Examples with TaskClusterAttempt org.apache.hyracks.control.cc.job.TaskClusterAttempt used on opensource projects

Example 6 with TaskClusterAttempt

use of org.apache.hyracks.control.cc.job.TaskClusterAttempt in project asterixdb by apache.

the class JobExecutor method findRunnableTaskClusterRoots.

private void findRunnableTaskClusterRoots(Set<TaskCluster> frontier, ActivityCluster candidate) throws HyracksException {
    boolean depsComplete = true;
    for (ActivityCluster depAC : candidate.getDependencies()) {
        if (!isPlanned(depAC)) {
            depsComplete = false;
            findRunnableTaskClusterRoots(frontier, depAC);
        } else {
            boolean tcRootsComplete = true;
            for (TaskCluster tc : getActivityClusterPlan(depAC).getTaskClusters()) {
                if (!tc.getProducedPartitions().isEmpty()) {
                    continue;
                }
                TaskClusterAttempt tca = findLastTaskClusterAttempt(tc);
                if (tca == null || tca.getStatus() != TaskClusterAttempt.TaskClusterStatus.COMPLETED) {
                    tcRootsComplete = false;
                    break;
                }
            }
            if (!tcRootsComplete) {
                depsComplete = false;
                findRunnableTaskClusterRoots(frontier, depAC);
            }
        }
    }
    if (!depsComplete) {
        return;
    }
    if (!isPlanned(candidate)) {
        ActivityClusterPlanner acp = new ActivityClusterPlanner(this);
        ActivityClusterPlan acPlan = acp.planActivityCluster(candidate);
        jobRun.getActivityClusterPlanMap().put(candidate.getId(), acPlan);
        partitionProducingTaskClusterMap.putAll(acp.getPartitionProducingTaskClusterMap());
    }
    for (TaskCluster tc : getActivityClusterPlan(candidate).getTaskClusters()) {
        if (!tc.getProducedPartitions().isEmpty()) {
            continue;
        }
        TaskClusterAttempt tca = findLastTaskClusterAttempt(tc);
        if (tca == null || tca.getStatus() != TaskClusterAttempt.TaskClusterStatus.COMPLETED) {
            frontier.add(tc);
        }
    }
}

Example 7 with TaskClusterAttempt

use of org.apache.hyracks.control.cc.job.TaskClusterAttempt in project asterixdb by apache.

the class JobExecutor method assignRunnabilityRank.

/*
     * Runnability rank has the following semantics
     * Runnability(Runnable TaskCluster depending on completed TaskClusters) = {RUNNABLE, 0}
     * Runnability(Runnable TaskCluster) = max(Rank(Dependent TaskClusters)) + 1
     * Runnability(Non-schedulable TaskCluster) = {NOT_RUNNABLE, _}
     */
private Runnability assignRunnabilityRank(TaskCluster goal, Map<TaskCluster, Runnability> runnabilityMap) {
    if (LOGGER.isLoggable(Level.FINE)) {
        LOGGER.fine("Computing runnability: " + goal);
    }
    if (runnabilityMap.containsKey(goal)) {
        return runnabilityMap.get(goal);
    }
    TaskClusterAttempt lastAttempt = findLastTaskClusterAttempt(goal);
    if (lastAttempt != null) {
        if (LOGGER.isLoggable(Level.FINE)) {
            LOGGER.fine("Last Attempt Status: " + lastAttempt.getStatus());
        }
        if (lastAttempt.getStatus() == TaskClusterAttempt.TaskClusterStatus.COMPLETED) {
            Runnability runnability = new Runnability(Runnability.Tag.COMPLETED, Integer.MIN_VALUE);
            runnabilityMap.put(goal, runnability);
            return runnability;
        }
        if (lastAttempt.getStatus() == TaskClusterAttempt.TaskClusterStatus.RUNNING) {
            Runnability runnability = new Runnability(Runnability.Tag.RUNNING, Integer.MIN_VALUE);
            runnabilityMap.put(goal, runnability);
            return runnability;
        }
    }
    Map<ConnectorDescriptorId, IConnectorPolicy> connectorPolicyMap = jobRun.getConnectorPolicyMap();
    PartitionMatchMaker pmm = jobRun.getPartitionMatchMaker();
    Runnability aggregateRunnability = new Runnability(Runnability.Tag.RUNNABLE, 0);
    for (PartitionId pid : goal.getRequiredPartitions()) {
        if (LOGGER.isLoggable(Level.FINE)) {
            LOGGER.fine("Inspecting required partition: " + pid);
        }
        Runnability runnability;
        ConnectorDescriptorId cdId = pid.getConnectorDescriptorId();
        IConnectorPolicy cPolicy = connectorPolicyMap.get(cdId);
        PartitionState maxState = pmm.getMaximumAvailableState(pid);
        if (LOGGER.isLoggable(Level.FINE)) {
            LOGGER.fine("Policy: " + cPolicy + " maxState: " + maxState);
        }
        if (PartitionState.COMMITTED.equals(maxState)) {
            runnability = new Runnability(Runnability.Tag.RUNNABLE, 0);
        } else if (PartitionState.STARTED.equals(maxState) && !cPolicy.consumerWaitsForProducerToFinish()) {
            runnability = new Runnability(Runnability.Tag.RUNNABLE, 1);
        } else {
            runnability = assignRunnabilityRank(partitionProducingTaskClusterMap.get(pid), runnabilityMap);
            switch(runnability.getTag()) {
                case RUNNABLE:
                    if (cPolicy.consumerWaitsForProducerToFinish()) {
                        runnability = new Runnability(Runnability.Tag.NOT_RUNNABLE, Integer.MAX_VALUE);
                    } else {
                        runnability = new Runnability(Runnability.Tag.RUNNABLE, runnability.getPriority() + 1);
                    }
                    break;
                case NOT_RUNNABLE:
                    break;
                case RUNNING:
                    if (cPolicy.consumerWaitsForProducerToFinish()) {
                        runnability = new Runnability(Runnability.Tag.NOT_RUNNABLE, Integer.MAX_VALUE);
                    } else {
                        runnability = new Runnability(Runnability.Tag.RUNNABLE, 1);
                    }
                    break;
                default:
                    break;
            }
        }
        aggregateRunnability = Runnability.getWorstCase(aggregateRunnability, runnability);
        if (aggregateRunnability.getTag() == Runnability.Tag.NOT_RUNNABLE) {
            // already not runnable -- cannot get better. bail.
            break;
        }
        if (LOGGER.isLoggable(Level.FINE)) {
            LOGGER.fine("aggregateRunnability: " + aggregateRunnability);
        }
    }
    runnabilityMap.put(goal, aggregateRunnability);
    return aggregateRunnability;
}

