Flink – Trigger,Evictor
org.apache.flink.streaming.api.windowing.triggers;
Trigger
public abstract class Trigger<T, W extends Window> implements Serializable { /**<br/> * Called for every element that gets added to a pane. The result of this will determine<br/> * whether the pane is evaluated to emit results.<br/> *<br/> * @param element The element that arrived.<br/> * @param timestamp The timestamp of the element that arrived.<br/> * @param window The window to which the element is being added.<br/> * @param ctx A context object that can be used to register timer callbacks.<br/> */<br/> public abstract TriggerResult onElement(T element, long timestamp, W window, TriggerContext ctx) throws Exception; /**<br/> * Called when a processing-time timer that was set using the trigger context fires.<br/> *<br/> * @param time The timestamp at which the timer fired.<br/> * @param window The window for which the timer fired.<br/> * @param ctx A context object that can be used to register timer callbacks.<br/> */<br/> public abstract TriggerResult onProcessingTime(long time, W window, TriggerContext ctx) throws Exception; /**<br/> * Called when an event-time timer that was set using the trigger context fires.<br/> *<br/> * @param time The timestamp at which the timer fired.<br/> * @param window The window for which the timer fired.<br/> * @param ctx A context object that can be used to register timer callbacks.<br/> */<br/> public abstract TriggerResult onEventTime(long time, W window, TriggerContext ctx) throws Exception; /**<br/> * Called when several windows have been merged into one window by the<br/> * {@link org.apache.flink.streaming.api.windowing.assigners.WindowAssigner}.<br/> *<br/> * @param window The new window that results from the merge.<br/> * @param ctx A context object that can be used to register timer callbacks and access state.<br/> */<br/> public TriggerResult onMerge(W window, OnMergeContext ctx) throws Exception {<br/> throw new RuntimeException("This trigger does not support merging.");<br/> }
Trigger决定pane何时被evaluated,实现一系列接口,来判断各种情况下是否需要trigger
看看具体的trigger的实现,
ProcessingTimeTrigger
/**<br/> * A {@link Trigger} that fires once the current system time passes the end of the window<br/> * to which a pane belongs.<br/> */<br/> public class ProcessingTimeTrigger implements Trigger<Object, TimeWindow> {<br/> private static final long serialVersionUID = 1L; private ProcessingTimeTrigger() {} @Override<br/> public TriggerResult onElement(Object element, long timestamp, TimeWindow window, TriggerContext ctx) {<br/> ctx.registerProcessingTimeTimer(window.maxTimestamp()); //对于processingTime,element的trigger时间是current+window,所以这里需要注册定时器去触发<br/> return TriggerResult.CONTINUE;<br/> } @Override<br/> public TriggerResult onEventTime(long time, TimeWindow window, TriggerContext ctx) throws Exception {<br/> return TriggerResult.CONTINUE;<br/> } @Override<br/> public TriggerResult onProcessingTime(long time, TimeWindow window, TriggerContext ctx) {//触发后调用<br/> return TriggerResult.FIRE_AND_PURGE;<br/> } @Override<br/> public String toString() {<br/> return "ProcessingTimeTrigger()";<br/> } /**<br/> * Creates a new trigger that fires once system time passes the end of the window.<br/> */<br/> public static ProcessingTimeTrigger create() {<br/> return new ProcessingTimeTrigger();<br/> }<br/> }
可以看到只有在onProcessingTime的时候,是FIRE_AND_PURGE,其他时候都是continue
再看个CountTrigger,
public class CountTrigger<W extends Window> extends Trigger<Object, W> { private final long maxCount; private final ReducingStateDescriptor<Long> stateDesc =<br/> new ReducingStateDescriptor<>("count", new Sum(), LongSerializer.INSTANCE); private CountTrigger(long maxCount) {<br/> this.maxCount = maxCount;<br/> } @Override<br/> public TriggerResult onElement(Object element, long timestamp, W window, TriggerContext ctx) throws Exception {<br/> ReducingState<Long> count = ctx.getPartitionedState(stateDesc); //从backend取出conunt state<br/> count.add(1L); //加1<br/> if (count.get() >= maxCount) {<br/> count.clear();<br/> return TriggerResult.FIRE;<br/> }<br/> return TriggerResult.CONTINUE;<br/> } @Override<br/> public TriggerResult onEventTime(long time, W window, TriggerContext ctx) {<br/> return TriggerResult.CONTINUE;<br/> } @Override<br/> public TriggerResult onProcessingTime(long time, W window, TriggerContext ctx) throws Exception {<br/> return TriggerResult.CONTINUE;<br/> } @Override<br/> public TriggerResult onMerge(W window, OnMergeContext ctx) throws Exception {<br/> ctx.mergePartitionedState(stateDesc); //先调用merge,底层backend里面的window进行merge<br/> ReducingState<Long> count = ctx.getPartitionedState(stateDesc); //merge后再取出state,count,进行判断<br/> if (count.get() >= maxCount) {<br/> return TriggerResult.FIRE;<br/> }<br/> return TriggerResult.CONTINUE;<br/> }
