dbw9580 commented on a change in pull request #2084: URL: https://github.com/apache/drill/pull/2084#discussion_r446247251
########## File path: contrib/storage-ipfs/src/main/java/org/apache/drill/exec/store/ipfs/IPFSGroupScan.java ########## @@ -0,0 +1,456 @@ +/* + * Copyright (c) 2018-2020 Bowen Ding, Yuedong Xu, Liang Wang + * + * 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.drill.exec.store.ipfs; + + +import com.fasterxml.jackson.annotation.JacksonInject; +import com.fasterxml.jackson.annotation.JsonCreator; +import com.fasterxml.jackson.annotation.JsonIgnore; +import com.fasterxml.jackson.annotation.JsonProperty; +import com.fasterxml.jackson.annotation.JsonTypeName; +import org.apache.drill.shaded.guava.com.google.common.cache.LoadingCache; +import org.apache.drill.shaded.guava.com.google.common.base.Preconditions; +import org.apache.drill.shaded.guava.com.google.common.collect.ArrayListMultimap; +import org.apache.drill.shaded.guava.com.google.common.collect.ImmutableList; +import org.apache.drill.shaded.guava.com.google.common.collect.ListMultimap; +import org.apache.drill.shaded.guava.com.google.common.collect.Lists; +import io.ipfs.api.MerkleNode; +import io.ipfs.multihash.Multihash; +import org.apache.drill.common.exceptions.ExecutionSetupException; +import org.apache.drill.common.expression.SchemaPath; +import org.apache.drill.common.util.DrillVersionInfo; +import org.apache.drill.exec.coord.ClusterCoordinator; +import org.apache.drill.exec.physical.EndpointAffinity; +import org.apache.drill.exec.physical.base.AbstractGroupScan; +import org.apache.drill.exec.physical.base.PhysicalOperator; +import org.apache.drill.exec.physical.base.ScanStats; +import org.apache.drill.exec.proto.CoordinationProtos.DrillbitEndpoint; +import org.apache.drill.exec.store.StoragePluginRegistry; +import org.apache.drill.exec.store.schedule.AffinityCreator; +import org.apache.drill.exec.store.schedule.AssignmentCreator; +import org.apache.drill.exec.store.schedule.CompleteWork; +import org.apache.drill.exec.store.schedule.EndpointByteMap; +import org.apache.drill.exec.store.schedule.EndpointByteMapImpl; +import org.apache.drill.shaded.guava.com.google.common.base.Stopwatch; +import org.slf4j.Logger; +import org.slf4j.LoggerFactory; + +import java.io.IOException; +import java.util.HashMap; +import java.util.LinkedHashMap; +import java.util.LinkedList; +import java.util.List; +import java.util.Map; +import java.util.Optional; +import java.util.Random; +import java.util.concurrent.ForkJoinPool; +import java.util.concurrent.RecursiveTask; +import java.util.concurrent.TimeUnit; +import java.util.stream.Collectors; + +import static org.apache.drill.exec.store.ipfs.IPFSStoragePluginConfig.IPFSTimeOut.FETCH_DATA; + +@JsonTypeName("ipfs-scan") +public class IPFSGroupScan extends AbstractGroupScan { + private static final Logger logger = LoggerFactory.getLogger(IPFSGroupScan.class); + private IPFSContext ipfsContext; + private IPFSScanSpec ipfsScanSpec; + private IPFSStoragePluginConfig config; + private List<SchemaPath> columns; + + private static long DEFAULT_NODE_SIZE = 1000l; + + private ListMultimap<Integer, IPFSWork> assignments; + private List<IPFSWork> ipfsWorkList = Lists.newArrayList(); + private Map<String, List<IPFSWork>> endpointWorksMap; + private List<EndpointAffinity> affinities; + + @JsonCreator + public IPFSGroupScan(@JsonProperty("IPFSScanSpec") IPFSScanSpec ipfsScanSpec, + @JsonProperty("IPFSStoragePluginConfig") IPFSStoragePluginConfig ipfsStoragePluginConfig, + @JsonProperty("columns") List<SchemaPath> columns, + @JacksonInject StoragePluginRegistry pluginRegistry) throws