Search in sources :

Example 1 with TaskClusterId

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

the class ActivityClusterPlanner method buildConnectorPolicyUnawareTaskClusters.

private TaskCluster[] buildConnectorPolicyUnawareTaskClusters(ActivityCluster ac, Map<ActivityId, ActivityPlan> activityPlanMap) {
    List<Task> taskStates = new ArrayList<>();
    for (ActivityId anId : ac.getActivityMap().keySet()) {
        ActivityPlan ap = activityPlanMap.get(anId);
        Task[] tasks = ap.getTasks();
        taskStates.addAll(Arrays.asList(tasks));
    }
    TaskCluster tc = new TaskCluster(new TaskClusterId(ac.getId(), 0), ac, taskStates.toArray(new Task[taskStates.size()]));
    for (Task t : tc.getTasks()) {
        t.setTaskCluster(tc);
    }
    return new TaskCluster[] { tc };
}
Also used : Task(org.apache.hyracks.control.cc.job.Task) ActivityId(org.apache.hyracks.api.dataflow.ActivityId) ActivityPlan(org.apache.hyracks.control.cc.job.ActivityPlan) ArrayList(java.util.ArrayList) TaskCluster(org.apache.hyracks.control.cc.job.TaskCluster) TaskClusterId(org.apache.hyracks.control.cc.job.TaskClusterId)

Example 2 with TaskClusterId

use of org.apache.hyracks.control.cc.job.TaskClusterId 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)

Aggregations

ArrayList (java.util.ArrayList)2 ActivityId (org.apache.hyracks.api.dataflow.ActivityId)2 ActivityPlan (org.apache.hyracks.control.cc.job.ActivityPlan)2 Task (org.apache.hyracks.control.cc.job.Task)2 TaskCluster (org.apache.hyracks.control.cc.job.TaskCluster)2 TaskClusterId (org.apache.hyracks.control.cc.job.TaskClusterId)2 BitSet (java.util.BitSet)1 HashMap (java.util.HashMap)1 HashSet (java.util.HashSet)1 List (java.util.List)1 Map (java.util.Map)1 Set (java.util.Set)1 ConnectorDescriptorId (org.apache.hyracks.api.dataflow.ConnectorDescriptorId)1 TaskId (org.apache.hyracks.api.dataflow.TaskId)1 IConnectorPolicy (org.apache.hyracks.api.dataflow.connectors.IConnectorPolicy)1 JobRun (org.apache.hyracks.control.cc.job.JobRun)1