很简单,既然是算count,那么和time相关的自然都是continue
对于count,是在onElement中触发,每次来element都会走到这个逻辑
当累积的count > 设定的count时,就会返回Fire,注意,这里这是fire,并不会purge
并将计数清0
TriggerResult
TriggerResult是个枚举,
enum TriggerResult {<br/> CONTINUE(false, false), FIRE_AND_PURGE(true, true), FIRE(true, false), PURGE(false, true); private final boolean fire;<br/> private final boolean purge;<br/> }
两个选项,fire,purge,2×2,所以4种可能性
两个Result可以merge,
/**<br/> * Merges two {@code TriggerResults}. This specifies what should happen if we have<br/> * two results from a Trigger, for example as a result from<br/> * {@link Trigger#onElement(Object, long, Window, Trigger.TriggerContext)} and<br/> * {@link Trigger#onEventTime(long, Window, Trigger.TriggerContext)}.<br/> *<br/> * <p><br/> * For example, if one result says {@code CONTINUE} while the other says {@code FIRE}<br/> * then {@code FIRE} is the combined result;<br/> */<br/> public static TriggerResult merge(TriggerResult a, TriggerResult b) {<br/> if (a.purge || b.purge) {<br/> if (a.fire || b.fire) {<br/> return FIRE_AND_PURGE;<br/> } else {<br/> return PURGE;<br/> }<br/> } else if (a.fire || b.fire) {<br/> return FIRE;<br/> } else {<br/> return CONTINUE;<br/> }<br/> }
TriggerContext
为Trigger做些环境的工作,比如管理timer,和处理state
这些接口在,Trigger中的接口逻辑里面都会用到,所以在Trigger的所有接口上,都需要传入context
/**<br/> * A context object that is given to {@link Trigger} methods to allow them to register timer<br/> * callbacks and deal with state.<br/> */<br/> public interface TriggerContext { long getCurrentProcessingTime();<br/> long getCurrentWatermark(); /**<br/> * Register a system time callback. When the current system time passes the specified<br/> * time {@link Trigger#onProcessingTime(long, Window, TriggerContext)} is called with the time specified here.<br/> *<br/> * @param time The time at which to invoke {@link Trigger#onProcessingTime(long, Window, TriggerContext)}<br/> */<br/> void registerProcessingTimeTimer(long time);<br/> void registerEventTimeTimer(long time); void deleteProcessingTimeTimer(long time);<br/> void deleteEventTimeTimer(long time); <S extends State> S getPartitionedState(StateDescriptor<S, ?> stateDescriptor);<br/> }
OnMergeContext 仅仅是多了一个接口,
public interface OnMergeContext extends TriggerContext {<br/> <S extends MergingState<?, ?>> void mergePartitionedState(StateDescriptor<S, ?> stateDescriptor);<br/> }
WindowOperator.Context作为TriggerContext的一个实现,
/**<br/> * {@code Context} is a utility for handling {@code Trigger} invocations. It can be reused<br/> * by setting the {@code key} and {@code window} fields. No internal state must be kept in<br/> * the {@code Context}<br/> */<br/> public class Context implements Trigger.OnMergeContext {<br/> protected K key; //Context对应的window上下文<br/> protected W window; protected Collection<W> mergedWindows; //onMerge中被赋值 @SuppressWarnings("unchecked")<br/> public <S extends State> S getPartitionedState(StateDescriptor<S, ?> stateDescriptor) {<br/> try {<br/> return WindowOperator.this.getPartitionedState(window, windowSerializer, stateDescriptor); //从backend里面读出改window的状态,即window buffer<br/> } catch (Exception e) {<br/> throw new RuntimeException("Could not retrieve state", e);<br/> }<br/> } @Override<br/> public <S extends MergingState<?, ?>> void mergePartitionedState(StateDescriptor<S, ?> stateDescriptor) {<br/> if (mergedWindows != null && mergedWindows.size() > 0) {<br/> try {<br/> WindowOperator.this.getStateBackend().mergePartitionedStates(window, //在backend层面把mergedWindows merge到window中<br/> mergedWindows,<br/> windowSerializer,<br/> stateDescriptor);<br/> } catch (Exception e) {<br/> throw new RuntimeException("Error while merging state.", e);<br/> }<br/> }<br/> } @Override<br/> public void registerProcessingTimeTimer(long time) {<br/> Timer<K, W> timer = new Timer<>(time, key, window);<br/> // make sure we only put one timer per key into the queue<br/> if (processingTimeTimers.add(timer)) {<br/> processingTimeTimersQueue.add(timer);<br/> //If this is the first timer added for this timestamp register a TriggerTask<br/> if (processingTimeTimerTimestamps.add(time, 1) == 0) { //如果这个window是第一次注册的话<br/> ScheduledFuture<?