Examples with TaskStats io.trino.operator.TaskStats used on opensource projects

Example 1 with TaskStats

use of io.trino.operator.TaskStats in project trino by trinodb.

the class TaskSystemTable method cursor.

@Override
public RecordCursor cursor(ConnectorTransactionHandle transactionHandle, ConnectorSession session, TupleDomain<Integer> constraint) {
    Builder table = InMemoryRecordSet.builder(TASK_TABLE);
    for (TaskInfo taskInfo : taskManager.getAllTaskInfo()) {
        TaskStats stats = taskInfo.getStats();
        TaskStatus taskStatus = taskInfo.getTaskStatus();
        table.addRow(nodeId, taskStatus.getTaskId().toString(), taskStatus.getTaskId().getStageId().toString(), taskStatus.getTaskId().getQueryId().toString(), taskStatus.getState().toString(), (long) stats.getTotalDrivers(), (long) stats.getQueuedDrivers(), (long) stats.getRunningDrivers(), (long) stats.getCompletedDrivers(), toMillis(stats.getTotalScheduledTime()), toMillis(stats.getTotalCpuTime()), toMillis(stats.getTotalBlockedTime()), toBytes(stats.getRawInputDataSize()), stats.getRawInputPositions(), toBytes(stats.getProcessedInputDataSize()), stats.getProcessedInputPositions(), toBytes(stats.getOutputDataSize()), stats.getOutputPositions(), toBytes(stats.getPhysicalInputDataSize()), toBytes(stats.getPhysicalWrittenDataSize()), toTimestampWithTimeZoneMillis(stats.getCreateTime()), toTimestampWithTimeZoneMillis(stats.getFirstStartTime()), toTimestampWithTimeZoneMillis(taskInfo.getLastHeartbeat()), toTimestampWithTimeZoneMillis(stats.getEndTime()));
    }
    return table.build().cursor();
}

Example 2 with TaskStats

use of io.trino.operator.TaskStats in project trino by trinodb.

the class SqlTask method createTaskInfo.

private TaskInfo createTaskInfo(TaskHolder taskHolder) {
    // create task status first to prevent potentially seeing incomplete stats for a done task state
    TaskStatus taskStatus = createTaskStatus(taskHolder);
    TaskStats taskStats = getTaskStats(taskHolder);
    Set<PlanNodeId> noMoreSplits = getNoMoreSplits(taskHolder);
    return new TaskInfo(taskStatus, lastHeartbeat.get(), outputBuffer.getInfo(), noMoreSplits, taskStats, needsPlan.get());
}

Example 3 with TaskStats

use of io.trino.operator.TaskStats in project trino by trinodb.

the class SqlTask method getTaskStats.

private TaskStats getTaskStats(TaskHolder taskHolder) {
    TaskInfo finalTaskInfo = taskHolder.getFinalTaskInfo();
    if (finalTaskInfo != null) {
        return finalTaskInfo.getStats();
    }
    SqlTaskExecution taskExecution = taskHolder.getTaskExecution();
    if (taskExecution != null) {
        return taskExecution.getTaskContext().getTaskStats();
    }
    // if the task completed without creation, set end time
    DateTime endTime = taskStateMachine.getState().isDone() ? DateTime.now() : null;
    return new TaskStats(taskStateMachine.getCreatedTime(), endTime);
}

Example 4 with TaskStats

use of io.trino.operator.TaskStats in project trino by trinodb.

the class SqlTask method createTaskStatus.