IOException, ExecutionSetupException { + this( + ((IPFSStoragePlugin) pluginRegistry.getPlugin(ipfsStoragePluginConfig)).getIPFSContext(), + ipfsScanSpec, + columns + ); + } + + public IPFSGroupScan(IPFSContext ipfsContext, + IPFSScanSpec ipfsScanSpec, + List<SchemaPath> columns) { + super((String) null); + this.ipfsContext = ipfsContext; + this.ipfsScanSpec = ipfsScanSpec; + this.config = ipfsContext.getStoragePluginConfig(); + logger.debug("GroupScan constructor called with columns {}", columns); + this.columns = columns == null || columns.size() == 0? ALL_COLUMNS : columns; + init(); + } + + private void init() { + IPFSHelper ipfsHelper = ipfsContext.getIPFSHelper(); + ipfsHelper.setMaxPeersPerLeaf(config.getMaxNodesPerLeaf()); + ipfsHelper.setTimeouts(config.getIpfsTimeouts()); + endpointWorksMap = new HashMap<>(); + + Multihash topHash = ipfsScanSpec.getTargetHash(ipfsHelper); + LoadingCache<Multihash, IPFSPeer> peerMap = ipfsContext.getIPFSPeerCache(); + + try { + //TODO detect and warn about loops/recursions in a malformed tree + class IPFSTreeFlattener extends RecursiveTask<Map<Multihash, String>> { + private Multihash hash; + private boolean isProvider; + private Map<Multihash, String> ret = new LinkedHashMap<>(); + + public IPFSTreeFlattener(Multihash hash, boolean isProvider) { + this.hash = hash; + this.isProvider = isProvider; + } + + @Override + public Map<Multihash, String> compute() { + try { + if (isProvider) { + IPFSPeer peer = peerMap.getUnchecked(hash); + ret.put(hash, peer.hasDrillbitAddress() ? peer.getDrillbitAddress().get() : null); + return ret; + } + + MerkleNode metaOrSimpleNode = ipfsHelper.timedFailure(ipfsHelper.getClient().object::links, hash, config.getIpfsTimeout(FETCH_DATA)); + if (metaOrSimpleNode.links.size() > 0) { + logger.debug("{} is a meta node", hash); + //TODO do something useful with leaf size, e.g. hint Drill about operation costs + List<Multihash> intermediates = metaOrSimpleNode.links.stream().map(x -> x.hash).collect(Collectors.toList()); + + ImmutableList.Builder<IPFSTreeFlattener> builder = ImmutableList.builder(); + for (Multihash intermediate : intermediates.subList(1, intermediates.size())) { + builder.add(new IPFSTreeFlattener(intermediate, false)); + } + ImmutableList<IPFSTreeFlattener> subtasks = builder.build(); + subtasks.forEach(IPFSTreeFlattener::fork); + + IPFSTreeFlattener first = new IPFSTreeFlattener(intermediates.get(0), false); + ret.putAll(first.compute()); + subtasks.reverse().forEach( + subtask -> ret.putAll(subtask.join()) + ); + + } else { + logger.debug("{} is a simple node", hash); + List<IPFSPeer> providers = ipfsHelper.findprovsTimeout(hash).stream() + .map(id -> + peerMap.getUnchecked(id) + ) + .collect(Collectors.toList()); + //FIXME isDrillReady may block threads + providers = providers.stream() + .filter(IPFSPeer::isDrillReady) + .collect(Collectors.toList()); + if (providers.size() < 1) { + logger.warn("No drill-ready provider found for leaf {}, adding foreman as the provider", hash); + providers.add(ipfsContext.getMyself()); + } + + logger.debug("Got {} providers for {} from IPFS", providers.size(), hash); + ImmutableList.Builder<IPFSTreeFlattener> builder = ImmutableList.builder(); + for (IPFSPeer provider : providers.subList(1, providers.size())) { + builder.add(new IPFSTreeFlattener(provider.getId(), true)); + } + ImmutableList<IPFSTreeFlattener> subtasks = builder.build(); + subtasks.forEach(IPFSTreeFlattener::fork); + + List<String> possibleAddrs = new LinkedList<>(); + Multihash firstProvider = providers.get(0).getId(); + IPFSTreeFlattener firstTask = new IPFSTreeFlattener(firstProvider, true); + String