> scheduledFuture = WindowOperator.this.registerTimer(time, WindowOperator.this); //对于processTime必须注册定时器主动触发<br/> processingTimeTimerFutures.put(time, scheduledFuture);<br/> }<br/> }<br/> } @Override<br/> public void registerEventTimeTimer(long time) {<br/> Timer<K, W> timer = new Timer<>(time, key, window);<br/> if (watermarkTimers.add(timer)) {<br/> watermarkTimersQueue.add(timer);<br/> }<br/> } //封装一遍trigger的接口,并把self作为context传入trigger的接口中<br/> public TriggerResult onElement(StreamRecord<IN> element) throws Exception {<br/> return trigger.onElement(element.getValue(), element.getTimestamp(), window, this);<br/> } public TriggerResult onProcessingTime(long time) throws Exception {<br/> return trigger.onProcessingTime(time, window, this);<br/> } public TriggerResult onEventTime(long time) throws Exception {<br/> return trigger.onEventTime(time, window, this);<br/> } public TriggerResult onMerge(Collection<W> mergedWindows) throws Exception {<br/> this.mergedWindows = mergedWindows;<br/> return trigger.onMerge(window, this);<br/> } }
Evictor
/**<br/> * An {@code Evictor} can remove elements from a pane before it is being processed and after<br/> * window evaluation was triggered by a<br/> * {@link org.apache.flink.streaming.api.windowing.triggers.Trigger}.<br/> *<br/> * <p><br/> * A pane is the bucket of elements that have the same key (assigned by the<br/> * {@link org.apache.flink.api.java.functions.KeySelector}) and same {@link Window}. An element can<br/> * be in multiple panes of it was assigned to multiple windows by the<br/> * {@link org.apache.flink.streaming.api.windowing.assigners.WindowAssigner}. These panes all<br/> * have their own instance of the {@code Evictor}.<br/> *<br/> * @param <T> The type of elements that this {@code Evictor} can evict.<br/> * @param <W> The type of {@link Window Windows} on which this {@code Evictor} can operate.<br/> */<br/> public interface Evictor<T, W extends Window> extends Serializable { /**<br/> * Computes how many elements should be removed from the pane. The result specifies how<br/> * many elements should be removed from the beginning.<br/> *<br/> * @param elements The elements currently in the pane.<br/> * @param size The current number of elements in the pane.<br/> * @param window The {@link Window}<br/> */<br/> int evict(Iterable<StreamRecord<T>> elements, int size, W window);<br/> }
Evictor的目的就是在Trigger fire后,但在element真正被处理前,从pane中remove掉一些数据
比如你虽然是每小时触发一次,但是只是想处理最后10分钟的数据,而不是所有数据。。。
CountEvictor
/**<br/> * An {@link Evictor} that keeps only a certain amount of elements.<br/> *<br/> * @param <W> The type of {@link Window Windows} on which this {@code Evictor} can operate.<br/> */<br/> public class CountEvictor<W extends Window> implements Evictor<Object, W> {<br/> private static final long serialVersionUID = 1L; private final long maxCount; private CountEvictor(long count) {<br/> this.maxCount = count;<br/> } @Override<br/> public int evict(Iterable<StreamRecord<Object>> elements, int size, W window) {<br/> if (size > maxCount) {<br/> return (int) (size - maxCount);<br/> } else {<br/> return 0;<br/> }<br/> } /**<br/> * Creates a {@code CountEvictor} that keeps the given number of elements.<br/> *<br/> * @param maxCount The number of elements to keep in the pane.<br/> */<br/> public static <W extends Window> CountEvictor<W> of(long maxCount) {<br/> return new CountEvictor<>(maxCount);<br/> }<br/> }
初始化count,表示想保留多少elements(from end)
evict返回需要删除的elements数目(from begining)
如果element数大于保留数,我们需要删除size – maxCount(from begining)
反之,就全保留
TimeEvictor
/**<br/> * An {@link Evictor} that keeps elements for a certain amount of time. Elements older<br/> * than {@code current_time - keep_time} are evicted.<br/> *<br/> * @param <W> The type of {@link Window Windows} on which this {@code Evictor} can operate.<br/> */<br/> public class TimeEvictor<W extends Window> implements Evictor<Object, W> {<br/> private static final long serialVersionUID = 1L; private final long windowSize; public TimeEvictor(long windowSize) {<br/> this.windowSize = windowSize;<br/> } @Override<br/> public int evict(Iterable<StreamRecord<Object>> elements, int size, W window) {<br/> int toEvict = 0;<br/> long currentTime = Iterables.getLast(elements).getTimestamp();<br/> long evictCutoff = currentTime - windowSize;<br/> for (StreamRecord<Object> record: elements) {<br/> if (record.getTimestamp() > evictCutoff) {<br/> break;<br/> }<br/> toEvict++;<br/> }<br/> return toEvict;<br/> }<br/> }
TimeEvictor设置需要保留的时间,
用最后一条的时间作为current,current-windowSize,作为界限,小于这个时间的要evict掉
转发申明:
本文转自互联网,由小站整理并发布,在于分享相关技术和知识。版权归原作者所有,如有侵权,请联系本站 top8488@163.com,将在24小时内删除。谢谢