Examples with BlockingSendException edu.iu.dsc.tws.api.exceptions.net.BlockingSendException used on opensource projects

Example 1 with BlockingSendException

use of edu.iu.dsc.tws.api.exceptions.net.BlockingSendException in project twister2 by DSC-SPIDAL.

the class ExecutionPlanBuilder method build.

@Override
public ExecutionPlan build(Config cfg, ComputeGraph taskGraph, TaskSchedulePlan taskSchedule) {
    // we need to build the task plan
    LogicalPlan logicalPlan = TaskPlanBuilder.build(workerId, workerInfoList, taskSchedule, taskIdGenerator);
    ParallelOperationFactory opFactory = new ParallelOperationFactory(cfg, network, logicalPlan);
    Map<Integer, WorkerSchedulePlan> containersMap = taskSchedule.getContainersMap();
    WorkerSchedulePlan conPlan = containersMap.get(workerId);
    if (conPlan == null) {
        LOG.log(Level.INFO, "Cannot find worker in the task plan: " + workerId);
        return null;
    }
    ExecutionPlan execution = new ExecutionPlan();
    Set<TaskInstancePlan> instancePlan = conPlan.getTaskInstances();
    long tasksVersion = 0L;
    if (CheckpointingContext.isCheckpointingEnabled(cfg)) {
        Set<Integer> globalTasks = Collections.emptySet();
        if (workerId == 0) {
            globalTasks = containersMap.values().stream().flatMap(containerPlan -> containerPlan.getTaskInstances().stream()).filter(ip -> taskGraph.vertex(ip.getTaskName()).getTask() instanceof CheckpointableTask && !(taskGraph.vertex(ip.getTaskName()).getTask() instanceof CheckpointingSGatherSink)).map(TaskInstancePlan::getTaskId).collect(Collectors.toSet());
        }
        try {
            Checkpoint.FamilyInitializeResponse familyInitializeResponse = this.checkpointingClient.initFamily(workerId, containersMap.size(), taskGraph.getGraphName(), globalTasks);
            tasksVersion = familyInitializeResponse.getVersion();
        } catch (BlockingSendException e) {
            throw new RuntimeException("Failed to register tasks with Checkpoint Manager", e);
        }
        LOG.info("Tasks will start with version " + tasksVersion);
    }
    // for each task we are going to create the communications
    for (TaskInstancePlan ip : instancePlan) {
        Vertex v = taskGraph.vertex(ip.getTaskName());
        Map<String, Set<String>> inEdges = new HashMap<>();
        Map<String, String> outEdges = new HashMap<>();
        if (v == null) {
            throw new RuntimeException("Non-existing task scheduled: " + ip.getTaskName());
        }
        INode node = v.getTask();
        if (node instanceof ICompute || node instanceof ISource) {
            // lets get the communication
            Set<Edge> edges = taskGraph.outEdges(v);
            // now lets create the communication object
            for (Edge e : edges) {
                Vertex child = taskGraph.childOfTask(v, e.getName());
                // lets figure out the parents task id
                Set<Integer> srcTasks = taskIdGenerator.getTaskIds(v, ip.getTaskId());
                Set<Integer> tarTasks = taskIdGenerator.getTaskIds(child, getTaskIdOfTask(child.getName(), taskSchedule));
                Map<Integer, Integer> srcGlobalToIndex = taskIdGenerator.getGlobalTaskToIndex(v, ip.getTaskId());
                Map<Integer, Integer> tarGlobaToIndex = taskIdGenerator.getGlobalTaskToIndex(child, getTaskIdOfTask(child.getName(), taskSchedule));
                createCommunication(child, e, v, srcTasks, tarTasks, srcGlobalToIndex, tarGlobaToIndex);
                outEdges.put(e.getName(), child.getName());
            }
        }
        if (node instanceof ICompute) {
            // lets get the parent tasks
            Set<Edge> parentEdges = taskGraph.inEdges(v);
            for (Edge e : parentEdges) {
                Vertex parent = taskGraph.getParentOfTask(v, e.getName());
                // lets figure out the parents task id
