聊聊storm的WindowedBoltExecutor

序

本文主要研究一下storm的WindowedBoltExecutor

WindowedBoltExecutor

storm-2.0.0/storm-client/src/jvm/org/apache/storm/topology/WindowedBoltExecutor.java

/**
 * An {@link IWindowedBolt} wrapper that does the windowing of tuples.
 */
public class WindowedBoltExecutor implements IRichBolt {
    public static final String LATE_TUPLE_FIELD = "late_tuple";
    private static final Logger LOG = LoggerFactory.getLogger(WindowedBoltExecutor.class);
    private static final int DEFAULT_WATERMARK_EVENT_INTERVAL_MS = 1000; // 1s
    private static final int DEFAULT_MAX_LAG_MS = 0; // no lag
    private final IWindowedBolt bolt;
    // package level for unit tests
    transient WaterMarkEventGenerator<Tuple> waterMarkEventGenerator;
    private transient WindowedOutputCollector windowedOutputCollector;
    private transient WindowLifecycleListener<Tuple> listener;
    private transient WindowManager<Tuple> windowManager;
    private transient int maxLagMs;
    private TimestampExtractor timestampExtractor;
    private transient String lateTupleStream;
    private transient TriggerPolicy<Tuple, ?> triggerPolicy;
    private transient EvictionPolicy<Tuple, ?> evictionPolicy;
    private transient Duration windowLengthDuration;

    public WindowedBoltExecutor(IWindowedBolt bolt) {
        this.bolt = bolt;
        timestampExtractor = bolt.getTimestampExtractor();
    }

    @Override
    public void prepare(Map<String, Object> topoConf, TopologyContext context, OutputCollector collector) {
        doPrepare(topoConf, context, collector, new ConcurrentLinkedQueue<>(), false);
    }

    // NOTE: the queue has to be thread safe.
    protected void doPrepare(Map<String, Object> topoConf, TopologyContext context, OutputCollector collector,
                             Collection<Event<Tuple>> queue, boolean stateful) {
        Objects.requireNonNull(topoConf);
        Objects.requireNonNull(context);
        Objects.requireNonNull(collector);
        Objects.requireNonNull(queue);
        this.windowedOutputCollector = new WindowedOutputCollector(collector);
        bolt.prepare(topoConf, context, windowedOutputCollector);
        this.listener = newWindowLifecycleListener();
        this.windowManager = initWindowManager(listener, topoConf, context, queue, stateful);
        start();
        LOG.info("Initialized window manager {} ", windowManager);
    }

    @Override
    public void execute(Tuple input) {
        if (isTupleTs()) {
            long ts = timestampExtractor.extractTimestamp(input);
            if (waterMarkEventGenerator.track(input.getSourceGlobalStreamId(), ts)) {
                windowManager.add(input, ts);
            } else {
                if (lateTupleStream != null) {
                    windowedOutputCollector.emit(lateTupleStream, input, new Values(input));
                } else {
                    LOG.info("Received a late tuple {} with ts {}. This will not be processed.", input, ts);
                }
                windowedOutputCollector.ack(input);
            }
        } else {
            windowManager.add(input);
        }
    }

    @Override
    public void cleanup() {
        if (waterMarkEventGenerator != null) {
            waterMarkEventGenerator.shutdown();
        }
        windowManager.shutdown();
        bolt.cleanup();
    }

    // for unit tests
    WindowManager<Tuple> getWindowManager() {
        return windowManager;
    }

    @Override
    public void declareOutputFields(OutputFieldsDeclarer declarer) {
        String lateTupleStream = (String) getComponentConfiguration().get(Config.TOPOLOGY_BOLTS_LATE_TUPLE_STREAM);
        if (lateTupleStream != null) {
            declarer.declareStream(lateTupleStream, new Fields(LATE_TUPLE_FIELD));
        }
        bolt.declareOutputFields(declarer);
    }

    @Override
    public Map<String, Object> getComponentConfiguration() {
        return bolt.getComponentConfiguration();
    }

