Search in sources :

Example 11 with TaskCluster

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

the class ActivityClusterPlanner method buildConnectorPolicyAwareTaskClusters.

private TaskCluster[] buildConnectorPolicyAwareTaskClusters(ActivityCluster ac, Map<ActivityId, ActivityPlan> activityPlanMap, Map<TaskId, List<Pair<TaskId, ConnectorDescriptorId>>> taskConnectivity) {
    Map<TaskId, Set<TaskId>> taskClusterMap = new HashMap<>();
    for (ActivityId anId : ac.getActivityMap().keySet()) {
        ActivityPlan ap = activityPlanMap.get(anId);
        Task[] tasks = ap.getTasks();
        for (Task t : tasks) {
            Set<TaskId> cluster = new HashSet<>();
            TaskId tid = t.getTaskId();
            cluster.add(tid);
            taskClusterMap.put(tid, cluster);
        }
    }
    JobRun jobRun = executor.getJobRun();
    Map<ConnectorDescriptorId, IConnectorPolicy> connectorPolicies = jobRun.getConnectorPolicyMap();
    for (Map.Entry<TaskId, List<Pair<TaskId, ConnectorDescriptorId>>> e : taskConnectivity.entrySet()) {
        Set<TaskId> cluster = taskClusterMap.get(e.getKey());
        for (Pair<TaskId, ConnectorDescriptorId> p : e.getValue()) {
            IConnectorPolicy cPolicy = connectorPolicies.get(p.getRight());
            if (cPolicy.requiresProducerConsumerCoscheduling()) {
                cluster.add(p.getLeft());
            }
        }
    }
    /*
         * We compute the transitive closure of this (producer-consumer) relation to find the largest set of
         * tasks that need to be co-scheduled.
         */
    int counter = 0;
    TaskId[] ordinalList = new TaskId[taskClusterMap.size()];
    Map<TaskId, Integer> ordinalMap = new HashMap<>();
    for (TaskId tid : taskClusterMap.keySet()) {
        ordinalList[counter] = tid;
        ordinalMap.put(tid, counter);
        ++counter;
    }
    int n = ordinalList.length;
    BitSet[] paths = new BitSet[n];
    for (Map.Entry<TaskId, Set<TaskId>> e : taskClusterMap.entrySet()) {
        int i = ordinalMap.get(e.getKey());
        BitSet bsi = paths[i];
        if (bsi == null) {
            bsi = new BitSet(n);
            paths[i] = bsi;
        }
        for (TaskId ttid : e.getValue()) {
            int j = ordinalMap.get(ttid);
            paths[i].set(j);
            BitSet bsj = paths[j];
            if (bsj == null) {
                bsj = new BitSet(n);
                paths[j] = bsj;
            }
            bsj.set(i);
        }
    }
    for (int k = 0; k < n; ++k) {
        for (int i = paths[k].nextSetBit(0); i >= 0; i = paths[k].nextSetBit(i + 1)) {
            for (int j = paths[i].nextClearBit(0); j < n && j >= 0; j = paths[i].nextClearBit(j + 1)) {
                paths[i].set(j, paths[k].get(j));
                paths[j].set(i, paths[i].get(j));
            }
        }
    }
    BitSet pending = new BitSet(n);
    pending.set(0, n);
    List<List<TaskId>> clusters = new ArrayList<>();
    for (int i = pending.nextSetBit(0); i >= 0; i = pending.nextSetBit(i)) {
        List<TaskId> cluster = new ArrayList<>();
        for (int j = paths[i].nextSetBit(0); j >= 0; j = paths[i].nextSetBit(j + 1)) {
            cluster.add(ordinalList[j]);
            pending.clear(j);
        }
        clusters.add(cluster);
    }
    List<TaskCluster> tcSet = new ArrayList<>();
    counter = 0;
    for (List<TaskId> cluster : clusters) {
        List<Task> taskStates = new ArrayList<>();
        for (TaskId tid : cluster) {
            taskStates.add(activityPlanMap.get(tid.getActivityId()).getTasks()[tid.getPartition()]);
        }
        TaskCluster tc = new TaskCluster(new TaskClusterId(ac.getId(), counter++), ac, taskStates.toArray(new Task[taskStates.size()]));
        tcSet.add(tc);
        for (TaskId tid : cluster) {
            activityPlanMap.get(tid.getActivityId()).getTasks()[tid.getPartition()].setTaskCluster(tc);
        }
    }
    return tcSet.toArray(new TaskCluster[tcSet.size()]);
}
Also used : Task(org.apache.hyracks.control.cc.job.Task) TaskId(org.apache.hyracks.api.dataflow.TaskId) HashSet(java.util.HashSet) Set(java.util.Set) BitSet(java.util.BitSet) HashMap(java.util.HashMap) ActivityId(org.apache.hyracks.api.dataflow.ActivityId) ConnectorDescriptorId(org.apache.hyracks.api.dataflow.ConnectorDescriptorId) ArrayList(java.util.ArrayList) ArrayList(java.util.ArrayList) List(java.util.List) HashSet(java.util.HashSet) IConnectorPolicy(org.apache.hyracks.api.dataflow.connectors.IConnectorPolicy) ActivityPlan(org.apache.hyracks.control.cc.job.ActivityPlan) BitSet(java.util.BitSet) TaskCluster(org.apache.hyracks.control.cc.job.TaskCluster) HashMap(java.util.HashMap) Map(java.util.Map) TaskClusterId(org.apache.hyracks.control.cc.job.TaskClusterId) JobRun(org.apache.hyracks.control.cc.job.JobRun)