Example 8 with TaskClusterAttempt

use of org.apache.hyracks.control.cc.job.TaskClusterAttempt in project asterixdb by apache.

the class AbstractTaskLifecycleWork method runWork.

@Override
public final void runWork() {
    IJobManager jobManager = ccs.getJobManager();
    JobRun run = jobManager.get(jobId);
    if (run != null) {
        TaskId tid = taId.getTaskId();
        Map<ActivityId, ActivityCluster> activityClusterMap = run.getActivityClusterGraph().getActivityMap();
        ActivityCluster ac = activityClusterMap.get(tid.getActivityId());
        if (ac != null) {
            Map<ActivityId, ActivityPlan> taskStateMap = run.getActivityClusterPlanMap().get(ac.getId()).getActivityPlanMap();
            Task[] taskStates = taskStateMap.get(tid.getActivityId()).getTasks();
            if (taskStates != null && taskStates.length > tid.getPartition()) {
                Task ts = taskStates[tid.getPartition()];
                TaskCluster tc = ts.getTaskCluster();
                List<TaskClusterAttempt> taskClusterAttempts = tc.getAttempts();
                if (taskClusterAttempts != null && taskClusterAttempts.size() > taId.getAttempt()) {
                    TaskClusterAttempt tca = taskClusterAttempts.get(taId.getAttempt());
                    TaskAttempt ta = tca.getTaskAttempts().get(tid);
                    if (ta != null) {
                        performEvent(ta);
                    }
                }
            }
        }
    }
}

Example 9 with TaskClusterAttempt

use of org.apache.hyracks.control.cc.job.TaskClusterAttempt in project asterixdb by apache.

the class JobExecutor method assignTaskLocations.