    //......
}

• WindowedBoltExecutor實現了IRichBolt接口，在prepare的時候初始化windowedOutputCollector、listener、windowManager，調用了bolt.prepare；在cleanup的時候對waterMarkEventGenerator、windowManager、bolt進行清理；TopologyBuilder在setBolt的時候，對原始的IWindowedBolt的實現類進行了一次包裝，用WindowedBoltExecutor替代
• declareOutputFields採用的是bolt.declareOutputFields(declarer)；getComponentConfiguration也返回的是bolt.getComponentConfiguration();
• execute方法主要是將tuple添加到windowManager，對於不歸入window的tuple則馬上進行ack

WindowedOutputCollector

storm-2.0.0/storm-client/src/jvm/org/apache/storm/topology/WindowedBoltExecutor.java

/**
     * Creates an {@link OutputCollector} wrapper that automatically anchors the tuples to inputTuples while emitting.
     */
    private static class WindowedOutputCollector extends OutputCollector {
        private List<Tuple> inputTuples;

        WindowedOutputCollector(IOutputCollector delegate) {
            super(delegate);
        }

        void setContext(List<Tuple> inputTuples) {
            this.inputTuples = inputTuples;
        }

        @Override
        public List<Integer> emit(String streamId, List<Object> tuple) {
            return emit(streamId, inputTuples, tuple);
        }

        @Override
        public void emitDirect(int taskId, String streamId, List<Object> tuple) {
            emitDirect(taskId, streamId, inputTuples, tuple);
        }
    }

• WindowedOutputCollector繼承了OutputCollector，能夠看到這裏重寫了emit計emitDirect方法，默認對inputTuples進行anchor

WindowLifecycleListener

storm-2.0.0/storm-client/src/jvm/org/apache/storm/windowing/WindowLifecycleListener.java

/**
 * A callback for expiry, activation of events tracked by the {@link WindowManager}
 *
 * @param <T> The type of Event in the window (e.g. Tuple).
 */
public interface WindowLifecycleListener<T> {
    /**
     * Called on expiry of events from the window due to {@link EvictionPolicy}
     *
     * @param events the expired events
     */
    void onExpiry(List<T> events);

    /**
     * Called on activation of the window due to the {@link TriggerPolicy}
     *
     * @param events        the list of current events in the window.
     * @param newEvents     the newly added events since last activation.
     * @param expired       the expired events since last activation.
     * @param referenceTime the reference (event or processing) time that resulted in activation
     */
    default void onActivation(List<T> events, List<T> newEvents, List<T> expired, Long referenceTime) {
        throw new UnsupportedOperationException("Not implemented");
    }

    /**
     * Called on activation of the window due to the {@link TriggerPolicy}. This is typically invoked when the windows are persisted in
     * state and is huge to be loaded entirely in memory.
     *
     * @param eventsIt      a supplier of iterator over the list of current events in the window
     * @param newEventsIt   a supplier of iterator over the newly added events since the last ativation
     * @param expiredIt     a supplier of iterator over the expired events since the last activation
     * @param referenceTime the reference (event or processing) time that resulted in activation
     */
    default void onActivation(Supplier<Iterator<T>> eventsIt, Supplier<Iterator<T>> newEventsIt, Supplier<Iterator<T>> expiredIt,
                              Long referenceTime) {
        throw new UnsupportedOperationException("Not implemented");
    }
}

• WindowLifecycleListener定義了幾個回調方法，分別是onExpiry、onActivation
• 它們分別是由EvictionPolicy、TriggerPolicy兩種策略來觸發

EvictionPolicy

storm-2.0.0/storm-client/src/jvm/org/apache/storm/windowing/EvictionPolicy.java

/**
 * Eviction policy tracks events and decides whether an event should be evicted from the window or not.
 *
 * @param <T> the type of event that is tracked.
 */
public interface EvictionPolicy<T, S> {
    /**
     * Decides if an event should be expired from the window, processed in the current window or kept for later processing.
     *
     * @param event the input event
     * @return the {@link org.apache.storm.windowing.EvictionPolicy.Action} to be taken based on the input event
     */
    Action evict(Event<T> event);