firstAddr = firstTask.compute().get(firstProvider); + if (firstAddr != null) { + possibleAddrs.add(firstAddr); + } + + subtasks.reverse().forEach( + subtask -> { + String addr = subtask.join().get(subtask.hash); + if (addr != null) { + possibleAddrs.add(addr); + } + } + ); + + if (possibleAddrs.size() < 1) { + logger.error("All attempts to find an appropriate provider address for {} have failed", hash); + throw new RuntimeException("No address found for any provider for leaf " + hash); + } else { + Random random = new Random(); + String chosenAddr = possibleAddrs.get(random.nextInt(possibleAddrs.size())); + ret.clear(); + ret.put(hash, chosenAddr); + logger.debug("Got peer host {} for leaf {}", chosenAddr, hash); + } + } + } catch (IOException e) { + throw new RuntimeException(e); + } + return ret; + } + } + + logger.debug("start to recursively expand nested IPFS hashes, topHash={}", topHash); + + Stopwatch watch = Stopwatch.createStarted(); + //FIXME parallelization width magic number, maybe a config entry? + ForkJoinPool forkJoinPool = new ForkJoinPool(config.getNumWorkerThreads()); + IPFSTreeFlattener topTask = new IPFSTreeFlattener(topHash, false); + Map<Multihash, String> leafAddrMap = forkJoinPool.invoke(topTask); + + logger.debug("Took {} ms to expand hash leaves", watch.elapsed(TimeUnit.MILLISECONDS)); + logger.debug("Iterating on {} leaves...", leafAddrMap.size()); + ClusterCoordinator coordinator = ipfsContext.getStoragePlugin().getContext().getClusterCoordinator(); + for (Multihash leaf : leafAddrMap.keySet()) { + String peerHostname = leafAddrMap.get(leaf); + + Optional<DrillbitEndpoint> oep = coordinator.getAvailableEndpoints() + .stream() + .filter(a -> a.getAddress().equals(peerHostname)) + .findAny(); + DrillbitEndpoint ep; + if (oep.isPresent()) { + ep = oep.get(); + logger.debug("Using existing endpoint {}", ep.getAddress()); + } else { + logger.debug("created new endpoint on the fly {}", peerHostname); + //TODO read ports & version info from IPFS instead of hard-coded + ep = DrillbitEndpoint.newBuilder() + .setAddress(peerHostname) + .setUserPort(31010) + .setControlPort(31011) + .setDataPort(31012) + .setHttpPort(8047) + .setVersion(DrillVersionInfo.getVersion()) + .setState(DrillbitEndpoint.State.ONLINE) + .build(); + //TODO how to safely remove endpoints that are no longer needed once the query is completed? + ClusterCoordinator.RegistrationHandle handle = coordinator.register(ep); + } + + IPFSWork work = new IPFSWork(leaf.toBase58()); + logger.debug("added endpoint {} to work {}", ep.getAddress(), work); + work.getByteMap().add(ep, DEFAULT_NODE_SIZE); + work.setOnEndpoint(ep); + + if(endpointWorksMap.containsKey(ep.getAddress())) { + endpointWorksMap.get(ep.getAddress()).add(work); + } else { + List<IPFSWork> ipfsWorks = Lists.newArrayList(); + ipfsWorks.add(work); + endpointWorksMap.put(ep.getAddress(), ipfsWorks); + } + ipfsWorkList.add(work); + } + }catch (Exception e) { + logger.debug("exception in init"); + throw new RuntimeException(e); + } + } + + private IPFSGroupScan(IPFSGroupScan that) { + super(that); + this.ipfsContext = that.ipfsContext; + this.ipfsScanSpec = that.ipfsScanSpec; + this.config = that.config; + this.assignments = that.assignments; + this.ipfsWorkList = that.ipfsWorkList; + this.endpointWorksMap = that.endpointWorksMap; + this.columns = that.columns; + } + + @JsonProperty + public List<SchemaPath> getColumns() { + return columns; + } + + @JsonIgnore + public IPFSStoragePlugin getStoragePlugin() { + return ipfsContext.getStoragePlugin(); + } + + @JsonProperty + public IPFSScanSpec getIPFSScanSpec() { + return