wchevreuil commented on a change in pull request #3488: URL: https://github.com/apache/hbase/pull/3488#discussion_r675663329
########## File path: hbase-server/src/main/java/org/apache/hadoop/hbase/regionserver/DirectStoreFSWriteStrategy.java ########## @@ -0,0 +1,258 @@ +/** + * 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.hadoop.hbase.regionserver; + +import org.apache.hadoop.fs.FileSystem; +import org.apache.hadoop.fs.Path; +import org.apache.hadoop.hbase.client.ConnectionFactory; +import org.apache.hadoop.hbase.client.RegionInfo; +import org.apache.yetus.audience.InterfaceAudience; + +import java.io.IOException; +import java.util.ArrayList; +import java.util.HashMap; +import java.util.List; +import java.util.Map; +import java.util.concurrent.ConcurrentHashMap; + +/** + * <code>HRegionFileSystemWriteStrategy</code> implementation to be used in combination with + * <code>PersistedStoreEngine</code> to avoid renames when splitting and merging regions. + * + * To use it, define the following properties under master configuration: + * 1) <property> + * <name>hbase.hregion.file.write.strategy</name> + * <value>org.apache.hadoop.hbase.regionserver.DirectStoreFSWriteStrategy</value> + * </property> + * 2) <property> + * <name>hbase.hregion.merge.strategy</name> + * <value>org.apache.hadoop.hbase.master.assignment.DirectStoreMergeRegionsStrategy</value> + * </property> + * + * This will create the resulting merging and splitting regions directory straight under + * the table dir, instead of creating it under the temporary ".tmp" or ".merges" dirs, + * as done by the default implementation. + */ +@InterfaceAudience.Private +public class DirectStoreFSWriteStrategy extends HRegionFileSystemWriteStrategy { + private StoreFilePathAccessor accessor; + private Map<String, Map<String,List<Path>>> regionSplitReferences = new ConcurrentHashMap<>(); + private Map<String, List<Path>> mergeReferences = new HashMap(); + + public DirectStoreFSWriteStrategy(HRegionFileSystem fileSystem) throws IOException { + super(fileSystem); + this.accessor = StoreFileTrackingUtils.createStoreFilePathAccessor(fileSystem.conf, + ConnectionFactory.createConnection(fileSystem.conf)); + } + + /** + * The parent directory where to create the splits dirs is + * the table directory itself, in this case. + * @return Path representing the table directory. + */ + @Override + public Path getParentSplitsDir() { + return fileSystem.getTableDir(); + } + + /** + * The parent directory where to create the merge dir is + * the table directory itself, in this case. + * @return Path representing the table directory. + */ + @Override + public Path getParentMergesDir() { + return fileSystem.getTableDir(); + } + + /** + * Creates the directories for the respective split daughters directly under the + * table directory, instead of default behaviour of doing it under temp dirs, initially. + * @param daughterA the first half of the split region + * @param daughterB the second half of the split region + * + * @throws IOException if directories creation fails. + */ + @Override + public void createSplitsDir(RegionInfo daughterA, RegionInfo daughterB) + throws IOException { + Path splitdir = getParentSplitsDir(); + // splitDir doesn't exists now. No need to do an exists() call for it. + if (!fileSystem.getFileSystem().exists(splitdir)) { + throw new IOException("Table dir for splitting region not found: " + splitdir); + } + Path daughterADir = getSplitsDir(daughterA); + if (!fileSystem.createDir(daughterADir)) { + throw new IOException("Failed create of " + daughterADir); + } + Path daughterBDir = getSplitsDir(daughterB); + if (!fileSystem.createDir(daughterBDir)) { + throw new IOException("Failed create of " + daughterBDir); + } + } + + /** + * Just validates that merges parent, the actual table dir in this case, exists. + * @throws IOException if table dir doesn't exist. + */ + @Override + public void createMergesDir() throws IOException { + //When writing directly, avoiding renames, merges parent is the table dir itself, so it + // should exist already, so just validate it exist then do nothing + Path mergesdir = getParentMergesDir(); + if (!fileSystem.fs.exists(mergesdir)) { + throw new IOException("Table dir for merging region not found: " + mergesdir); + } + } + + /** + * Wraps <code>super.splitStoreFile</code>, so that it can map the resulting split reference to + * the related split region and column family, in order to add this to 'storefile' system table + * for the tracking logic of PersisedStoreFileManager later on <code>commitDaughterRegion</code> + * method. + * @param parentRegionInfo {@link RegionInfo} of the parent split region. + * @param daughterRegionInfo {@link RegionInfo} of the resulting split region. + * @param familyName Column Family Name + * @param f File to split. + * @param splitRow Split Row + * @param top True if we are referring to the top half of the hfile. + * @param splitPolicy A split policy instance; be careful! May not be full populated; e.g. if + * this method is invoked on the Master side, then the RegionSplitPolicy will + * NOT have a reference to a Region. + * @param fs FileSystem instance for creating the actual reference file. + * @return + * @throws IOException + */ + @Override + public Path splitStoreFile(RegionInfo parentRegionInfo, RegionInfo daughterRegionInfo, + String familyName, HStoreFile f, byte[] splitRow, + boolean top, RegionSplitPolicy splitPolicy, FileSystem fs) throws IOException { + Path path = super.splitStoreFile(parentRegionInfo, daughterRegionInfo, + familyName, f, splitRow, top, splitPolicy, fs); + Map<String,List<Path>> splitReferences = regionSplitReferences. + get(daughterRegionInfo.getEncodedName()); + if(splitReferences==null){ + splitReferences = new HashMap<>(); + regionSplitReferences.put(daughterRegionInfo.getEncodedName(), splitReferences); + } + List<Path> references = splitReferences.get(familyName); + if(references==null){ + references = new ArrayList<>(); + splitReferences.put(familyName,references); + } + references.add(path); + return path; + } + + /** + * Wraps <code>super.mergeStoreFile</code>, so that it can map the resulting merge reference to + * the related merged region and column family, in order to add this to 'storefile' system table + * for the tracking logic of PersisedStoreFileManager later in <code>commitMergedRegion</code>. + * @param regionToMerge {@link RegionInfo} for one of the regions being merged. + * @param mergedRegion {@link RegionInfo} of the merged region + * @param familyName Column Family Name + * @param f File to create reference. + * @param mergedDir + * @param fs FileSystem instance for creating the actual reference file. + * @return + * @throws IOException + */ + @Override + public Path mergeStoreFile(RegionInfo regionToMerge, RegionInfo mergedRegion, String familyName, + HStoreFile f, Path mergedDir, FileSystem fs) throws IOException { +// Path path = super.mergeStoreFile(regionToMerge, mergedRegion, familyName, f, mergedDir, fs); + + Path path = (this.fileSystem.regionInfoForFs.equals(regionToMerge)) ? + super.mergeStoreFile(mergedRegion, regionToMerge, familyName, f, mergedDir, fs) + : super.mergeStoreFile(regionToMerge, mergedRegion, familyName, f, mergedDir, fs); + + List<Path> referenceFiles = mergeReferences.get(familyName); + if(referenceFiles==null){ + referenceFiles = new ArrayList<>(); + mergeReferences.put(familyName, referenceFiles); + } + referenceFiles.add(path); + return path; + } + + /** + * Do nothing. Here the split dir is the table dir itself, we cannot delete it. + * @throws IOException + */ + @Override + public void cleanupSplitsDir() throws IOException { + //do nothing, the split dir is the store dir itself, so we cannot delete it. + } + + /** + * Do nothing. Here the merge dir is the table dir itself, we cannot delete it. + * @throws IOException + */ + @Override + public void cleanupMergesDir() throws IOException { + //do nothing, the merges dir is the store dir itself, so we cannot delete it. + } Review comment: On a second thought about this, it doesn't help implement anything here. As mentioned above, the split cleanups only get called at region opening, which is fine for the default strategy, that will find rubbish under a valid, opening region dir ".split" dir. But in the context of direct store, those cleanup would only be called for splits that succeeded, which is not what we are seeking. And adding an additional check that queries system table every region open could be terrible. -- 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...@hbase.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org
