feaNum);
siMat.setMaxRowNumInBlock(blockRowNum);
siMat.setMaxColNumInBlock(blockColNum);
siMat.setRowType(RowType.T_INT_SPARSE);
siMat.setPartitionNum(partNum);
siMat.setPartitionerClass(HashPartitioner.class);
angelClient.addMatrix(siMat);
// add dense long matrix
MatrixContext dlMat = new MatrixContext();
dlMat.setName(DENSE_LONG_MAT);
dlMat.setRowNum(rowNum);
dlMat.setColNum(feaNum);
dlMat.setMaxRowNumInBlock(blockRowNum);
dlMat.setMaxColNumInBlock(blockColNum);
dlMat.setRowType(RowType.T_LONG_DENSE);
dlMat.setPartitionerClass(ColumnRangePartitioner.class);
angelClient.addMatrix(dlMat);
// add sparse long matrix
MatrixContext slMat = new MatrixContext();
slMat.setName(SPARSE_LONG_MAT);
slMat.setRowNum(rowNum);
slMat.setColNum(feaNum);
slMat.setMaxRowNumInBlock(blockRowNum);
slMat.setMaxColNumInBlock(blockColNum);
slMat.setRowType(RowType.T_LONG_SPARSE);
slMat.setPartitionNum(partNum);
slMat.setPartitionerClass(HashPartitioner.class);
angelClient.addMatrix(slMat);
// add sparse long-key double matrix
MatrixContext dLongKeysMatrix = new MatrixContext();
dLongKeysMatrix.setName(SPARSE_DOUBLE_LONG_MAT);
dLongKeysMatrix.setRowNum(rowNum);
dLongKeysMatrix.setColNum(feaNum);
dLongKeysMatrix.setMaxRowNumInBlock(blockRowNum);
dLongKeysMatrix.setMaxColNumInBlock(blockColNum);
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.setColNum(feaNum);
slfMatrix.setMaxRowNumInBlock(blockRowNum);
slfMatrix.setMaxColNumInBlock(blockColNum);
slfMatrix.setRowType(RowType.T_FLOAT_SPARSE_LONGKEY);
slfMatrix.setPartitionNum(partNum);
slfMatrix.setPartitionerClass(HashPartitioner.class);
angelClient.addMatrix(slfMatrix);
// add sparse long-key int matrix
MatrixContext sliMatrix = new MatrixContext();
sliMatrix.setName(SPARSE_INT_LONG_MAT);
sliMatrix.setRowNum(rowNum);
sliMatrix.setColNum(feaNum);
sliMatrix.setMaxRowNumInBlock(blockRowNum);
sliMatrix.setMaxColNumInBlock(blockColNum);
sliMatrix.setRowType(RowType.T_INT_SPARSE_LONGKEY);
sliMatrix.setPartitionNum(partNum);
sliMatrix.setPartitionerClass(HashPartitioner.class);
angelClient.addMatrix(sliMatrix);
// add sparse long-key long matrix
MatrixContext sllMatrix = new MatrixContext();
sllMatrix.setName(SPARSE_LONG_LONG_MAT);
sllMatrix.setRowNum(rowNum);
sllMatrix.setColNum(feaNum);
sllMatrix.setMaxRowNumInBlock(blockRowNum);
sllMatrix.setMaxColNumInBlock(blockColNum);
sllMatrix.setRowType(RowType.T_LONG_SPARSE_LONGKEY);
sllMatrix.setPartitionNum(partNum);
sllMatrix.setPartitionerClass(HashPartitioner.class);
angelClient.addMatrix(sllMatrix);
// Start PS
angelClient.startPSServer();
// Start to run application
angelClient.run();
Thread.sleep(2000);
psId = new ParameterServerId(0);
psAttempt0Id = new PSAttemptId(psId, 0);
WorkerGroupId workerGroupId = new WorkerGroupId(0);
workerId = new WorkerId(workerGroupId, 0);
workerAttempt0Id = new WorkerAttemptId(workerId, 0);
}
use of com.tencent.angel.worker.WorkerAttemptId in project angel by Tencent.
the class MasterRecoverTest 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_AM_WRITE_STATE_INTERVAL_MS, 1000);
conf.setInt(AngelConf.ANGEL_WORKER_HEARTBEAT_INTERVAL_MS, 1000);
conf.setInt(AngelConf.ANGEL_PS_HEARTBEAT_INTERVAL_MS, 1000);
// 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;
}
}
use of com.tencent.angel.worker.WorkerAttemptId in project angel by Tencent.