                Set<Integer> srcTasks = taskIdGenerator.getTaskIds(parent, getTaskIdOfTask(parent.getName(), taskSchedule));
                Set<Integer> tarTasks = taskIdGenerator.getTaskIds(v, ip.getTaskId());
                Map<Integer, Integer> srcGlobalToIndex = taskIdGenerator.getGlobalTaskToIndex(parent, getTaskIdOfTask(parent.getName(), taskSchedule));
                Map<Integer, Integer> tarGlobalToIndex = taskIdGenerator.getGlobalTaskToIndex(v, ip.getTaskId());
                createCommunication(v, e, parent, srcTasks, tarTasks, srcGlobalToIndex, tarGlobalToIndex);
                // if we are a grouped edge, we have to use the group name
                String inEdge;
                if (e.getTargetEdge() == null) {
                    inEdge = e.getName();
                } else {
                    inEdge = e.getTargetEdge();
                }
                Set<String> parents = inEdges.get(inEdge);
                if (parents == null) {
                    parents = new HashSet<>();
                }
                parents.add(inEdge);
                inEdges.put(inEdge, parents);
            }
        }
        // lets create the instance
        INodeInstance iNodeInstance = createInstances(cfg, taskGraph.getGraphName(), ip, v, taskGraph.getOperationMode(), inEdges, outEdges, taskSchedule, tasksVersion);
        // add to execution
        execution.addNodes(v.getName(), taskIdGenerator.generateGlobalTaskId(ip.getTaskId(), ip.getTaskIndex()), iNodeInstance);
    }
    // now lets create the queues and start the execution
    for (Table.Cell<String, String, Communication> cell : parOpTable.cellSet()) {
        Communication c = cell.getValue();
        // lets create the communication
        OperationMode operationMode = taskGraph.getOperationMode();
        IParallelOperation op;
        assert c != null;
        c.build();
        if (c.getEdge().size() == 1) {
            op = opFactory.build(c.getEdge(0), c.getSourceTasks(), c.getTargetTasks(), operationMode, c.srcGlobalToIndex, c.tarGlobalToIndex);
        } else if (c.getEdge().size() > 1) {
            // just join op for now. Could change in the future
            // here the sources should be separated out for left and right edge
            Set<Integer> sourceTasks = c.getSourceTasks();
            Set<Integer> leftSources = new HashSet<>();
            Set<Integer> rightSources = new HashSet<>();
            if (!sourceTasks.isEmpty()) {
                // just to safely do .get() calls without isPresent()
                int minBin = (sourceTasks.stream().min(Integer::compareTo).get() / TaskIdGenerator.TASK_OFFSET) * TaskIdGenerator.TASK_OFFSET;
                for (Integer source : sourceTasks) {
                    if ((source / TaskIdGenerator.TASK_OFFSET) * TaskIdGenerator.TASK_OFFSET == minBin) {
                        leftSources.add(source);
                    } else {
                        rightSources.add(source);
                    }
                }
            }
            // now determine, which task is connected to which edge
            Edge leftEdge = c.getEdge(0);
            Edge rightEdge = c.getEdge(1);
            op = opFactory.build(leftEdge, rightEdge, leftSources, rightSources, c.getTargetTasks(), operationMode, c.srcGlobalToIndex, c.tarGlobalToIndex);
        } else {
            throw new RuntimeException("Cannot have communication with 0 edges");
        }
        // now lets check the sources and targets that are in this executor
        Set<Integer> sourcesOfThisWorker = intersectionOfTasks(conPlan, c.getSourceTasks());
        Set<Integer> targetsOfThisWorker = intersectionOfTasks(conPlan, c.getTargetTasks());
        // we use the target edge as the group name
        String targetEdge;
        if (c.getEdge().size() > 1) {
            targetEdge = c.getEdge(0).getTargetEdge();
        } else {
            targetEdge = c.getEdge(0).getName();
        }