    /**
     * Tracks the event to later decide whether {@link EvictionPolicy#evict(Event)} should evict it or not.
     *
     * @param event the input event to be tracked
     */
    void track(Event<T> event);

    /**
     * Returns the current context that is part of this eviction policy.
     *
     * @return the eviction context
     */
    EvictionContext getContext();

    /**
     * Sets a context in the eviction policy that can be used while evicting the events. E.g. For TimeEvictionPolicy, this could be used to
     * set the reference timestamp.
     *
     * @param context the eviction context
     */
    void setContext(EvictionContext context);

    /**
     * Resets the eviction policy.
     */
    void reset();

    /**
     * Return runtime state to be checkpointed by the framework for restoring the eviction policy in case of failures.
     *
     * @return the state
     */
    S getState();

    /**
     * Restore the eviction policy from the state that was earlier checkpointed by the framework.
     *
     * @param state the state
     */
    void restoreState(S state);

    /**
     * The action to be taken when {@link EvictionPolicy#evict(Event)} is invoked.
     */
    public enum Action {
        /**
         * expire the event and remove it from the queue.
         */
        EXPIRE,
        /**
         * process the event in the current window of events.
         */
        PROCESS,
        /**
         * don't include in the current window but keep the event in the queue for evaluating as a part of future windows.
         */
        KEEP,
        /**
         * stop processing the queue, there cannot be anymore events satisfying the eviction policy.
         */
        STOP
    }
}

• EvictionPolicy主要負責追蹤event，而後判斷event是否該從window中移除
• EvictionPolicy有幾個實現類：CountEvictionPolicy、TimeEvictionPolicy、WatermarkCountEvictionPolicy、WatermarkTimeEvictionPolicy

TriggerPolicy

storm-2.0.0/storm-client/src/jvm/org/apache/storm/windowing/TriggerPolicy.java

/**
 * Triggers the window calculations based on the policy.
 *
 * @param <T> the type of the event that is tracked
 */
public interface TriggerPolicy<T, S> {
    /**
     * Tracks the event and could use this to invoke the trigger.
     *
     * @param event the input event
     */
    void track(Event<T> event);

    /**
     * resets the trigger policy.
     */
    void reset();

    /**
     * Starts the trigger policy. This can be used during recovery to start the triggers after recovery is complete.
     */
    void start();

    /**
     * Any clean up could be handled here.
     */
    void shutdown();

    /**
     * Return runtime state to be checkpointed by the framework for restoring the trigger policy in case of failures.
     *
     * @return the state
     */
    S getState();

    /**
     * Restore the trigger policy from the state that was earlier checkpointed by the framework.
     *
     * @param state the state
     */
    void restoreState(S state);
}

• TriggerPolicy主要是負責window的計算
• TriggerPolicy有幾個實現類：CountTriggerPolicy、TimeTriggerPolicy、WatermarkCountTriggerPolicy、WatermarkTimeTriggerPolicy

WindowedBoltExecutor.newWindowLifecycleListener

storm-2.0.0/storm-client/src/jvm/org/apache/storm/topology/WindowedBoltExecutor.java

protected WindowLifecycleListener<Tuple> newWindowLifecycleListener() {
        return new WindowLifecycleListener<Tuple>() {
            @Override
            public void onExpiry(List<Tuple> tuples) {
                for (Tuple tuple : tuples) {
                    windowedOutputCollector.ack(tuple);
                }
            }

            @Override
            public void onActivation(List<Tuple> tuples, List<Tuple> newTuples, List<Tuple> expiredTuples, Long timestamp) {
                windowedOutputCollector.setContext(tuples);
                boltExecute(tuples, newTuples, expiredTuples, timestamp);
            }

