Examples with BatchTask org.dkpro.lab.task.BatchTask used on opensource projects

Example 1 with BatchTask

use of org.dkpro.lab.task.BatchTask in project dkpro-lab by dkpro.

the class MultiThreadBatchTaskEngine method executeConfiguration.

@Override
protected void executeConfiguration(BatchTask aConfiguration, TaskContext aContext, Map<String, Object> aConfig, Set<String> aExecutedSubtasks) throws ExecutionException, LifeCycleException {
    if (log.isTraceEnabled()) {
        // Show all subtasks executed so far
        for (String est : aExecutedSubtasks) {
            log.trace("-- Already executed: " + est);
        }
    }
    // Set up initial scope used by sub-batch-tasks using the inherited scope. The scope is
    // extended as the subtasks of this batch are executed with the present configuration.
    // FIXME: That means that sub-batch-tasks in two different configurations cannot see
    // each other. Is that intended? Mind that the "executedSubtasks" set is intentionally
    // maintained *across* configurations, so maybe the scope should also be maintained
    // *across* configurations? - REC 2014-06-15
    Set<String> scope = new HashSet<>();
    if (aConfiguration.getScope() != null) {
        scope.addAll(aConfiguration.getScope());
    }
    // Configure subtasks
    for (Task task : aConfiguration.getTasks()) {
        // Now the setup is complete
        aContext.getLifeCycleManager().configure(aContext, task, aConfig);
    }
    Queue<Task> queue = new LinkedList<>(aConfiguration.getTasks());
    // keeps track of the execution threads;
    // TODO MW: do we really need this or can we work with the futures list only?
    Map<Task, ExecutionThread> threads = new HashMap<>();
    // keeps track of submitted Futures and their associated tasks
    Map<Future<?>, Task> futures = new HashMap<Future<?>, Task>();
    // will be instantiated with all exceptions from current loop
    ConcurrentMap<Task, Throwable> exceptionsFromLastLoop = null;
    ConcurrentMap<Task, Throwable> exceptionsFromCurrentLoop = new ConcurrentHashMap<>();
    int outerLoopCounter = 0;
    // main loop
    do {
        outerLoopCounter++;
        threads.clear();
        futures.clear();
        ExecutorService executor = Executors.newFixedThreadPool(maxThreads);
        // set the exceptions from the last loop
        exceptionsFromLastLoop = new ConcurrentHashMap<>(exceptionsFromCurrentLoop);
        // Fix MW: Clear exceptionsFromCurrentLoop; otherwise the loop with run at most twice.
        exceptionsFromCurrentLoop.clear();
        // process all tasks from the queue
        while (!queue.isEmpty()) {
            Task task = queue.poll();
            TaskContextMetadata execution = getExistingExecution(aConfiguration, aContext, task, aConfig, aExecutedSubtasks);
            // does already exist ...
            if (execution == null) {
                // ... otherwise execute it with the present configuration
                log.info("Executing task [" + task.getType() + "]");
                // set scope here so that the inherited scopes are considered
                if (task instanceof BatchTask) {
                    ((BatchTask) task).setScope(scope);
                }
                ExecutionThread thread = new ExecutionThread(aContext, task, aConfig, aExecutedSubtasks);
                threads.put(task, thread);
                futures.put(executor.submit(thread), task);
            } else {
                log.debug("Using existing execution [" + execution.getId() + "]");
                // Record new/existing execution
                aExecutedSubtasks.add(execution.getId());
                scope.add(execution.getId());
            }
        }
        // try and get results from all futures to check for failed executions
        for (Map.Entry<Future<?>, Task> entry : futures.entrySet()) {
            try {
                entry.getKey().get();
            } catch (java.util.concurrent.ExecutionException ex) {
                Task task = entry.getValue();
                // TODO MW: add a retry-counter here to prevent endless loops?
                log.info("Task exec failed for [" + task.getType() + "]");
                // record the failed task, so that it can be re-added to the queue
                exceptionsFromCurrentLoop.put(task, ex);
            } catch (InterruptedException ex) {
                // thread interrupted, exit
                throw new RuntimeException(ex);
            }
        }
        log.debug("Calling shutdown");
        executor.shutdown();
        log.debug("All threads finished");
        // collect the results
        for (Map.Entry<Task, ExecutionThread> entry : threads.entrySet()) {
            Task task = entry.getKey();
            ExecutionThread thread = entry.getValue();
            TaskContextMetadata execution = thread.getTaskContextMetadata();
            // probably failed
            if (execution == null) {
                Throwable exception = exceptionsFromCurrentLoop.get(task);
                if (!(exception instanceof UnresolvedImportException) && !(exception instanceof java.util.concurrent.ExecutionException)) {
                    throw new RuntimeException(exception);
                }
                exceptionsFromCurrentLoop.put(task, exception);
                // re-add to the queue
                queue.add(task);
            } else {
                // Record new/existing execution
                aExecutedSubtasks.add(execution.getId());
                scope.add(execution.getId());
            }
        }
    } while (// finish if the same tasks failed again
    !exceptionsFromCurrentLoop.keySet().equals(exceptionsFromLastLoop.keySet()));
    if (!exceptionsFromCurrentLoop.isEmpty()) {
        // collect all details
        StringBuilder details = new StringBuilder();
        for (Throwable throwable : exceptionsFromCurrentLoop.values()) {
            details.append("\n -");
            details.append(throwable.getMessage());
        }
        // we re-throw the first exception
        Throwable next = exceptionsFromCurrentLoop.values().iterator().next();
        if (next instanceof RuntimeException) {
            throw (RuntimeException) next;
        }
        // otherwise wrap it
        throw new RuntimeException(details.toString(), next);
    }
    log.info("MultiThreadBatchTask completed successfully. Total number of outer loop runs: " + outerLoopCounter);
}