private TaskStatus createTaskStatus(TaskHolder taskHolder) {
    // Obtain task status version before building actual TaskStatus object.
    // This way any task updates won't be lost since all updates happen
    // before version number is increased.
    long versionNumber = taskStatusVersion.get();
    TaskState state = taskStateMachine.getState();
    List<ExecutionFailureInfo> failures = ImmutableList.of();
    if (state == FAILED) {
        failures = toFailures(taskStateMachine.getFailureCauses());
    }
    int queuedPartitionedDrivers = 0;
    long queuedPartitionedSplitsWeight = 0L;
    int runningPartitionedDrivers = 0;
    long runningPartitionedSplitsWeight = 0L;
    DataSize physicalWrittenDataSize = DataSize.ofBytes(0);
    DataSize userMemoryReservation = DataSize.ofBytes(0);
    DataSize revocableMemoryReservation = DataSize.ofBytes(0);
    // TODO: add a mechanism to avoid sending the whole completedDriverGroups set over the wire for every task status reply
    Set<Lifespan> completedDriverGroups = ImmutableSet.of();
    long fullGcCount = 0;
    Duration fullGcTime = new Duration(0, MILLISECONDS);
    long dynamicFiltersVersion = INITIAL_DYNAMIC_FILTERS_VERSION;
    if (taskHolder.getFinalTaskInfo() != null) {
        TaskInfo taskInfo = taskHolder.getFinalTaskInfo();
        TaskStats taskStats = taskInfo.getStats();
        queuedPartitionedDrivers = taskStats.getQueuedPartitionedDrivers();
        queuedPartitionedSplitsWeight = taskStats.getQueuedPartitionedSplitsWeight();
        runningPartitionedDrivers = taskStats.getRunningPartitionedDrivers();
        runningPartitionedSplitsWeight = taskStats.getRunningPartitionedSplitsWeight();
        physicalWrittenDataSize = taskStats.getPhysicalWrittenDataSize();
        userMemoryReservation = taskStats.getUserMemoryReservation();
        revocableMemoryReservation = taskStats.getRevocableMemoryReservation();
        fullGcCount = taskStats.getFullGcCount();
        fullGcTime = taskStats.getFullGcTime();
    } else if (taskHolder.getTaskExecution() != null) {
        long physicalWrittenBytes = 0;
        TaskContext taskContext = taskHolder.getTaskExecution().getTaskContext();
        for (PipelineContext pipelineContext : taskContext.getPipelineContexts()) {
            PipelineStatus pipelineStatus = pipelineContext.getPipelineStatus();
            queuedPartitionedDrivers += pipelineStatus.getQueuedPartitionedDrivers();
            queuedPartitionedSplitsWeight += pipelineStatus.getQueuedPartitionedSplitsWeight();
            runningPartitionedDrivers += pipelineStatus.getRunningPartitionedDrivers();
            runningPartitionedSplitsWeight += pipelineStatus.getRunningPartitionedSplitsWeight();
            physicalWrittenBytes += pipelineContext.getPhysicalWrittenDataSize();
        }
        physicalWrittenDataSize = succinctBytes(physicalWrittenBytes);
        userMemoryReservation = taskContext.getMemoryReservation();
        revocableMemoryReservation = taskContext.getRevocableMemoryReservation();
        completedDriverGroups = taskContext.getCompletedDriverGroups();
        fullGcCount = taskContext.getFullGcCount();
        fullGcTime = taskContext.getFullGcTime();
        dynamicFiltersVersion = taskContext.getDynamicFiltersVersion();
    }
    return new TaskStatus(taskStateMachine.getTaskId(), taskInstanceId, versionNumber, state, location, nodeId, completedDriverGroups, failures, queuedPartitionedDrivers, runningPartitionedDrivers, isOutputBufferOverutilized(), physicalWrittenDataSize, userMemoryReservation, revocableMemoryReservation, fullGcCount, fullGcTime, dynamicFiltersVersion, queuedPartitionedSplitsWeight, runningPartitionedSplitsWeight);
}

Example 5 with TaskStats

use of io.trino.operator.TaskStats in project trino by trinodb.

the class StageStateMachine method getStageInfo.

