mxm commented on code in PR #904: URL: https://github.com/apache/flink-kubernetes-operator/pull/904#discussion_r1819378159
########## flink-autoscaler/src/main/java/org/apache/flink/autoscaler/ParallelismAdjuster.java: ########## @@ -0,0 +1,148 @@ +/* + * 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.autoscaler; + +import org.apache.flink.autoscaler.config.AutoScalerOptions; +import org.apache.flink.autoscaler.event.AutoScalerEventHandler; +import org.apache.flink.autoscaler.topology.ShipStrategy; +import org.apache.flink.configuration.DescribedEnum; +import org.apache.flink.configuration.description.InlineElement; +import org.apache.flink.runtime.jobgraph.JobVertexID; + +import java.util.Collection; + +import static org.apache.flink.autoscaler.JobVertexScaler.SCALE_LIMITED_MESSAGE_FORMAT; +import static org.apache.flink.autoscaler.JobVertexScaler.SCALING_LIMITED; +import static org.apache.flink.autoscaler.ParallelismAdjuster.KeyGroupOrPartitionsAdjustMode.MAXIMIZE_UTILISATION; +import static org.apache.flink.autoscaler.config.AutoScalerOptions.SCALING_EVENT_INTERVAL; +import static org.apache.flink.autoscaler.topology.ShipStrategy.HASH; +import static org.apache.flink.configuration.description.TextElement.text; + +/** + * Component responsible adjusts the parallelism of a vertex. + * + * <p>When input vertex {@link ShipStrategy} is {@link ShipStrategy#HASH} or knows the number of + * current partitions of vertex. We hope to adjust the parallelism of the current vertex according + * to the number of key groups or partitions to achieve the goal of evenly distributing data among + * subtasks or maximizing utilization. + */ +public class ParallelismAdjuster { + + public static <KEY, Context extends JobAutoScalerContext<KEY>> int adjust( + JobVertexID vertex, + Context context, + AutoScalerEventHandler<KEY, Context> eventHandler, + int maxParallelism, + int numSourcePartitions, + int newParallelism, + int upperBound, + int parallelismLowerLimit, + Collection<ShipStrategy> inputShipStrategies) { + var adjustByMaxParallelismOrPartitions = + numSourcePartitions > 0 || inputShipStrategies.contains(HASH); + if (!adjustByMaxParallelismOrPartitions) { + return newParallelism; + } + var numKeyGroupsOrPartitions = + numSourcePartitions <= 0 ? maxParallelism : numSourcePartitions; + + KeyGroupOrPartitionsAdjustMode mode = + context.getConfiguration() + .get(AutoScalerOptions.SCALING_KEY_GROUP_PARTITIONS_ADJUST_MODE); + + var upperBoundForAlignment = Math.min(numKeyGroupsOrPartitions, upperBound); + + // When the shuffle type of vertex inputs contains keyBy or vertex is a source, + // we try to adjust the parallelism such that it divides + // the numKeyGroupsOrPartitions without a remainder => data is evenly spread across subtasks + for (int p = newParallelism; p <= upperBoundForAlignment; p++) { + if (numKeyGroupsOrPartitions % p == 0 + || + // When Mode is MAXIMIZE_UTILISATION , Try to find the smallest parallelism + // that can satisfy the current consumption rate. + (mode == MAXIMIZE_UTILISATION + && numKeyGroupsOrPartitions / p + < numKeyGroupsOrPartitions / newParallelism)) { Review Comment: From what I can tell, this is the only change in this PR, apart from the refactoring. The assumption here is that for cases where a parallelism such that `numKeyGroupsOrPartitions % parallelism == 0` cannot be found, we at least pick a parallelism which leads to fewer state/partition imbalance. I understand the idea behind this but I wonder whether it has the desired effect. The autoscaling algorithm isn't aware of any state/partition imbalance, it assumes linear scale through state/partition balance can be achieved. A slight adjustment to the parallelism won't drastically improve the situation. It looks like this change could help in situations where the number of partition / key groups do not have many divisors, but its also kind of hard to reason about. -- 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. To unsubscribe, e-mail: issues-unsubscr...@flink.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org