        };
    }

    protected void boltExecute(List<Tuple> tuples, List<Tuple> newTuples, List<Tuple> expiredTuples, Long timestamp) {
        bolt.execute(new TupleWindowImpl(tuples, newTuples, expiredTuples, getWindowStartTs(timestamp), timestamp));
    }

• 這裏建立了一個匿名的WindowLifecycleListener實現
• 在onExpiry的時候挨個對tuple進行ack，在onActivation的時候，調用了boltExecute，構造TupleWindowImpl，傳遞給bolt進行執行

WindowedBoltExecutor.initWindowManager

storm-2.0.0/storm-client/src/jvm/org/apache/storm/topology/WindowedBoltExecutor.java

private WindowManager<Tuple> initWindowManager(WindowLifecycleListener<Tuple> lifecycleListener, Map<String, Object> topoConf,
                                                   TopologyContext context, Collection<Event<Tuple>> queue, boolean stateful) {

        WindowManager<Tuple> manager = stateful ?
            new StatefulWindowManager<>(lifecycleListener, queue)
            : new WindowManager<>(lifecycleListener, queue);

        Count windowLengthCount = null;
        Duration slidingIntervalDuration = null;
        Count slidingIntervalCount = null;
        // window length
        if (topoConf.containsKey(Config.TOPOLOGY_BOLTS_WINDOW_LENGTH_COUNT)) {
            windowLengthCount = new Count(((Number) topoConf.get(Config.TOPOLOGY_BOLTS_WINDOW_LENGTH_COUNT)).intValue());
        } else if (topoConf.containsKey(Config.TOPOLOGY_BOLTS_WINDOW_LENGTH_DURATION_MS)) {
            windowLengthDuration = new Duration(
                ((Number) topoConf.get(Config.TOPOLOGY_BOLTS_WINDOW_LENGTH_DURATION_MS)).intValue(),
                TimeUnit.MILLISECONDS);
        }
        // sliding interval
        if (topoConf.containsKey(Config.TOPOLOGY_BOLTS_SLIDING_INTERVAL_COUNT)) {
            slidingIntervalCount = new Count(((Number) topoConf.get(Config.TOPOLOGY_BOLTS_SLIDING_INTERVAL_COUNT)).intValue());
        } else if (topoConf.containsKey(Config.TOPOLOGY_BOLTS_SLIDING_INTERVAL_DURATION_MS)) {
            slidingIntervalDuration =
                new Duration(((Number) topoConf.get(Config.TOPOLOGY_BOLTS_SLIDING_INTERVAL_DURATION_MS)).intValue(), TimeUnit.MILLISECONDS);
        } else {
            // default is a sliding window of count 1
            slidingIntervalCount = new Count(1);
        }
        // tuple ts
        if (timestampExtractor != null) {
            // late tuple stream
            lateTupleStream = (String) topoConf.get(Config.TOPOLOGY_BOLTS_LATE_TUPLE_STREAM);
            if (lateTupleStream != null) {
                if (!context.getThisStreams().contains(lateTupleStream)) {
                    throw new IllegalArgumentException(
                        "Stream for late tuples must be defined with the builder method withLateTupleStream");
                }
            }
            // max lag
            if (topoConf.containsKey(Config.TOPOLOGY_BOLTS_TUPLE_TIMESTAMP_MAX_LAG_MS)) {
                maxLagMs = ((Number) topoConf.get(Config.TOPOLOGY_BOLTS_TUPLE_TIMESTAMP_MAX_LAG_MS)).intValue();
            } else {
                maxLagMs = DEFAULT_MAX_LAG_MS;
            }
            // watermark interval
            int watermarkInterval;
            if (topoConf.containsKey(Config.TOPOLOGY_BOLTS_WATERMARK_EVENT_INTERVAL_MS)) {
                watermarkInterval = ((Number) topoConf.get(Config.TOPOLOGY_BOLTS_WATERMARK_EVENT_INTERVAL_MS)).intValue();
            } else {
                watermarkInterval = DEFAULT_WATERMARK_EVENT_INTERVAL_MS;
            }
            waterMarkEventGenerator = new WaterMarkEventGenerator<>(manager, watermarkInterval,
                                                                    maxLagMs, getComponentStreams(context));
        } else {
            if (topoConf.containsKey(Config.TOPOLOGY_BOLTS_LATE_TUPLE_STREAM)) {
                throw new IllegalArgumentException("Late tuple stream can be defined only when specifying a timestamp field");
            }
        }
        // validate
        validate(topoConf, windowLengthCount, windowLengthDuration,
                 slidingIntervalCount, slidingIntervalDuration);
        evictionPolicy = getEvictionPolicy(windowLengthCount, windowLengthDuration);
        triggerPolicy = getTriggerPolicy(slidingIntervalCount, slidingIntervalDuration,
                                         manager, evictionPolicy);
        manager.setEvictionPolicy(evictionPolicy);
        manager.setTriggerPolicy(triggerPolicy);
        return manager;
    }