private void assignTaskLocations(TaskCluster tc, Map<String, List<TaskAttemptDescriptor>> taskAttemptMap) throws HyracksException {
    ActivityClusterGraph acg = jobRun.getActivityClusterGraph();
    Task[] tasks = tc.getTasks();
    List<TaskClusterAttempt> tcAttempts = tc.getAttempts();
    int attempts = tcAttempts.size();
    TaskClusterAttempt tcAttempt = new TaskClusterAttempt(tc, attempts);
    Map<TaskId, TaskAttempt> taskAttempts = new HashMap<>();
    Map<TaskId, LValueConstraintExpression> locationMap = new HashMap<>();
    for (int i = 0; i < tasks.length; ++i) {
        Task ts = tasks[i];
        TaskId tid = ts.getTaskId();
        TaskAttempt taskAttempt = new TaskAttempt(tcAttempt, new TaskAttemptId(new TaskId(tid.getActivityId(), tid.getPartition()), attempts), ts);
        taskAttempt.setStatus(TaskAttempt.TaskStatus.INITIALIZED, null);
        locationMap.put(tid, new PartitionLocationExpression(tid.getActivityId().getOperatorDescriptorId(), tid.getPartition()));
        taskAttempts.put(tid, taskAttempt);
    }
    tcAttempt.setTaskAttempts(taskAttempts);
    solver.solve(locationMap.values());
    for (int i = 0; i < tasks.length; ++i) {
        Task ts = tasks[i];
        TaskId tid = ts.getTaskId();
        TaskAttempt taskAttempt = taskAttempts.get(tid);
        String nodeId = assignLocation(acg, locationMap, tid, taskAttempt);
        taskAttempt.setNodeId(nodeId);
        taskAttempt.setStatus(TaskAttempt.TaskStatus.RUNNING, null);
        taskAttempt.setStartTime(System.currentTimeMillis());
        List<TaskAttemptDescriptor> tads = taskAttemptMap.get(nodeId);
        if (tads == null) {
            tads = new ArrayList<>();
            taskAttemptMap.put(nodeId, tads);
        }
        OperatorDescriptorId opId = tid.getActivityId().getOperatorDescriptorId();
        jobRun.registerOperatorLocation(opId, tid.getPartition(), nodeId);
        ActivityPartitionDetails apd = ts.getActivityPlan().getActivityPartitionDetails();
        TaskAttemptDescriptor tad = new TaskAttemptDescriptor(taskAttempt.getTaskAttemptId(), apd.getPartitionCount(), apd.getInputPartitionCounts(), apd.getOutputPartitionCounts());
        tads.add(tad);
    }
    tcAttempt.initializePendingTaskCounter();
    tcAttempts.add(tcAttempt);
    /**
         * Improvement for reducing master/slave message communications, for each TaskAttemptDescriptor,
         * we set the NetworkAddress[][] partitionLocations, in which each row is for an incoming connector descriptor
         * and each column is for an input channel of the connector.
         */
    INodeManager nodeManager = ccs.getNodeManager();
    for (Map.Entry<String, List<TaskAttemptDescriptor>> e : taskAttemptMap.entrySet()) {
        List<TaskAttemptDescriptor> tads = e.getValue();
        for (TaskAttemptDescriptor tad : tads) {
            TaskAttemptId taid = tad.getTaskAttemptId();
            int attempt = taid.getAttempt();
            TaskId tid = taid.getTaskId();
            ActivityId aid = tid.getActivityId();
            List<IConnectorDescriptor> inConnectors = acg.getActivityInputs(aid);
            int[] inPartitionCounts = tad.getInputPartitionCounts();
            if (inPartitionCounts == null) {
                continue;
            }
            NetworkAddress[][] partitionLocations = new NetworkAddress[inPartitionCounts.length][];
            for (int i = 0; i < inPartitionCounts.length; ++i) {
                ConnectorDescriptorId cdId = inConnectors.get(i).getConnectorId();
                IConnectorPolicy policy = jobRun.getConnectorPolicyMap().get(cdId);
                /**
                     * carry sender location information into a task
                     * when it is not the case that it is an re-attempt and the send-side
                     * is materialized blocking.
                     */
                if (attempt > 0 && policy.materializeOnSendSide() && policy.consumerWaitsForProducerToFinish()) {
                    continue;
                }
                ActivityId producerAid = acg.getProducerActivity(cdId);
                partitionLocations[i] = new NetworkAddress[inPartitionCounts[i]];
                for (int j = 0; j < inPartitionCounts[i]; ++j) {
                    TaskId producerTaskId = new TaskId(producerAid, j);
                    String nodeId = findTaskLocation(producerTaskId);
                    partitionLocations[i][j] = nodeManager.getNodeControllerState(nodeId).getDataPort();
                }
            }
            tad.setInputPartitionLocations(partitionLocations);
        }
    }
    tcAttempt.setStatus(TaskClusterAttempt.TaskClusterStatus.RUNNING);
    tcAttempt.setStartTime(System.currentTimeMillis());
    inProgressTaskClusters.add(tc);
}

Example 10 with TaskClusterAttempt

use of org.apache.hyracks.control.cc.job.TaskClusterAttempt in project asterixdb by apache.

the class JobExecutor method notifyTaskComplete.

public void notifyTaskComplete(TaskAttempt ta) throws HyracksException {
    TaskAttemptId taId = ta.getTaskAttemptId();
    TaskCluster tc = ta.getTask().getTaskCluster();
    TaskClusterAttempt lastAttempt = findLastTaskClusterAttempt(tc);
    if (lastAttempt == null || taId.getAttempt() != lastAttempt.getAttempt()) {
        LOGGER.warning("Ignoring task complete notification: " + taId + " -- Current last attempt = " + lastAttempt);
        return;
    }
    TaskAttempt.TaskStatus taStatus = ta.getStatus();
    if (taStatus != TaskAttempt.TaskStatus.RUNNING) {
        LOGGER.warning("Spurious task complete notification: " + taId + " Current state = " + taStatus);
        return;
    }
    ta.setStatus(TaskAttempt.TaskStatus.COMPLETED, null);
    ta.setEndTime(System.currentTimeMillis());
    if (lastAttempt.decrementPendingTasksCounter() == 0) {
        lastAttempt.setStatus(TaskClusterAttempt.TaskClusterStatus.COMPLETED);
        lastAttempt.setEndTime(System.currentTimeMillis());
        inProgressTaskClusters.remove(tc);
        startRunnableActivityClusters();
    }
}