zhuzhurk commented on code in PR #25414: URL: https://github.com/apache/flink/pull/25414#discussion_r1846417226
########## flink-runtime/src/main/java/org/apache/flink/streaming/api/graph/DefaultStreamGraphContext.java: ########## @@ -0,0 +1,222 @@ +/* + * 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.streaming.api.graph; + +import org.apache.flink.annotation.Internal; +import org.apache.flink.runtime.jobgraph.forwardgroup.ForwardGroupComputeUtil; +import org.apache.flink.runtime.jobgraph.forwardgroup.StreamNodeForwardGroup; +import org.apache.flink.streaming.api.graph.util.ImmutableStreamGraph; +import org.apache.flink.streaming.api.graph.util.StreamEdgeUpdateRequestInfo; +import org.apache.flink.streaming.runtime.partitioner.ForwardPartitioner; +import org.apache.flink.streaming.runtime.partitioner.RescalePartitioner; +import org.apache.flink.streaming.runtime.partitioner.StreamPartitioner; + +import org.slf4j.Logger; +import org.slf4j.LoggerFactory; + +import java.util.List; +import java.util.Map; +import java.util.Set; +import java.util.stream.Collectors; + +import static org.apache.flink.util.Preconditions.checkState; + +/** Default implementation for {@link StreamGraphContext}. */ +@Internal +public class DefaultStreamGraphContext implements StreamGraphContext { + + private static final Logger LOG = LoggerFactory.getLogger(DefaultStreamGraphContext.class); + + private final StreamGraph streamGraph; + private final ImmutableStreamGraph immutableStreamGraph; + private final Map<Integer, StreamNodeForwardGroup> startAndEndNodeIdToForwardGroupMap; + private final Map<Integer, Integer> frozenNodeToStartNodeMap; + private final Map<Integer, Map<StreamEdge, NonChainedOutput>> opIntermediateOutputsCaches; + + public DefaultStreamGraphContext( + StreamGraph streamGraph, + Map<Integer, StreamNodeForwardGroup> startAndEndNodeIdToForwardGroupMap, + Map<Integer, Integer> frozenNodeToStartNodeMap, + Map<Integer, Map<StreamEdge, NonChainedOutput>> opIntermediateOutputsCaches) { + this.streamGraph = streamGraph; + this.immutableStreamGraph = new ImmutableStreamGraph(streamGraph); + this.startAndEndNodeIdToForwardGroupMap = startAndEndNodeIdToForwardGroupMap; + this.frozenNodeToStartNodeMap = frozenNodeToStartNodeMap; + this.opIntermediateOutputsCaches = opIntermediateOutputsCaches; + } + + @Override + public ImmutableStreamGraph getStreamGraph() { + return immutableStreamGraph; + } + + @Override + public boolean modifyStreamEdge(List<StreamEdgeUpdateRequestInfo> requestInfos) { + // We first verify the legality of all requestInfos to ensure that all requests can be + // modified atomically. + for (StreamEdgeUpdateRequestInfo requestInfo : requestInfos) { + if (!modifyStreamEdgeValidate(requestInfo)) { + return false; + } + } + + for (StreamEdgeUpdateRequestInfo requestInfo : requestInfos) { + StreamEdge targetEdge = + getStreamEdge( + requestInfo.getSourceId(), + requestInfo.getTargetId(), + requestInfo.getEdgeId()); + StreamPartitioner<?> newPartitioner = requestInfo.getOutputPartitioner(); + if (newPartitioner != null) { + modifyOutputPartitioner(targetEdge, newPartitioner); + } + } + + return true; + } + + private boolean modifyStreamEdgeValidate(StreamEdgeUpdateRequestInfo requestInfo) { + Integer sourceNodeId = requestInfo.getSourceId(); + Integer targetNodeId = requestInfo.getTargetId(); + + StreamEdge targetEdge = getStreamEdge(sourceNodeId, targetNodeId, requestInfo.getEdgeId()); + + if (targetEdge == null) { + return false; + } + + // Modification is not allowed when the subscribing output is reused. + Map<StreamEdge, NonChainedOutput> opIntermediateOutputs = + opIntermediateOutputsCaches.get(sourceNodeId); + NonChainedOutput output = + opIntermediateOutputs != null ? opIntermediateOutputs.get(targetEdge) : null; + if (output != null) { + Set<StreamEdge> consumerStreamEdges = + opIntermediateOutputs.entrySet().stream() + .filter(entry -> entry.getValue().equals(output)) + .map(Map.Entry::getKey) + .collect(Collectors.toSet()); + if (consumerStreamEdges.size() != 1) { + LOG.info( + "Modification for edge {} is not allowed as the subscribing output is reused.", + targetEdge); + return false; + } + } + + if (frozenNodeToStartNodeMap.containsKey(targetNodeId)) { + LOG.info( + "Modification for edge {} is not allowed as the target node with id {} is in frozen list.", + targetEdge, + targetNodeId); + return false; + } + + StreamPartitioner<?> newPartitioner = requestInfo.getOutputPartitioner(); + if (newPartitioner != null + && targetEdge.getPartitioner().getClass().equals(ForwardPartitioner.class)) { + LOG.info( + "Modification for edge {} is not allowed as the origin partitioner is ForwardPartitioner.", + targetEdge); + return false; + } + + return true; + } + + private void modifyOutputPartitioner( + StreamEdge targetEdge, StreamPartitioner<?> newPartitioner) { + if (newPartitioner == null || targetEdge == null) { + return; + } + Integer sourceNodeId = targetEdge.getSourceId(); + Integer targetNodeId = targetEdge.getTargetId(); + + StreamPartitioner<?> oldPartitioner = targetEdge.getPartitioner(); + + targetEdge.setPartitioner(newPartitioner); + + // For non-chainable edges, we change the ForwardPartitioner to RescalePartitioner to avoid + // limiting the parallelism of the downstream node by the forward edge. + // 1. If the upstream job vertex is created. + if (targetEdge.getPartitioner() instanceof ForwardPartitioner + && frozenNodeToStartNodeMap.containsKey(sourceNodeId)) { + targetEdge.setPartitioner(new RescalePartitioner<>()); + } + // 2. If the source and target are non-chainable. + if (targetEdge.getPartitioner() instanceof ForwardPartitioner + && !StreamingJobGraphGenerator.isChainable(targetEdge, streamGraph)) { + targetEdge.setPartitioner(new RescalePartitioner<>()); + } + // 3. If the forward group cannot be merged. + if (targetEdge.getPartitioner() instanceof ForwardPartitioner + && !mergeForwardGroups(sourceNodeId, targetNodeId)) { + targetEdge.setPartitioner(new RescalePartitioner<>()); + } + + Map<StreamEdge, NonChainedOutput> opIntermediateOutputs = + opIntermediateOutputsCaches.get(sourceNodeId); + NonChainedOutput output = + opIntermediateOutputs != null ? opIntermediateOutputs.get(targetEdge) : null; + if (output != null) { + output.setPartitioner(targetEdge.getPartitioner()); + } + LOG.info( + "The original partitioner of the edge {} is: {} , requested change to: {} , and finally modified to: {}.", + targetEdge, + oldPartitioner, + newPartitioner, + targetEdge.getPartitioner()); + } + + private boolean mergeForwardGroups(Integer sourceNodeId, Integer targetNodeId) { + StreamNodeForwardGroup sourceForwardGroup = + startAndEndNodeIdToForwardGroupMap.get(sourceNodeId); + StreamNodeForwardGroup targetForwardGroup = + startAndEndNodeIdToForwardGroupMap.get(targetNodeId); + if (sourceForwardGroup == null || targetForwardGroup == null) { + return false; + } + + if (!ForwardGroupComputeUtil.canTargetMergeIntoSourceForwardGroup( + sourceForwardGroup, targetForwardGroup)) { + return false; + } + + // sanity check Review Comment: The "sanity check" here is a bit misleading because here is not only a sanity check but also a critical production operation. Maybe remove the above check as mentioned in the following comment, and just returns false if the merge method returns false? -- 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: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