    private EvictionPolicy<Tuple, ?> getEvictionPolicy(Count windowLengthCount, Duration windowLengthDuration) {
        if (windowLengthCount != null) {
            if (isTupleTs()) {
                return new WatermarkCountEvictionPolicy<>(windowLengthCount.value);
            } else {
                return new CountEvictionPolicy<>(windowLengthCount.value);
            }
        } else {
            if (isTupleTs()) {
                return new WatermarkTimeEvictionPolicy<>(windowLengthDuration.value, maxLagMs);
            } else {
                return new TimeEvictionPolicy<>(windowLengthDuration.value);
            }
        }
    }

    private TriggerPolicy<Tuple, ?> getTriggerPolicy(Count slidingIntervalCount, Duration slidingIntervalDuration,
                                                     WindowManager<Tuple> manager, EvictionPolicy<Tuple, ?> evictionPolicy) {
        if (slidingIntervalCount != null) {
            if (isTupleTs()) {
                return new WatermarkCountTriggerPolicy<>(slidingIntervalCount.value, manager, evictionPolicy, manager);
            } else {
                return new CountTriggerPolicy<>(slidingIntervalCount.value, manager, evictionPolicy);
            }
        } else {
            if (isTupleTs()) {
                return new WatermarkTimeTriggerPolicy<>(slidingIntervalDuration.value, manager, evictionPolicy, manager);
            } else {
                return new TimeTriggerPolicy<>(slidingIntervalDuration.value, manager, evictionPolicy);
            }
        }
    }

• 對於WindowedBoltExecutor來講，stateful爲false，這裏建立的是WindowManager
• 這裏默認的DEFAULT_MAX_LAG_MS爲0，即沒有lag，默認的DEFAULT_WATERMARK_EVENT_INTERVAL_MS爲1000，即1秒
• 這裏根據windowLength及slidingInterval指定的參數類型，來獲取相應的EvictionPolicy及TriggerPolicy，對於有配置timestampField的，參數是Duration的，則建立的是WatermarkTimeEvictionPolicy以及WatermarkTimeTriggerPolicy

WindowManager

storm-2.0.0/storm-client/src/jvm/org/apache/storm/windowing/WindowManager.java

/**
 * Tracks a window of events and fires {@link WindowLifecycleListener} callbacks on expiry of events or activation of the window due to
 * {@link TriggerPolicy}.
 *
 * @param <T> the type of event in the window.
 */
public class WindowManager<T> implements TriggerHandler {

    protected final Collection<Event<T>> queue;

    private final AtomicInteger eventsSinceLastExpiry;

    //......
    /**
     * Add an event into the window, with the given ts as the tracking ts.
     *
     * @param event the event to track
     * @param ts    the timestamp
     */
    public void add(T event, long ts) {
        add(new EventImpl<T>(event, ts));
    }

