wuchong commented on a change in pull request #8244: [FLINK-11945] [table-runtime-blink] Support over aggregation for blink streaming runtime URL: https://github.com/apache/flink/pull/8244#discussion_r281022808
########## File path: flink-table/flink-table-runtime-blink/src/main/java/org/apache/flink/table/runtime/aggregate/RowTimeUnboundedOver.java ########## @@ -0,0 +1,266 @@ +/* + * 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.flink.table.runtime.aggregate; + +import org.apache.flink.api.common.state.MapState; +import org.apache.flink.api.common.state.MapStateDescriptor; +import org.apache.flink.api.common.state.ValueState; +import org.apache.flink.api.common.state.ValueStateDescriptor; +import org.apache.flink.api.common.typeinfo.Types; +import org.apache.flink.api.java.typeutils.ListTypeInfo; +import org.apache.flink.configuration.Configuration; +import org.apache.flink.streaming.api.functions.KeyedProcessFunction; +import org.apache.flink.table.api.TableConfig; +import org.apache.flink.table.dataformat.BaseRow; +import org.apache.flink.table.dataformat.JoinedRow; +import org.apache.flink.table.dataview.PerKeyStateDataViewStore; +import org.apache.flink.table.generated.AggsHandleFunction; +import org.apache.flink.table.generated.GeneratedAggsHandleFunction; +import org.apache.flink.table.type.InternalType; +import org.apache.flink.table.typeutils.BaseRowTypeInfo; +import org.apache.flink.util.Collector; + +import org.slf4j.Logger; +import org.slf4j.LoggerFactory; + +import java.util.ArrayList; +import java.util.Iterator; +import java.util.LinkedList; +import java.util.List; +import java.util.ListIterator; +import java.util.Map; + +/** + * A ProcessFunction to support unbounded event-time over-window. + */ +public class RowTimeUnboundedOver<K> extends + ProcessFunctionWithCleanupState<K, BaseRow, BaseRow> { + private static final Logger LOG = LoggerFactory.getLogger(RowTimeUnboundedOver.class); + + private GeneratedAggsHandleFunction genAggsHandler; + private InternalType[] accTypes; + private InternalType[] inputFieldTypes; + private int rowTimeIdx; + + protected JoinedRow output; + // state to hold the accumulators of the aggregations + private ValueState<BaseRow> accState; + // state to hold rows until the next watermark arrives + private MapState<Long, List<BaseRow>> inputState; + // list to sort timestamps to access rows in timestamp order + private LinkedList<Long> sortedTimestamps; + + protected AggsHandleFunction function; + + public RowTimeUnboundedOver( + GeneratedAggsHandleFunction genAggsHandler, + InternalType[] accTypes, + InternalType[] inputFieldTypes, + int rowTimeIdx, + TableConfig tableConfig) { + super(tableConfig); + this.genAggsHandler = genAggsHandler; + this.accTypes = accTypes; + this.inputFieldTypes = inputFieldTypes; + this.rowTimeIdx = rowTimeIdx; + } + + @Override + public void open(Configuration parameters) throws Exception { + LOG.debug("Compiling AggregateHelper: " + genAggsHandler.getClassName() + " \n\n" + + "Code:\n" + genAggsHandler.getCode()); + function = genAggsHandler.newInstance(getRuntimeContext().getUserCodeClassLoader()); + function.open(new PerKeyStateDataViewStore(getRuntimeContext())); + + output = new JoinedRow(); + + sortedTimestamps = new LinkedList<Long>(); + + // initialize accumulator state + BaseRowTypeInfo accTypeInfo = new BaseRowTypeInfo(accTypes); + ValueStateDescriptor<BaseRow> accStateDesc = + new ValueStateDescriptor<BaseRow>("accState", accTypeInfo); + accState = getRuntimeContext().getState(accStateDesc); + + // input element are all binary row as they are came from network + BaseRowTypeInfo inputType = new BaseRowTypeInfo(inputFieldTypes); + ListTypeInfo<BaseRow> rowListTypeInfo = new ListTypeInfo<BaseRow>(inputType); + MapStateDescriptor<Long, List<BaseRow>> inputStateDesc = new MapStateDescriptor<Long, List<BaseRow>>( + "inputState", + Types.LONG, + rowListTypeInfo); + inputState = getRuntimeContext().getMapState(inputStateDesc); + + initCleanupTimeState("RowTimeUnboundedOverCleanupTime"); + } + + /** + * Puts an element from the input stream into state if it is not late. + * Registers a timer for the next watermark. + * + * @param input The input value. + * @param ctx A {@link Context} that allows querying the timestamp of the element and getting + * a {@link TimerService} for registering timers and querying the time. The + * context is only valid during the invocation of this method, do not store it. + * @param out The collector for returning result values. + * @throws Exception + */ + @Override + public void processElement( + BaseRow input, + KeyedProcessFunction<K, BaseRow, BaseRow>.Context ctx, + Collector<BaseRow> out) throws Exception { + // register state-cleanup timer + registerProcessingCleanupTimer(ctx, ctx.timerService().currentProcessingTime()); + + Long timestamp = input.getLong(rowTimeIdx); + Long curWatermark = ctx.timerService().currentWatermark(); + + // discard late record + if (timestamp > curWatermark) { + // ensure every key just registers one timer + // default watermark is Long.Min, avoid overflow we use zero when watermark < 0 + Long triggerTs = curWatermark < 0 ? 