Example 2 with BatchTask

use of org.dkpro.lab.task.BatchTask in project dkpro-lab by dkpro.

the class BatchTaskEngine method executeConfiguration.

/**
 * Locate the latest task execution compatible with the given task configuration.
 *
 * @param aContext
 *            the context of the current batch task.
 * @param aConfig
 *            the current parameter configuration.
 * @param aExecutedSubtasks
 *            already executed subtasks.
 */
protected void executeConfiguration(BatchTask aConfiguration, TaskContext aContext, Map<String, Object> aConfig, Set<String> aExecutedSubtasks) throws ExecutionException, LifeCycleException {
    if (log.isTraceEnabled()) {
        // Show all subtasks executed so far
        for (String est : aExecutedSubtasks) {
            log.trace("-- Already executed: " + est);
        }
    }
    // Set up initial scope used by sub-batch-tasks using the inherited scope. The scope is
    // extended as the subtasks of this batch are executed with the present configuration.
    // FIXME: That means that sub-batch-tasks in two different configurations cannot see
    // each other. Is that intended? Mind that the "executedSubtasks" set is intentionally
    // maintained *across* configurations, so maybe the scope should also be maintained
    // *across* configurations? - REC 2014-06-15
    Set<String> scope = new HashSet<String>();
    if (aConfiguration.getScope() != null) {
        scope.addAll(aConfiguration.getScope());
    }
    // Configure subtasks
    for (Task task : aConfiguration.getTasks()) {
        aContext.getLifeCycleManager().configure(aContext, task, aConfig);
    }
    Queue<Task> queue = new LinkedList<Task>(aConfiguration.getTasks());
    Set<Task> loopDetection = new HashSet<Task>();
    List<UnresolvedImportException> deferralReasons = new ArrayList<UnresolvedImportException>();
    while (!queue.isEmpty()) {
        Task task = queue.poll();
        try {
            // Check if a subtask execution compatible with the present configuration has
            // does already exist ...
            TaskContextMetadata execution = getExistingExecution(aConfiguration, aContext, task, aConfig, aExecutedSubtasks);
            if (execution == null) {
                // ... otherwise execute it with the present configuration
                log.info("Executing task [" + task.getType() + "]");
                // set scope here so that tasks added to scope in this loop are considered
                if (task instanceof BatchTask) {
                    ((BatchTask) task).setScope(scope);
                }
                execution = runNewExecution(aContext, task, aConfig, aExecutedSubtasks);
            } else {
                log.debug("Using existing execution [" + execution.getId() + "]");
            }
            // Record new/existing execution
            aExecutedSubtasks.add(execution.getId());
            scope.add(execution.getId());
            loopDetection.clear();
            deferralReasons.clear();
        } catch (UnresolvedImportException e) {
            // Add task back to queue
            log.debug("Deferring execution of task [" + task.getType() + "]: " + e.getMessage());
            queue.add(task);
            // Detect endless loop
            if (loopDetection.contains(task)) {
                StringBuilder details = new StringBuilder();
                for (UnresolvedImportException r : deferralReasons) {
                    details.append("\n -");
                    details.append(r.getMessage());
                }
                // needs to be executed first
                throw new UnresolvedImportException(e, details.toString());
            }
            // Record failed execution
            loopDetection.add(task);
            deferralReasons.add(e);
        }
    }
}