    /**
     * Tracks a window event
     *
     * @param windowEvent the window event to track
     */
    public void add(Event<T> windowEvent) {
        // watermark events are not added to the queue.
        if (!windowEvent.isWatermark()) {
            queue.add(windowEvent);
        } else {
            LOG.debug("Got watermark event with ts {}", windowEvent.getTimestamp());
        }
        track(windowEvent);
        compactWindow();
    }

    /**
     * feed the event to the eviction and trigger policies for bookkeeping and optionally firing the trigger.
     */
    private void track(Event<T> windowEvent) {
        evictionPolicy.track(windowEvent);
        triggerPolicy.track(windowEvent);
    }

    /**
     * expires events that fall out of the window every EXPIRE_EVENTS_THRESHOLD so that the window does not grow too big.
     */
    protected void compactWindow() {
        if (eventsSinceLastExpiry.incrementAndGet() >= EXPIRE_EVENTS_THRESHOLD) {
            scanEvents(false);
        }
    }

    /**
     * Scan events in the queue, using the expiration policy to check if the event should be evicted or not.
     *
     * @param fullScan if set, will scan the entire queue; if not set, will stop as soon as an event not satisfying the expiration policy is
     *                 found
     * @return the list of events to be processed as a part of the current window
     */
    private List<Event<T>> scanEvents(boolean fullScan) {
        LOG.debug("Scan events, eviction policy {}", evictionPolicy);
        List<T> eventsToExpire = new ArrayList<>();
        List<Event<T>> eventsToProcess = new ArrayList<>();
        try {
            lock.lock();
            Iterator<Event<T>> it = queue.iterator();
            while (it.hasNext()) {
                Event<T> windowEvent = it.next();
                Action action = evictionPolicy.evict(windowEvent);
                if (action == EXPIRE) {
                    eventsToExpire.add(windowEvent.get());
                    it.remove();
                } else if (!fullScan || action == STOP) {
                    break;
                } else if (action == PROCESS) {
                    eventsToProcess.add(windowEvent);
                }
            }
            expiredEvents.addAll(eventsToExpire);
        } finally {
            lock.unlock();
        }
        eventsSinceLastExpiry.set(0);
        LOG.debug("[{}] events expired from window.", eventsToExpire.size());
        if (!eventsToExpire.isEmpty()) {
            LOG.debug("invoking windowLifecycleListener.onExpiry");
            windowLifecycleListener.onExpiry(eventsToExpire);
        }
        return eventsToProcess;
    }

    //......
}

• WindowedBoltExecutor的execute主要是將tuple添加到windowManager
• EventImpl的isWatermark返回false，這裏主要是執行track及compactWindow操做
• track主要是委託給evictionPolicy以及triggerPolicy進行track，compactWindow在events超過指定閾值的時候，會觸發scanEvents，不是fullScan的話，檢測到一個非過時的event就跳出遍歷，而後檢測eventsToExpire是否爲空若是有則觸發windowLifecycleListener.onExpiry(eventsToExpire);

WaterMarkEventGenerator

storm-2.0.0/storm-client/src/jvm/org/apache/storm/windowing/WaterMarkEventGenerator.java

/**
 * Tracks tuples across input streams and periodically emits watermark events. Watermark event timestamp is the minimum of the latest tuple
 * timestamps across all the input streams (minus the lag). Once a watermark event is emitted any tuple coming with an earlier timestamp can
 * be considered as late events.
 */
public class WaterMarkEventGenerator<T> implements Runnable {
    private static final Logger LOG = LoggerFactory.getLogger(WaterMarkEventGenerator.class);
    private final WindowManager<T> windowManager;
    private final int eventTsLag;
    private final Set<GlobalStreamId> inputStreams;
    private final Map<GlobalStreamId, Long> streamToTs;
    private final ScheduledExecutorService executorService;
    private final int interval;
    private ScheduledFuture<?> executorFuture;
    private volatile long lastWaterMarkTs;

    //......

    public void start() {
        this.executorFuture = executorService.scheduleAtFixedRate(this, interval, interval, TimeUnit.MILLISECONDS);
    }