Example 12 with TaskCluster

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

the class ActivityClusterPlanner method computeTaskClusters.

private TaskCluster[] computeTaskClusters(ActivityCluster ac, JobRun jobRun, Map<ActivityId, ActivityPlan> activityPlanMap) {
    Set<ActivityId> activities = ac.getActivityMap().keySet();
    Map<TaskId, List<Pair<TaskId, ConnectorDescriptorId>>> taskConnectivity = computeTaskConnectivity(jobRun, activityPlanMap, activities);
    TaskCluster[] taskClusters = ac.getActivityClusterGraph().isUseConnectorPolicyForScheduling() ? buildConnectorPolicyAwareTaskClusters(ac, activityPlanMap, taskConnectivity) : buildConnectorPolicyUnawareTaskClusters(ac, activityPlanMap);
    for (TaskCluster tc : taskClusters) {
        Set<TaskCluster> tcDependencyTaskClusters = tc.getDependencyTaskClusters();
        for (Task ts : tc.getTasks()) {
            TaskId tid = ts.getTaskId();
            List<Pair<TaskId, ConnectorDescriptorId>> cInfoList = taskConnectivity.get(tid);
            if (cInfoList != null) {
                for (Pair<TaskId, ConnectorDescriptorId> p : cInfoList) {
                    Task targetTS = activityPlanMap.get(p.getLeft().getActivityId()).getTasks()[p.getLeft().getPartition()];
                    TaskCluster targetTC = targetTS.getTaskCluster();
                    if (targetTC != tc) {
                        ConnectorDescriptorId cdId = p.getRight();
                        PartitionId pid = new PartitionId(jobRun.getJobId(), cdId, tid.getPartition(), p.getLeft().getPartition());
                        tc.getProducedPartitions().add(pid);
                        targetTC.getRequiredPartitions().add(pid);
                        partitionProducingTaskClusterMap.put(pid, tc);
                    }
                }
            }
            for (TaskId dTid : ts.getDependencies()) {
                TaskCluster dTC = getTaskCluster(dTid);
                dTC.getDependentTaskClusters().add(tc);
                tcDependencyTaskClusters.add(dTC);
            }
        }
    }
    return taskClusters;
}
Also used : Task(org.apache.hyracks.control.cc.job.Task) TaskId(org.apache.hyracks.api.dataflow.TaskId) ActivityId(org.apache.hyracks.api.dataflow.ActivityId) ConnectorDescriptorId(org.apache.hyracks.api.dataflow.ConnectorDescriptorId) PartitionId(org.apache.hyracks.api.partitions.PartitionId) TaskCluster(org.apache.hyracks.control.cc.job.TaskCluster) ArrayList(java.util.ArrayList) List(java.util.List) Pair(org.apache.commons.lang3.tuple.Pair)

Example 13 with TaskCluster

use of org.apache.hyracks.control.cc.job.TaskCluster 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);
                    }
                }
            }
        }
    }
}
Also used : Task(org.apache.hyracks.control.cc.job.Task) TaskId(org.apache.hyracks.api.dataflow.TaskId) TaskClusterAttempt(org.apache.hyracks.control.cc.job.TaskClusterAttempt) ActivityId(org.apache.hyracks.api.dataflow.ActivityId) ActivityPlan(org.apache.hyracks.control.cc.job.ActivityPlan) IJobManager(org.apache.hyracks.control.cc.job.IJobManager) ActivityCluster(org.apache.hyracks.api.job.ActivityCluster) TaskCluster(org.apache.hyracks.control.cc.job.TaskCluster) TaskAttempt(org.apache.hyracks.control.cc.job.TaskAttempt) JobRun(org.apache.hyracks.control.cc.job.JobRun)

Example 14 with TaskCluster

use of org.apache.hyracks.control.cc.job.TaskCluster 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();
    }
}
Also used : TaskClusterAttempt(org.apache.hyracks.control.cc.job.TaskClusterAttempt) TaskAttemptId(org.apache.hyracks.api.dataflow.TaskAttemptId) TaskCluster(org.apache.hyracks.control.cc.job.TaskCluster) TaskAttempt(org.apache.hyracks.control.cc.job.TaskAttempt)

Aggregations

TaskCluster (org.apache.hyracks.control.cc.job.TaskCluster)14 TaskClusterAttempt (org.apache.hyracks.control.cc.job.TaskClusterAttempt)7 ArrayList (java.util.ArrayList)5 ActivityId (org.apache.hyracks.api.dataflow.ActivityId)5 HashSet (java.util.HashSet)4 List (java.util.List)4 ActivityPlan (org.apache.hyracks.control.cc.job.ActivityPlan)4 Task (org.apache.hyracks.control.cc.job.Task)4 TaskAttempt (org.apache.hyracks.control.cc.job.TaskAttempt)4 HashMap (java.util.HashMap)3 Map (java.util.Map)3 ConnectorDescriptorId (org.apache.hyracks.api.dataflow.ConnectorDescriptorId)3 TaskAttemptId (org.apache.hyracks.api.dataflow.TaskAttemptId)3 TaskId (org.apache.hyracks.api.dataflow.TaskId)3 ActivityCluster (org.apache.hyracks.api.job.ActivityCluster)3 JobRun (org.apache.hyracks.control.cc.job.JobRun)3 IConnectorPolicy (org.apache.hyracks.api.dataflow.connectors.IConnectorPolicy)2 HyracksException (org.apache.hyracks.api.exceptions.HyracksException)2 PartitionId (org.apache.hyracks.api.partitions.PartitionId)2 ActivityClusterPlan (org.apache.hyracks.control.cc.job.ActivityClusterPlan)2