johnyangk commented on a change in pull request #85: [NEMO-172] Implement one partition per one element partitioner URL: https://github.com/apache/incubator-nemo/pull/85#discussion_r208070263
########## File path: compiler/optimizer/src/main/java/edu/snu/nemo/compiler/optimizer/pass/compiletime/annotating/LargeShufflePartitionerPass.java ########## @@ -0,0 +1,56 @@ +/* + * Copyright (C) 2018 Seoul National University + * + * Licensed 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 edu.snu.nemo.compiler.optimizer.pass.compiletime.annotating; + +import edu.snu.nemo.common.dag.DAG; +import edu.snu.nemo.common.ir.edge.IREdge; +import edu.snu.nemo.common.ir.edge.executionproperty.CommunicationPatternProperty; +import edu.snu.nemo.common.ir.edge.executionproperty.PartitionerProperty; +import edu.snu.nemo.common.ir.vertex.IRVertex; + +import java.util.Collections; +import java.util.List; + +/** + * A pass to support Sailfish-like shuffle by tagging edges. + * This pass modifies the partitioner property from {@link edu.snu.nemo.common.ir.vertex.transform.RelayTransform} + * to write an element as a partition. + */ +public final class LargeShufflePartitionerPass extends AnnotatingPass { + /** + * Default constructor. + */ + public LargeShufflePartitionerPass() { + super(PartitionerProperty.class, Collections.singleton(CommunicationPatternProperty.class)); + } + + @Override + public DAG<IRVertex, IREdge> apply(final DAG<IRVertex, IREdge> dag) { + dag.getVertices().forEach(vertex -> { + final List<IREdge> inEdges = dag.getIncomingEdgesOf(vertex); + inEdges.forEach(edge -> { + if (edge.getPropertyValue(CommunicationPatternProperty.class).get() + .equals(CommunicationPatternProperty.Value.Shuffle)) { + dag.getOutgoingEdgesOf(edge.getDst()) + .forEach(edgeFromRelay -> { + edgeFromRelay.setProperty(PartitionerProperty.of(PartitionerProperty.Value.IncrementPartitioner)); Review comment: My understanding is that `edgeFromRelay` is OneToOne. Can you comment here why we need an `IncrementPartitioner`? ---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on 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
