lct45 commented on a change in pull request #9039: URL: https://github.com/apache/kafka/pull/9039#discussion_r478728829
########## File path: streams/src/main/java/org/apache/kafka/streams/kstream/internals/KStreamSlidingWindowAggregate.java ########## @@ -0,0 +1,303 @@ +/* + * Licensed to the Apache Software Foundation (ASF) under one or more + * contributor license agreements. See the NOTICE file distributed with + * this work for additional information regarding copyright ownership. + * The ASF licenses this file to You under the Apache License, Version 2.0 + * (the "License"); you may not use this file except in compliance with + * the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ +package org.apache.kafka.streams.kstream.internals; + +import org.apache.kafka.clients.consumer.ConsumerRecord; +import org.apache.kafka.common.metrics.Sensor; +import org.apache.kafka.streams.KeyValue; +import org.apache.kafka.streams.kstream.Aggregator; +import org.apache.kafka.streams.kstream.Initializer; +import org.apache.kafka.streams.kstream.Window; +import org.apache.kafka.streams.kstream.Windowed; +import org.apache.kafka.streams.kstream.SlidingWindows; +import org.apache.kafka.streams.processor.AbstractProcessor; +import org.apache.kafka.streams.processor.Processor; +import org.apache.kafka.streams.processor.ProcessorContext; +import org.apache.kafka.streams.processor.internals.InternalProcessorContext; +import org.apache.kafka.streams.processor.internals.metrics.StreamsMetricsImpl; +import org.apache.kafka.streams.state.KeyValueIterator; +import org.apache.kafka.streams.state.TimestampedWindowStore; +import org.apache.kafka.streams.state.ValueAndTimestamp; +import org.slf4j.Logger; +import org.slf4j.LoggerFactory; +import java.util.HashSet; +import java.util.Set; + +import static org.apache.kafka.streams.processor.internals.metrics.TaskMetrics.droppedRecordsSensorOrLateRecordDropSensor; +import static org.apache.kafka.streams.processor.internals.metrics.TaskMetrics.droppedRecordsSensorOrSkippedRecordsSensor; +import static org.apache.kafka.streams.state.ValueAndTimestamp.getValueOrNull; + +public class KStreamSlidingWindowAggregate<K, V, Agg> implements KStreamAggProcessorSupplier<K, Windowed<K>, V, Agg> { + private final Logger log = LoggerFactory.getLogger(getClass()); + + private final String storeName; + private final SlidingWindows windows; + private final Initializer<Agg> initializer; + private final Aggregator<? super K, ? super V, Agg> aggregator; + + private boolean sendOldValues = false; + + public KStreamSlidingWindowAggregate(final SlidingWindows windows, + final String storeName, + final Initializer<Agg> initializer, + final Aggregator<? super K, ? super V, Agg> aggregator) { + this.windows = windows; + this.storeName = storeName; + this.initializer = initializer; + this.aggregator = aggregator; + } + + @Override + public Processor<K, V> get() { + return new KStreamSlidingWindowAggregateProcessor(); + } + + public SlidingWindows windows() { + return windows; + } + + @Override + public void enableSendingOldValues() { + sendOldValues = true; + } + + private class KStreamSlidingWindowAggregateProcessor extends AbstractProcessor<K, V> { + private TimestampedWindowStore<K, Agg> windowStore; + private TimestampedTupleForwarder<Windowed<K>, Agg> tupleForwarder; + private StreamsMetricsImpl metrics; + private InternalProcessorContext internalProcessorContext; + private Sensor lateRecordDropSensor; + private Sensor droppedRecordsSensor; + private long observedStreamTime = ConsumerRecord.NO_TIMESTAMP; + + @SuppressWarnings("unchecked") + @Override + public void init(final ProcessorContext context) { + super.init(context); + internalProcessorContext = (InternalProcessorContext) context; + metrics = internalProcessorContext.metrics(); + final String threadId = Thread.currentThread().getName(); + lateRecordDropSensor = droppedRecordsSensorOrLateRecordDropSensor( + threadId, + context.taskId().toString(), + internalProcessorContext.currentNode().name(), + metrics + ); + droppedRecordsSensor = droppedRecordsSensorOrSkippedRecordsSensor(threadId, context.taskId().toString(), metrics); + windowStore = (TimestampedWindowStore<K, Agg>) context.getStateStore(storeName); + tupleForwarder = new TimestampedTupleForwarder<>( + windowStore, + context, + new TimestampedCacheFlushListener<>(context), + sendOldValues); + } + + @Override + public void process(final K key, final V value) { + if (key == null || value == null) { + log.warn( + "Skipping record due to null key or value. value=[{}] topic=[{}] partition=[{}] offset=[{}]", + value, context().topic(), context().partition(), context().offset() + ); + droppedRecordsSensor.record(); + return; + } + + final long timestamp = context().timestamp(); + //don't process records that don't fall within a full sliding window + if (timestamp < windows.timeDifferenceMs()) { + log.warn( + "Skipping record due to early arrival. value=[{}] topic=[{}] partition=[{}] offset=[{}]", + value, context().topic(), context().partition(), context().offset() + ); + droppedRecordsSensor.record(); + return; + } + processInOrder(key, value, timestamp); + } + + public void processInOrder(final K key, final V value, final long timestamp) { + + observedStreamTime = Math.max(observedStreamTime, timestamp); + final long closeTime = observedStreamTime - windows.gracePeriodMs(); + + //store start times of windows we find + final Set<Long> windowStartTimes = new HashSet<>(); + + // aggregate that will go in the current record’s left/right window (if needed) + ValueAndTimestamp<Agg> leftWinAgg = null; + ValueAndTimestamp<Agg> rightWinAgg = null; + + //if current record's left/right windows already exist + boolean leftWinAlreadyCreated = false; + boolean rightWinAlreadyCreated = false; + + // keep the left type window closest to the record + Window latestLeftTypeWindow = null; + try ( + final KeyValueIterator<Windowed<K>, ValueAndTimestamp<Agg>> iterator = windowStore.fetch( + key, + key, + timestamp - 2 * windows.timeDifferenceMs(), + // to catch the current record's right window, if it exists, without more calls to the store + timestamp + 1) + ) { + KeyValue<Windowed<K>, ValueAndTimestamp<Agg>> next; + while (iterator.hasNext()) { + next = iterator.next(); + windowStartTimes.add(next.key.window().start()); + final long startTime = next.key.window().start(); + final long endTime = startTime + windows.timeDifferenceMs(); + + if (endTime < timestamp) { + leftWinAgg = next.value; + if (isLeftWindow(next)) { + latestLeftTypeWindow = next.key.window(); + } + } else if (endTime == timestamp) { + leftWinAlreadyCreated = true; + putAndForward(next.key.window(), next.value, key, value, closeTime, timestamp); + } else if (endTime > timestamp && startTime <= timestamp) { + rightWinAgg = next.value; + putAndForward(next.key.window(), next.value, key, value, closeTime, timestamp); + } else { + rightWinAlreadyCreated = true; + } + } + } + + //create right window for previous record + if (latestLeftTypeWindow != null) { + final long rightWinStart = latestLeftTypeWindow.end() + 1; + if (!windowStartTimes.contains(rightWinStart)) { + final TimeWindow window = new TimeWindow(rightWinStart, rightWinStart + windows.timeDifferenceMs()); + final ValueAndTimestamp<Agg> valueAndTime = ValueAndTimestamp.make(initializer.apply(), timestamp); + putAndForward(window, valueAndTime, key, value, closeTime, timestamp); + } Review comment: While the range might give a window that starts and ends before the current record's timestamp, the current record would fall into the right window of the records _within_ those windows. EX: TimeDifference = 10, record @ 30, range from (10,31). The earliest start time of a window we can have is 10, so the earliest `leftTypeWindow` we can find is from [10,20]. If there's a record at 2, it's right window would be [21,31], which our record @ 30 would fall within. Because this is true for the furthest possible record, it'll be true for the others that we find. ---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org