        // so along with the operation mode, the windowing mode must be tested
        if (operationMode == OperationMode.STREAMING) {
            for (Integer i : sourcesOfThisWorker) {
                boolean found = false;
                // we can have multiple source tasks for an operation
                for (int sIndex = 0; sIndex < c.getSourceTask().size(); sIndex++) {
                    String sourceTask = c.getSourceTask().get(sIndex);
                    if (streamingTaskInstances.contains(sourceTask, i)) {
                        TaskStreamingInstance taskStreamingInstance = streamingTaskInstances.get(sourceTask, i);
                        taskStreamingInstance.registerOutParallelOperation(c.getEdge(sIndex).getName(), op);
                        op.registerSync(i, taskStreamingInstance);
                        found = true;
                    } else if (streamingSourceInstances.contains(sourceTask, i)) {
                        SourceStreamingInstance sourceStreamingInstance = streamingSourceInstances.get(sourceTask, i);
                        sourceStreamingInstance.registerOutParallelOperation(c.getEdge(sIndex).getName(), op);
                        found = true;
                    }
                    if (!found) {
                        throw new RuntimeException("Not found: " + c.getSourceTask());
                    }
                }
            }
            // we only have one target task always
            for (Integer i : targetsOfThisWorker) {
                if (streamingTaskInstances.contains(c.getTargetTask(), i)) {
                    TaskStreamingInstance taskStreamingInstance = streamingTaskInstances.get(c.getTargetTask(), i);
                    op.register(i, taskStreamingInstance.getInQueue());
                    taskStreamingInstance.registerInParallelOperation(targetEdge, op);
                    op.registerSync(i, taskStreamingInstance);
                } else {
                    throw new RuntimeException("Not found: " + c.getTargetTask());
                }
            }
            execution.addOps(op);
        }
        if (operationMode == OperationMode.BATCH) {
            for (Integer i : sourcesOfThisWorker) {
                boolean found = false;
                // we can have multiple source tasks for an operation
                for (int sIndex = 0; sIndex < c.getSourceTask().size(); sIndex++) {
                    String sourceTask = c.getSourceTask().get(sIndex);
                    if (batchTaskInstances.contains(sourceTask, i)) {
                        TaskBatchInstance taskBatchInstance = batchTaskInstances.get(sourceTask, i);
                        taskBatchInstance.registerOutParallelOperation(c.getEdge(sIndex).getName(), op);
                        found = true;
                    } else if (batchSourceInstances.contains(sourceTask, i)) {
                        SourceBatchInstance sourceBatchInstance = batchSourceInstances.get(sourceTask, i);
                        sourceBatchInstance.registerOutParallelOperation(c.getEdge(sIndex).getName(), op);
                        found = true;
                    }
                }
                if (!found) {
                    throw new RuntimeException("Not found: " + c.getSourceTask());
                }
            }
            for (Integer i : targetsOfThisWorker) {
                if (batchTaskInstances.contains(c.getTargetTask(), i)) {
                    TaskBatchInstance taskBatchInstance = batchTaskInstances.get(c.getTargetTask(), i);
                    op.register(i, taskBatchInstance.getInQueue());
                    taskBatchInstance.registerInParallelOperation(targetEdge, op);
                    op.registerSync(i, taskBatchInstance);
                } else {
                    throw new RuntimeException("Not found: " + c.getTargetTask());
                }
            }
            execution.addOps(op);
        }
    }
    return execution;
}