Example 3 with BatchTask

use of org.dkpro.lab.task.BatchTask in project dkpro-lab by dkpro.

the class BatchTaskEngine method run.

@Override
public String run(Task aConfiguration) throws ExecutionException, LifeCycleException {
    if (!(aConfiguration instanceof BatchTask)) {
        throw new ExecutionException("This engine can only execute [" + BatchTask.class.getName() + "]");
    }
    // Create persistence service for injection into analysis components
    TaskContext ctx = null;
    try {
        ctx = contextFactory.createContext(aConfiguration);
        // Now the setup is complete
        ctx.getLifeCycleManager().initialize(ctx, aConfiguration);
        // Start recording
        ctx.getLifeCycleManager().begin(ctx, aConfiguration);
        try {
            BatchTask cfg = (BatchTask) aConfiguration;
            ParameterSpace parameterSpace = cfg.getParameterSpace();
            // Try to calculate the parameter space size.
            int estimatedSize = 1;
            for (Dimension<?> d : parameterSpace.getDimensions()) {
                if (d instanceof FixedSizeDimension) {
                    FixedSizeDimension fsd = (FixedSizeDimension) d;
                    if (fsd.size() > 0) {
                        estimatedSize *= fsd.size();
                    }
                }
            }
            // A subtask execution may apply to multiple parameter space coordinates!
            Set<String> executedSubtasks = new LinkedHashSet<String>();
            ProgressMeter progress = new ProgressMeter(estimatedSize);
            for (Map<String, Object> config : parameterSpace) {
                if (cfg.getConfiguration() != null) {
                    for (Entry<String, Object> e : cfg.getConfiguration().entrySet()) {
                        if (!config.containsKey(e.getKey())) {
                            config.put(e.getKey(), e.getValue());
                        }
                    }
                }
                log.info("== Running new configuration [" + ctx.getId() + "] ==");
                List<String> keys = new ArrayList<String>(config.keySet());
                for (String key : keys) {
                    log.info("[" + key + "]: [" + StringUtils.abbreviateMiddle(Util.toString(config.get(key)), "…", 150) + "]");
                }
                executeConfiguration(cfg, ctx, config, executedSubtasks);
                progress.next();
                log.info("Completed configuration " + progress);
            }
            // Set the subtask property and persist again, so the property is available to
            // reports
            cfg.setAttribute(SUBTASKS_KEY, executedSubtasks.toString());
            cfg.persist(ctx);
        } catch (LifeCycleException e) {
            ctx.getLifeCycleManager().fail(ctx, aConfiguration, e);
            throw e;
        } catch (UnresolvedImportException e) {
            // HACK - pass unresolved import exceptions up to the outer batch task
            ctx.getLifeCycleManager().fail(ctx, aConfiguration, e);
            throw e;
        } catch (Throwable e) {
            ctx.getLifeCycleManager().fail(ctx, aConfiguration, e);
            throw new ExecutionException(e);
        }
        // End recording (here the reports will nbe done)
        ctx.getLifeCycleManager().complete(ctx, aConfiguration);
        return ctx.getId();
    } finally {
        if (ctx != null) {
            ctx.getLifeCycleManager().destroy(ctx, aConfiguration);
        }
    }
}