0 : curWatermark + 1; + ctx.timerService().registerEventTimeTimer(triggerTs); + + // put row into state + List<BaseRow> rowList = inputState.get(timestamp); + if (rowList == null) { + rowList = new ArrayList<BaseRow>(); + } + rowList.add(input); + inputState.put(timestamp, rowList); + } + } + + @Override + public void onTimer( + long timestamp, + KeyedProcessFunction<K, BaseRow, BaseRow>.OnTimerContext ctx, + Collector<BaseRow> out) throws Exception { + if (isProcessingTimeTimer(ctx)) { + if (needToCleanupState(timestamp)) { + + // we check whether there are still records which have not been processed yet + Boolean noRecordsToProcess = !inputState.contains(timestamp); + if (noRecordsToProcess) { + // we clean the state + cleanupState(inputState, accState); + function.cleanup(); + } else { + // There are records left to process because a watermark has not been received yet. + // This would only happen if the input stream has stopped. So we don't need to clean up. + // We leave the state as it is and schedule a new cleanup timer + registerProcessingCleanupTimer(ctx, ctx.timerService().currentProcessingTime()); + } + } + return; + } + + Iterator<Map.Entry<Long, List<BaseRow>>> keyIterator = inputState.iterator(); + if (keyIterator != null && keyIterator.hasNext()) { + Long curWatermark = ctx.timerService().currentWatermark(); + Boolean existEarlyRecord = false; + + // sort the record timestamps + do { + Long recordTime = keyIterator.next().getKey(); + // only take timestamps smaller/equal to the watermark + if (recordTime <= curWatermark) { + insertToSortedList(recordTime); + } else { + existEarlyRecord = true; + } + } while (keyIterator.hasNext()); + + // get last accumulator + BaseRow lastAccumulator = accState.value(); + if (lastAccumulator == null) { + // initialize accumulator + lastAccumulator = function.createAccumulators(); + } + // set accumulator in function context first + function.setAccumulators(lastAccumulator); + + // emit the rows in order + while (!sortedTimestamps.isEmpty()) { + Long curTimestamp = sortedTimestamps.removeFirst(); + List<BaseRow> curRowList = inputState.get(curTimestamp); + if (curRowList != null) { + // process the same timestamp datas, the mechanism is different according ROWS or RANGE + processElementsWithSameTimestamp(curRowList, out); + } else { + // Ignore the same timestamp datas if the state is cleared already. + LOG.warn("The state is cleared because of state ttl. " + + "This will result in incorrect result. " + + "You can increase the state ttl to avoid this."); + } + inputState.remove(curTimestamp); + } + + // update acc state + lastAccumulator = function.getAccumulators(); + accState.update(lastAccumulator); + + // if are are rows with timestamp > watermark, register a timer for the next watermark + if (existEarlyRecord) { + ctx.timerService().registerEventTimeTimer(curWatermark + 1); + } + } + + // update cleanup timer + registerProcessingCleanupTimer(ctx, ctx.timerService().currentProcessingTime()); + } + + /** + * Inserts timestamps in order into a linked list. + * If timestamps arrive in order (as in case of using the RocksDB state backend) this is just + * an append with O(1). + */ + private void insertToSortedList(Long recordTimestamp) { + ListIterator<Long> listIterator = sortedTimestamps.listIterator(sortedTimestamps.size()); + Boolean isContinue = true; + while (listIterator.hasPrevious() && isContinue) { + Long timestamp = listIterator.previous(); + if (recordTimestamp >= timestamp) { + listIterator.next(); + listIterator.add(recordTimestamp); + isContinue = false; + } + } + + if (isContinue) { + sortedTimestamps.addFirst(recordTimestamp); + } + } + + /** + * Process the same timestamp datas, the mechanism is different between + * rows and range window. + */ + public void processElementsWithSameTimestamp( Review comment: I think this method should be abstract, so that the sub-classes know to override it. ---------------------------------------------------------------- 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: [email protected] With regards, Apache Git Services