the class UpdatePSFTest 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, 1);
conf.setInt(AngelConf.ANGEL_WORKER_TASK_NUMBER, 1);
conf.setInt(AngelConf.ANGEL_MODEL_PARTITIONER_PARTITION_SIZE, 1000);
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.setBoolean("use.new.split", true);
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 dense double matrix
MatrixContext dMat = new MatrixContext();
dMat.setName(DENSE_DOUBLE_MAT);
dMat.setRowNum(1);
dMat.setColNum(feaNum);
dMat.setMaxColNumInBlock(feaNum / 3);
dMat.setRowType(RowType.T_DOUBLE_DENSE);
dMat.setPartitionerClass(ColumnRangePartitioner.class);
angelClient.addMatrix(dMat);
// add sparse double matrix
MatrixContext sMat = new MatrixContext();
sMat.setName(SPARSE_DOUBLE_MAT);
sMat.setRowNum(1);
sMat.setColNum(feaNum);
sMat.setMaxColNumInBlock(feaNum);
sMat.setRowType(RowType.T_DOUBLE_SPARSE);
angelClient.addMatrix(sMat);
// add dense float matrix
MatrixContext dfMat = new MatrixContext();
dfMat.setName(DENSE_FLOAT_MAT);
dfMat.setRowNum(1);
dfMat.setColNum(feaNum);
dfMat.setMaxColNumInBlock(feaNum / 3);
dfMat.setRowType(RowType.T_FLOAT_DENSE);
dfMat.setPartitionerClass(ColumnRangePartitioner.class);
angelClient.addMatrix(dfMat);
// add sparse float matrix
MatrixContext sfMat = new MatrixContext();
sfMat.setName(SPARSE_FLOAT_MAT);
sfMat.setRowNum(1);
sfMat.setColNum(feaNum);
sfMat.setMaxColNumInBlock(feaNum / 3);
sfMat.setRowType(RowType.T_FLOAT_SPARSE);
angelClient.addMatrix(sfMat);
// add dense float matrix
MatrixContext diMat = new MatrixContext();
diMat.setName(DENSE_INT_MAT);
diMat.setRowNum(1);
diMat.setColNum(feaNum);
diMat.setMaxColNumInBlock(feaNum / 3);
diMat.setRowType(RowType.T_INT_DENSE);
diMat.setPartitionerClass(ColumnRangePartitioner.class);
angelClient.addMatrix(diMat);
// add sparse float matrix
MatrixContext siMat = new MatrixContext();
siMat.setName(SPARSE_INT_MAT);
siMat.setRowNum(1);
siMat.setColNum(feaNum);
siMat.setMaxColNumInBlock(feaNum / 3);
siMat.setRowType(RowType.T_INT_SPARSE);
angelClient.addMatrix(siMat);
// add dense long matrix
MatrixContext dlMat = new MatrixContext();
dlMat.setName(DENSE_LONG_MAT);
dlMat.setRowNum(1);
dlMat.setColNum(feaNum);
dlMat.setMaxColNumInBlock(feaNum / 3);
dlMat.setRowType(RowType.T_LONG_DENSE);
dlMat.setPartitionerClass(ColumnRangePartitioner.class);
angelClient.addMatrix(dlMat);
// add sparse long matrix
MatrixContext slMat = new MatrixContext();
slMat.setName(SPARSE_LONG_MAT);
slMat.setRowNum(1);
slMat.setColNum(feaNum);
slMat.setMaxColNumInBlock(feaNum / 3);
slMat.setRowType(RowType.T_LONG_SPARSE);
angelClient.addMatrix(slMat);
// add sparse long-key float matrix
MatrixContext sldMatrix = new MatrixContext();
sldMatrix.setName(SPARSE_DOUBLE_LONG_MAT);
sldMatrix.setRowNum(1);
sldMatrix.setColNum(feaNum);
sldMatrix.setMaxColNumInBlock(feaNum / 3);
sldMatrix.setRowType(RowType.T_DOUBLE_SPARSE_LONGKEY);
angelClient.addMatrix(sldMatrix);
// add sparse long-key float matrix
MatrixContext slfMatrix = new MatrixContext();
slfMatrix.setName(SPARSE_FLOAT_LONG_MAT);
slfMatrix.setRowNum(1);
slfMatrix.setColNum(feaNum);
slfMatrix.setMaxColNumInBlock(feaNum / 3);
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(1);
sliMatrix.setColNum(feaNum);
sliMatrix.setMaxColNumInBlock(feaNum / 3);
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(1);
sllMatrix.setColNum(feaNum);
sllMatrix.setMaxColNumInBlock(feaNum / 3);
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);
}
Aggregations