    @Override
    public void run() {
        try {
            long waterMarkTs = computeWaterMarkTs();
            if (waterMarkTs > lastWaterMarkTs) {
                this.windowManager.add(new WaterMarkEvent<>(waterMarkTs));
                lastWaterMarkTs = waterMarkTs;
            }
        } catch (Throwable th) {
            LOG.error("Failed while processing watermark event ", th);
            throw th;
        }
    }
}

• WindowedBoltExecutor在start的時候會調用WaterMarkEventGenerator的start方法
• 該方法每隔watermarkInterval時間調度WaterMarkEventGenerator這個任務
• 其run方法就是計算watermark(這批數據最小值-lag)，當大於lastWaterMarkTs時，更新lastWaterMarkTs，往windowManager添加WaterMarkEvent(該event的isWatermark爲true)
• windowManager.add(new WaterMarkEvent<>(waterMarkTs))會觸發triggerPolicy.track(windowEvent)以及compactWindow操做

WatermarkTimeTriggerPolicy.track

storm-2.0.0/storm-client/src/jvm/org/apache/storm/windowing/WatermarkTimeTriggerPolicy.java

@Override
    public void track(Event<T> event) {
        if (started && event.isWatermark()) {
            handleWaterMarkEvent(event);
        }
    }

    /**
     * Invokes the trigger all pending windows up to the watermark timestamp. The end ts of the window is set in the eviction policy context
     * so that the events falling within that window can be processed.
     */
    private void handleWaterMarkEvent(Event<T> event) {
        long watermarkTs = event.getTimestamp();
        long windowEndTs = nextWindowEndTs;
        LOG.debug("Window end ts {} Watermark ts {}", windowEndTs, watermarkTs);
        while (windowEndTs <= watermarkTs) {
            long currentCount = windowManager.getEventCount(windowEndTs);
            evictionPolicy.setContext(new DefaultEvictionContext(windowEndTs, currentCount));
            if (handler.onTrigger()) {
                windowEndTs += slidingIntervalMs;
            } else {
                /*
                 * No events were found in the previous window interval.
                 * Scan through the events in the queue to find the next
                 * window intervals based on event ts.
                 */
                long ts = getNextAlignedWindowTs(windowEndTs, watermarkTs);
                LOG.debug("Next aligned window end ts {}", ts);
                if (ts == Long.MAX_VALUE) {
                    LOG.debug("No events to process between {} and watermark ts {}", windowEndTs, watermarkTs);
                    break;
                }
                windowEndTs = ts;
            }
        }
        nextWindowEndTs = windowEndTs;
    }

    /**
     * Computes the next window by scanning the events in the window and finds the next aligned window between the startTs and endTs. Return
     * the end ts of the next aligned window, i.e. the ts when the window should fire.
     *
     * @param startTs the start timestamp (excluding)
     * @param endTs   the end timestamp (including)
     * @return the aligned window end ts for the next window or Long.MAX_VALUE if there are no more events to be processed.
     */
    private long getNextAlignedWindowTs(long startTs, long endTs) {
        long nextTs = windowManager.getEarliestEventTs(startTs, endTs);
        if (nextTs == Long.MAX_VALUE || (nextTs % slidingIntervalMs == 0)) {
            return nextTs;
        }
        return nextTs + (slidingIntervalMs - (nextTs % slidingIntervalMs));
    }