public StageInfo getStageInfo(Supplier<Iterable<TaskInfo>> taskInfosSupplier) {
    Optional<StageInfo> finalStageInfo = this.finalStageInfo.get();
    if (finalStageInfo.isPresent()) {
        return finalStageInfo.get();
    }
    // stage state must be captured first in order to provide a
    // consistent view of the stage. For example, building this
    // information, the stage could finish, and the task states would
    // never be visible.
    StageState state = stageState.get();
    List<TaskInfo> taskInfos = ImmutableList.copyOf(taskInfosSupplier.get());
    int totalTasks = taskInfos.size();
    int runningTasks = 0;
    int completedTasks = 0;
    int failedTasks = 0;
    int totalDrivers = 0;
    int queuedDrivers = 0;
    int runningDrivers = 0;
    int blockedDrivers = 0;
    int completedDrivers = 0;
    long cumulativeUserMemory = 0;
    long failedCumulativeUserMemory = 0;
    long userMemoryReservation = 0;
    long revocableMemoryReservation = 0;
    long totalMemoryReservation = 0;
    long peakUserMemoryReservation = peakUserMemory.get();
    long peakRevocableMemoryReservation = peakRevocableMemory.get();
    long totalScheduledTime = 0;
    long failedScheduledTime = 0;
    long totalCpuTime = 0;
    long failedCpuTime = 0;
    long totalBlockedTime = 0;
    long physicalInputDataSize = 0;
    long failedPhysicalInputDataSize = 0;
    long physicalInputPositions = 0;
    long failedPhysicalInputPositions = 0;
    long physicalInputReadTime = 0;
    long failedPhysicalInputReadTime = 0;
    long internalNetworkInputDataSize = 0;
    long failedInternalNetworkInputDataSize = 0;
    long internalNetworkInputPositions = 0;
    long failedInternalNetworkInputPositions = 0;
    long rawInputDataSize = 0;
    long failedRawInputDataSize = 0;
    long rawInputPositions = 0;
    long failedRawInputPositions = 0;
    long processedInputDataSize = 0;
    long failedProcessedInputDataSize = 0;
    long processedInputPositions = 0;
    long failedProcessedInputPositions = 0;
    long bufferedDataSize = 0;
    long outputDataSize = 0;
    long failedOutputDataSize = 0;
    long outputPositions = 0;
    long failedOutputPositions = 0;
    long physicalWrittenDataSize = 0;
    long failedPhysicalWrittenDataSize = 0;
    int fullGcCount = 0;
    int fullGcTaskCount = 0;
    int minFullGcSec = 0;
    int maxFullGcSec = 0;
    int totalFullGcSec = 0;
    boolean fullyBlocked = true;
    Set<BlockedReason> blockedReasons = new HashSet<>();
    Map<String, OperatorStats> operatorToStats = new HashMap<>();
    for (TaskInfo taskInfo : taskInfos) {
        TaskState taskState = taskInfo.getTaskStatus().getState();
        if (taskState.isDone()) {
            completedTasks++;
        } else {
            runningTasks++;
        }
        if (taskState == TaskState.FAILED) {
            failedTasks++;
        }
        TaskStats taskStats = taskInfo.getStats();
        totalDrivers += taskStats.getTotalDrivers();
        queuedDrivers += taskStats.getQueuedDrivers();
        runningDrivers += taskStats.getRunningDrivers();
        blockedDrivers += taskStats.getBlockedDrivers();
        completedDrivers += taskStats.getCompletedDrivers();
        cumulativeUserMemory += taskStats.getCumulativeUserMemory();
        if (taskState == TaskState.FAILED) {
            failedCumulativeUserMemory += taskStats.getCumulativeUserMemory();
        }
        long taskUserMemory = taskStats.getUserMemoryReservation().toBytes();
        long taskRevocableMemory = taskStats.getRevocableMemoryReservation().toBytes();
        userMemoryReservation += taskUserMemory;
        revocableMemoryReservation += taskRevocableMemory;
        totalMemoryReservation += taskUserMemory + taskRevocableMemory;
        totalScheduledTime += taskStats.getTotalScheduledTime().roundTo(NANOSECONDS);
        totalCpuTime += taskStats.getTotalCpuTime().roundTo(NANOSECONDS);
        totalBlockedTime += taskStats.getTotalBlockedTime().roundTo(NANOSECONDS);
        if (taskState == TaskState.FAILED) {
            failedScheduledTime += taskStats.getTotalScheduledTime().roundTo(NANOSECONDS);
            failedCpuTime += taskStats.getTotalCpuTime().roundTo(NANOSECONDS);
        }
        if (!taskState.isDone()) {
            fullyBlocked &= taskStats.isFullyBlocked();
            blockedReasons.addAll(taskStats.getBlockedReasons());
        }
        physicalInputDataSize += taskStats.getPhysicalInputDataSize().toBytes();
        physicalInputPositions += taskStats.getPhysicalInputPositions();
        physicalInputReadTime += taskStats.getPhysicalInputReadTime().roundTo(NANOSECONDS);
        internalNetworkInputDataSize += taskStats.getInternalNetworkInputDataSize().toBytes();
