zentol commented on a change in pull request #8316: [FLINK-12369] [coordination] Implement a region failover strategy regarding the next version FailoverStrategy interfaces URL: https://github.com/apache/flink/pull/8316#discussion_r280758391
########## File path: flink-runtime/src/main/java/org/apache/flink/runtime/executiongraph/failover/flip1/RestartPipelinedRegionStrategy.java ########## @@ -0,0 +1,210 @@ +/* + * 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.runtime.executiongraph.failover.flip1; + +import org.apache.flink.annotation.VisibleForTesting; + +import org.apache.flink.runtime.scheduler.strategy.ExecutionVertexID; +import org.slf4j.Logger; +import org.slf4j.LoggerFactory; + +import java.util.ArrayList; +import java.util.Collections; +import java.util.HashMap; +import java.util.HashSet; +import java.util.IdentityHashMap; +import java.util.List; +import java.util.Map; +import java.util.Set; +import java.util.stream.Collectors; + +import static org.apache.flink.util.Preconditions.checkNotNull; + +/** + * A failover strategy that proposes to restart involved regions when a vertex fails. + * A region is defined by this strategy as tasks that communicate via pipelined data exchange. + */ +public class RestartPipelinedRegionStrategy implements FailoverStrategy { + + /** The log object used for debugging. */ + private static final Logger LOG = LoggerFactory.getLogger(RestartPipelinedRegionStrategy.class); + + /** The topology containing info about all the vertices and edges. */ + private final FailoverTopology topology; + + /** Maps execution vertex id to failover region. */ + private final Map<ExecutionVertexID, FailoverRegion> regions; + + /** + * Creates a new failover strategy to restart pipelined regions that works on the given topology. + * + * @param topology containing info about all the vertices and edges + */ + public RestartPipelinedRegionStrategy(FailoverTopology topology) { + this.topology = checkNotNull(topology); + this.regions = new HashMap<>(); + + // build regions based on the given topology + LOG.info("Start building failover regions."); + buildFailoverRegions(); + } + + // ------------------------------------------------------------------------ + // region building + // ------------------------------------------------------------------------ + + private void buildFailoverRegions() { + // currently we let a job with co-location constraints fail as one region + // putting co-located vertices in the same region with each other can be a future improvement + if (topology.containsCoLocationConstraints()) { + buildOneRegionForAllVertices(); + return; + } + + // we use the map (list -> null) to imitate an IdentityHashSet (which does not exist) + // this helps to optimize the building performance as it uses reference equality + final IdentityHashMap<FailoverVertex, HashSet<FailoverVertex>> vertexToRegion = new IdentityHashMap<>(); + + // iterate all the vertices which are topologically sorted + for (FailoverVertex vertex : topology.getFailoverVertices()) { + HashSet<FailoverVertex> currentRegion = null; + for (FailoverEdge inputEdge : vertex.getInputEdges()) { + if (inputEdge.getResultPartitionType().isPipelined()) { + final FailoverVertex producerVertex = inputEdge.getSourceVertex(); + final HashSet<FailoverVertex> producerRegion = vertexToRegion.get(producerVertex); + if (currentRegion == null) { + // use producer region as current region and add current vertex to it + if (producerRegion == null) { + throw new IllegalStateException("Producer task " + producerVertex.getExecutionVertexName() + + " failover region is null while calculating failover region for the consumer task " + + vertex.getExecutionVertexName() + ". This should be a failover region building bug."); + } + currentRegion = producerRegion; + currentRegion.add(vertex); + } else { + // check if it is the same as the producer region, if so skip the merge + // this check can significantly reduce compute complexity in All-to-All PIPELINED edge case + if (currentRegion != producerRegion) { + // merge current region and producer region + for (FailoverVertex v : producerRegion) { + vertexToRegion.put(v, currentRegion); + } + currentRegion.addAll(producerRegion); + } + } + } + } + + // generate a new region for a vertex which has no pipelined input + if (currentRegion == null) { + currentRegion = new HashSet<>(1); + currentRegion.add(vertex); + } + vertexToRegion.put(vertex, currentRegion); + } + + // find out all the distinct regions + final IdentityHashMap<HashSet<FailoverVertex>, Object> distinctRegions = new IdentityHashMap<>(); + for (HashSet<FailoverVertex> regionVertices : vertexToRegion.values()) { + distinctRegions.put(regionVertices, null); + } + + // creating all the failover regions and register them + for (HashSet<FailoverVertex> regionVertices : distinctRegions.keySet()) { + final FailoverRegion failoverRegion = new FailoverRegion(regionVertices); + for (FailoverVertex vertex : regionVertices) { + this.regions.put(vertex.getExecutionVertexID(), failoverRegion); + } + } + LOG.info("Created {} failover regions.", distinctRegions.size()); + } + + private void buildOneRegionForAllVertices() { + LOG.warn("Cannot decompose the topology into individual failover regions due to use of " + + "Co-Location constraints (iterations). Job will fail over as one holistic unit."); + + final List<FailoverVertex> allVertices = new ArrayList<>(); + for (FailoverVertex vertex : topology.getFailoverVertices()) { + allVertices.add(vertex); + } + + final FailoverRegion region = new FailoverRegion(allVertices); + for (FailoverVertex vertex : topology.getFailoverVertices()) { + regions.put(vertex.getExecutionVertexID(), region); + } + } + + // ------------------------------------------------------------------------ + // task failure handling + // ------------------------------------------------------------------------ + + /** + * Returns a set of IDs corresponding to the set of vertices that should be restarted. + * In this strategy, all task vertices in 'involved' regions are proposed to be restarted. + * The 'involved' regions are calculated with rules below: + * 1. The region containing the failed task is always involved + * 2. TODO: If an input result partition of an involved region is not available, i.e. Missing or Corrupted, + * the region containing the partition producer task is involved + * 3. TODO: If a region is involved, all of its consumer regions are involved + * + * @param executionVertexId ID of the failed task + * @param cause cause of the failure + * @return set of IDs of vertices to restart + */ + @Override + public Set<ExecutionVertexID> getTasksNeedingRestart(ExecutionVertexID executionVertexId, Throwable cause) { + final FailoverRegion failedRegion = regions.get(executionVertexId); + if (failedRegion == null) { + // TODO: show the task name in the log + throw new IllegalStateException("Can not find the failover region for task " + executionVertexId, cause); + } + + // TODO: if the failure cause is data consumption error, mark the corresponding data partition to be unavailable + + return getRegionsToRestart(failedRegion).stream().flatMap( + r -> r.getAllExecutionVertexIDs().stream()).collect(Collectors.toSet()); + } + + /** + * All 'involved' regions are proposed to be restarted. + * The 'involved' regions are calculated with rules below: + * 1. The region containing the failed task is always involved + * 2. TODO: If an input result partition of an involved region is not available, i.e. Missing or Corrupted, + * the region containing the partition producer task is involved + * 3. TODO: If a region is involved, all of its consumer regions are involved + */ + private Set<FailoverRegion> getRegionsToRestart(FailoverRegion regionToRestart) { + return Collections.singleton(regionToRestart); + + // TODO: implement backtracking logic + } + + // ------------------------------------------------------------------------ + // testing + // ------------------------------------------------------------------------ + + /** + * Returns the failover region that contains the given execution vertex. + * + * @return the failover region that contains the given execution vertex + */ + @VisibleForTesting + public FailoverRegion getFailoverRegion(ExecutionVertexID vertexID) { Review comment: can be package-private ---------------------------------------------------------------- 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 With regards, Apache Git Services