• handleWaterMarkEvent會觸發handler.onTrigger()方法

WindowManager.onTrigger

storm-2.0.0/storm-client/src/jvm/org/apache/storm/windowing/WindowManager.java

/**
     * The callback invoked by the trigger policy.
     */
    @Override
    public boolean onTrigger() {
        List<Event<T>> windowEvents = null;
        List<T> expired = null;
        try {
            lock.lock();
            /*
             * scan the entire window to handle out of order events in
             * the case of time based windows.
             */
            windowEvents = scanEvents(true);
            expired = new ArrayList<>(expiredEvents);
            expiredEvents.clear();
        } finally {
            lock.unlock();
        }
        List<T> events = new ArrayList<>();
        List<T> newEvents = new ArrayList<>();
        for (Event<T> event : windowEvents) {
            events.add(event.get());
            if (!prevWindowEvents.contains(event)) {
                newEvents.add(event.get());
            }
        }
        prevWindowEvents.clear();
        if (!events.isEmpty()) {
            prevWindowEvents.addAll(windowEvents);
            LOG.debug("invoking windowLifecycleListener onActivation, [{}] events in window.", events.size());
            windowLifecycleListener.onActivation(events, newEvents, expired, evictionPolicy.getContext().getReferenceTime());
        } else {
            LOG.debug("No events in the window, skipping onActivation");
        }
        triggerPolicy.reset();
        return !events.isEmpty();
    }

• onTrigger方法主要是計算出三類數據，events、expiredEvents、newEvents
• 當events不爲空時，觸發windowLifecycleListener.onActivation，也就是調用bolt的execute方法

小結

• WindowedBoltExecutor實現了IRichBolt接口，是一個bolt，TopologyBuilder在setBolt的時候，對用戶的IWindowedBolt的實現類進行了一次包裝，用WindowedBoltExecutor替代，它改造了execute方法，對於該歸入windows的調用windowManager.add添加，該丟棄的則進行ack，而真正的bolt的execute操做，則須要等待window的觸發
• WindowLifecycleListener有兩個回調操做，一個是由EvictionPolicy觸發的onExpiry，一個是由TriggerPolicy觸發的onActivation操做
• 因爲window的windowLength及slidingInterval參數有Duration及Count兩個維度，於是EvictionPolicy及TriggerPolicy也有這兩類維度，外加watermark屬性，於是每一個policy分別有4個實現類，EvictionPolicy有幾個實現類：CountEvictionPolicy、TimeEvictionPolicy、WatermarkCountEvictionPolicy、WatermarkTimeEvictionPolicy；TriggerPolicy有幾個實現類：CountTriggerPolicy、TimeTriggerPolicy、WatermarkCountTriggerPolicy、WatermarkTimeTriggerPolicy
• windowManager.add除了把tuple保存起來外，還調用了兩類trigger的track操做，而後進行compactWindow操做；WatermarkTimeEvictionPolicy的track目前沒有操做，而WatermarkTimeTriggerPolicy的track方法在event是WaterMarkEvent的時候會觸發window操做，調用WindowManager的onTrigger方法，進而篩選出window的數據，而後觸發windowLifecycleListener.onActivation操做，最後觸發windowedBolt的execute方法
• WindowManager的onTrigger方法以及add方法都會調用scanEvents，區別是前者是fullScan，後者不是；scanEvents會調用evictionPolicy.evict來判斷是否該剔除tuple，進而觸發windowLifecycleListener.onExpiry操做，該操做會對tuple進行ack，即過時的tuple在expired的時候會自動ack(理論上全部tuple都會過時，也就都會自動被ack，於是要求Config.TOPOLOGY_MESSAGE_TIMEOUT_SECS大於windowLength + slidingInterval，避免還沒ack就被認爲超時)
• WindowedBoltExecutor在start的時候會啓動WaterMarkEventGenerator，它會註冊一個定時任務，每隔watermarkInterval時間計算watermark(這批數據最小值-lag)，當大於lastWaterMarkTs時，更新lastWaterMarkTs，往windowManager添加WaterMarkEvent(該event的isWatermark爲true)，整個WindowManager的onTrigger方法(即windowLifecycleListener.onActivation操做)就是靠這裏來觸發的
• 關於ack的話，在WindowedBoltExecutor.execute方法對於未能進入window隊列的，沒有配置配置Config.TOPOLOGY_BOLTS_LATE_TUPLE_STREAM的話，則立馬ack；在tuple過時的時候會自ack；WindowedBoltExecutor使用了WindowedOutputCollector，它繼承了OutputCollector，對輸入的tuples作anchor操做

doc

• Windowing Support in Core Storm