Example 2 with BlockingSendException

use of edu.iu.dsc.tws.api.exceptions.net.BlockingSendException in project twister2 by DSC-SPIDAL.

the class RRClient method sendRequestWaitResponse.

/**
 * return requestID and response message
 * throw an exception with the failure reason
 * @param message message
 * @param waitLimit waitlimit
 * @return request id
 */
public Tuple<RequestID, Message> sendRequestWaitResponse(Message message, long waitLimit) throws BlockingSendException {
    // if this method is already called and waiting for a response
    if (requestIdOfWaitedResponse != null) {
        throw new BlockingSendException(BlockingSendFailureReason.ALREADY_SENDING_ANOTHER_MESSAGE, "Already sending another message.", null);
    }
    synchronized (responseWaitObject) {
        responseReceived = false;
        errorWhenSending = false;
        RequestID requestID = sendRequest(message);
        requestIdOfWaitedResponse = requestID;
        if (requestIdOfWaitedResponse == null) {
            throw new BlockingSendException(BlockingSendFailureReason.ERROR_WHEN_TRYING_TO_SEND, "Problem when trying to send the message.", null);
        }
        try {
            responseWaitObject.wait(waitLimit);
            if (errorWhenSending) {
                throw new BlockingSendException(BlockingSendFailureReason.CONNECTION_LOST_WHEN_SENDING, "Connection lost when sending the message.", null);
            }
            if (!responseReceived) {
                throw new BlockingSendException(BlockingSendFailureReason.TIME_LIMIT_REACHED, "Response not received for Wait limit[" + waitLimit + "ms].", null);
            }
        } catch (InterruptedException e) {
            throw new BlockingSendException(BlockingSendFailureReason.EXCEPTION_WHEN_WAITING, "Exception when waiting the response.", e);
        }
        return new Tuple(requestID, waitedResponseMessage);
    }
}

Example 3 with BlockingSendException

use of edu.iu.dsc.tws.api.exceptions.net.BlockingSendException in project twister2 by DSC-SPIDAL.

the class JMWorkerController method sendBarrierRequest.

private void sendBarrierRequest(JobMasterAPI.BarrierType barrierType, long timeLimit) throws TimeoutException {
    JobMasterAPI.BarrierRequest barrierRequest = JobMasterAPI.BarrierRequest.newBuilder().setWorkerID(workerInfo.getWorkerID()).setBarrierType(barrierType).setTimeout(timeLimit).build();
    LOG.fine("Sending BarrierRequest message: \n" + barrierRequest.toString());
    try {
        // set the local wait time for the barrier response to (2 * timeLimit)
        // if the requested time limit is more than half of the long max value,
        // set it to the long max value
        long tl = timeLimit > Long.MAX_VALUE / 2 ? Long.MAX_VALUE : timeLimit * 2;
        Tuple<RequestID, Message> response = rrClient.sendRequestWaitResponse(barrierRequest, tl);
        JobMasterAPI.BarrierResponse barrierResponse = (JobMasterAPI.BarrierResponse) response.getValue();
        if (barrierResponse.getResult() == JobMasterAPI.BarrierResult.SUCCESS) {
            return;
        } else if (barrierResponse.getResult() == JobMasterAPI.BarrierResult.JOB_FAULTY) {
            throw new JobFaultyException("Job became faulty and Default Barrier failed.");
        } else if (barrierResponse.getResult() == JobMasterAPI.BarrierResult.TIMED_OUT) {
            throw new TimeoutException("Barrier timed out. Not all workers arrived at the barrier " + "on the time limit: " + timeLimit + "ms");
        }
    } catch (BlockingSendException e) {
        throw new TimeoutException("Not all workers arrived at the barrier on the time limit: " + timeLimit + "ms.", e);
    }
}

Example 4 with BlockingSendException

use of edu.iu.dsc.tws.api.exceptions.net.BlockingSendException in project twister2 by DSC-SPIDAL.

the class JMWorkerController method sendWorkerListRequest.

private boolean sendWorkerListRequest(ListWorkersRequest.RequestType requestType, long timeLimit) {
    ListWorkersRequest listRequest = ListWorkersRequest.newBuilder().setWorkerID(workerInfo.getWorkerID()).setRequestType(requestType).build();
    LOG.fine("Sending ListWorkers message to the master: \n" + listRequest);
    try {
        rrClient.sendRequestWaitResponse(listRequest, timeLimit);
        return true;
    } catch (BlockingSendException e) {
        LOG.log(Level.SEVERE, e.getMessage(), e);
        return false;
    }
}