ipfsScanSpec; + } + + @Override + public List<EndpointAffinity> getOperatorAffinity() { + if (affinities == null) { + affinities = AffinityCreator.getAffinityMap(ipfsWorkList); + } + return affinities; + } + + @Override + public int getMaxParallelizationWidth() { + DrillbitEndpoint myself = ipfsContext.getStoragePlugin().getContext().getEndpoint(); + int width; + if (endpointWorksMap.containsKey(myself.getAddress())) { + // the foreman is also going to be a minor fragment worker under a UnionExchange operator + width = ipfsWorkList.size(); + } else { + // the foreman does not hold data, so we have to force parallelization + // to make sure there is a UnionExchange operator + width = ipfsWorkList.size() + 1; + } + logger.debug("getMaxParallelizationWidth: {}", width); + return width; + } + + @Override + public void applyAssignments(List<DrillbitEndpoint> incomingEndpoints) { + logger.debug("ipfsWorkList.size() = {}", ipfsWorkList.size()); + logger.debug("endpointWorksMap: {}", endpointWorksMap); + if (endpointWorksMap.size()>1) { //偶尔还会出错？ + //incomingEndpoints是已经排好顺序的endpoints,和fragment 顺序对应 + logger.debug("Use manual assignment"); + assignments = ArrayListMultimap.create(); + for (int fragmentId = 0; fragmentId < incomingEndpoints.size(); fragmentId++) { + String address = incomingEndpoints.get(fragmentId).getAddress(); + if (endpointWorksMap.containsKey(address)) { //如果对应的节点有工作 + for (IPFSWork work : endpointWorksMap.get(address)) { + assignments.put(fragmentId, work); + } + } else //如果对应的节点没有工作安排，分配一个空work + { + + } + } + } + else //如果出问题，按照系统默认分配模式？ + { + logger.debug("Use AssignmentCreator"); + assignments = AssignmentCreator.getMappings(incomingEndpoints, ipfsWorkList); + } + + for (int i = 0; i < incomingEndpoints.size(); i++) { + logger.debug("Fragment {} on endpoint {} is assigned with works: {}", i, incomingEndpoints.get(i).getAddress(), assignments.get(i)); + } + } + + @Override + public IPFSSubScan getSpecificScan(int minorFragmentId) { + logger.debug(String.format("getSpecificScan: minorFragmentId = %d", minorFragmentId)); + List<IPFSWork> workList = assignments.get(minorFragmentId); + logger.debug("workList == null: " + (workList == null? "true": "false")); + logger.debug(String.format("workList.size(): %d", workList.size())); + + List<Multihash> scanSpecList = Lists.newArrayList(); + + for (IPFSWork work : workList) { + scanSpecList.add(work.getPartialRootHash()); + } + + return new IPFSSubScan(ipfsContext, scanSpecList, ipfsScanSpec.getFormatExtension(), columns); + } + + @Override + public ScanStats getScanStats() { + //FIXME why 100000 * size? + long recordCount = 100000 * endpointWorksMap.size(); + return new ScanStats(ScanStats.GroupScanProperty.NO_EXACT_ROW_COUNT, recordCount, 1, recordCount); + } + + @Override + public IPFSGroupScan clone(List<SchemaPath> columns){ + logger.debug("IPFSGroupScan clone {}", columns); + IPFSGroupScan cloned = new IPFSGroupScan(this); + cloned.columns = columns; + return cloned; + } + + @Override + @JsonIgnore + public boolean canPushdownProjects(List<SchemaPath> columns) { + //FIXME what does this mean? + return true; + } + + @Override + @JsonIgnore + public PhysicalOperator getNewWithChildren(List<PhysicalOperator> children) { + Preconditions.checkArgument(children.isEmpty()); + logger.debug("getNewWithChildren called"); + return new IPFSGroupScan(this); + } + + + + @Override + public String getDigest() { + return toString(); + } + + @Override + public String toString() { Review comment: Changed in 2cca7fb. ---------------------------------------------------------------- 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