        internalNetworkInputPositions += taskStats.getInternalNetworkInputPositions();
        rawInputDataSize += taskStats.getRawInputDataSize().toBytes();
        rawInputPositions += taskStats.getRawInputPositions();
        processedInputDataSize += taskStats.getProcessedInputDataSize().toBytes();
        processedInputPositions += taskStats.getProcessedInputPositions();
        bufferedDataSize += taskInfo.getOutputBuffers().getTotalBufferedBytes();
        outputDataSize += taskStats.getOutputDataSize().toBytes();
        outputPositions += taskStats.getOutputPositions();
        physicalWrittenDataSize += taskStats.getPhysicalWrittenDataSize().toBytes();
        if (taskState == TaskState.FAILED) {
            failedPhysicalInputDataSize += taskStats.getPhysicalInputDataSize().toBytes();
            failedPhysicalInputPositions += taskStats.getPhysicalInputPositions();
            failedPhysicalInputReadTime += taskStats.getPhysicalInputReadTime().roundTo(NANOSECONDS);
            failedInternalNetworkInputDataSize += taskStats.getInternalNetworkInputDataSize().toBytes();
            failedInternalNetworkInputPositions += taskStats.getInternalNetworkInputPositions();
            failedRawInputDataSize += taskStats.getRawInputDataSize().toBytes();
            failedRawInputPositions += taskStats.getRawInputPositions();
            failedProcessedInputDataSize += taskStats.getProcessedInputDataSize().toBytes();
            failedProcessedInputPositions += taskStats.getProcessedInputPositions();
            failedOutputDataSize += taskStats.getOutputDataSize().toBytes();
            failedOutputPositions += taskStats.getOutputPositions();
            failedPhysicalWrittenDataSize += taskStats.getPhysicalWrittenDataSize().toBytes();
        }
        fullGcCount += taskStats.getFullGcCount();
        fullGcTaskCount += taskStats.getFullGcCount() > 0 ? 1 : 0;
        int gcSec = toIntExact(taskStats.getFullGcTime().roundTo(SECONDS));
        totalFullGcSec += gcSec;
        minFullGcSec = min(minFullGcSec, gcSec);
        maxFullGcSec = max(maxFullGcSec, gcSec);
        for (PipelineStats pipeline : taskStats.getPipelines()) {
            for (OperatorStats operatorStats : pipeline.getOperatorSummaries()) {
                String id = pipeline.getPipelineId() + "." + operatorStats.getOperatorId();
                operatorToStats.compute(id, (k, v) -> v == null ? operatorStats : v.add(operatorStats));
            }
        }
    }
    StageStats stageStats = new StageStats(schedulingComplete.get(), getSplitDistribution.snapshot(), totalTasks, runningTasks, completedTasks, failedTasks, totalDrivers, queuedDrivers, runningDrivers, blockedDrivers, completedDrivers, cumulativeUserMemory, failedCumulativeUserMemory, succinctBytes(userMemoryReservation), succinctBytes(revocableMemoryReservation), succinctBytes(totalMemoryReservation), succinctBytes(peakUserMemoryReservation), succinctBytes(peakRevocableMemoryReservation), succinctDuration(totalScheduledTime, NANOSECONDS), succinctDuration(failedScheduledTime, NANOSECONDS), succinctDuration(totalCpuTime, NANOSECONDS), succinctDuration(failedCpuTime, NANOSECONDS), succinctDuration(totalBlockedTime, NANOSECONDS), fullyBlocked && runningTasks > 0, blockedReasons, succinctBytes(physicalInputDataSize), succinctBytes(failedPhysicalInputDataSize), physicalInputPositions, failedPhysicalInputPositions, succinctDuration(physicalInputReadTime, NANOSECONDS), succinctDuration(failedPhysicalInputReadTime, NANOSECONDS), succinctBytes(internalNetworkInputDataSize), succinctBytes(failedInternalNetworkInputDataSize), internalNetworkInputPositions, failedInternalNetworkInputPositions, succinctBytes(rawInputDataSize), succinctBytes(failedRawInputDataSize), rawInputPositions, failedRawInputPositions, succinctBytes(processedInputDataSize), succinctBytes(failedProcessedInputDataSize), processedInputPositions, failedProcessedInputPositions, succinctBytes(bufferedDataSize), succinctBytes(outputDataSize), succinctBytes(failedOutputDataSize), outputPositions, failedOutputPositions, succinctBytes(physicalWrittenDataSize), succinctBytes(failedPhysicalWrittenDataSize), new StageGcStatistics(stageId.getId(), totalTasks, fullGcTaskCount, minFullGcSec, maxFullGcSec, totalFullGcSec, (int) (1.0 * totalFullGcSec / fullGcCount)), ImmutableList.copyOf(operatorToStats.values()));
    ExecutionFailureInfo failureInfo = null;
    if (state == FAILED) {
        failureInfo = failureCause.get();
    }
    return new StageInfo(stageId, state, fragment, fragment.getPartitioning().isCoordinatorOnly(), fragment.getTypes(), stageStats, taskInfos, ImmutableList.of(), tables, failureInfo);
}