[ https://issues.apache.org/jira/browse/FLINK-6364?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15993346#comment-15993346 ]
ASF GitHub Bot commented on FLINK-6364: --------------------------------------- Github user StefanRRichter commented on a diff in the pull request: https://github.com/apache/flink/pull/3801#discussion_r114359825 --- Diff: flink-contrib/flink-statebackend-rocksdb/src/main/java/org/apache/flink/contrib/streaming/state/RocksDBKeyedStateBackend.java --- @@ -621,6 +692,237 @@ private static void checkInterrupted() throws InterruptedException { } } + private static class RocksDBIncrementalSnapshotOperation { + + private final RocksDBKeyedStateBackend<?> stateBackend; + + private final CheckpointStreamFactory checkpointStreamFactory; + + private final long checkpointId; + + private final long checkpointTimestamp; + + private Map<String, StreamStateHandle> baseSstFiles; + + private List<KeyedBackendSerializationProxy.StateMetaInfo<?, ?>> stateMetaInfos = new ArrayList<>(); + + private FileSystem backupFileSystem; + private Path backupPath; + + private FSDataInputStream inputStream = null; + private CheckpointStreamFactory.CheckpointStateOutputStream outputStream = null; + + // new sst files since the last completed checkpoint + private Set<String> newSstFileNames = new HashSet<>(); + + // handles to the sst files in the current snapshot + private Map<String, StreamStateHandle> sstFiles = new HashMap<>(); + + // handles to the misc files in the current snapshot + private Map<String, StreamStateHandle> miscFiles = new HashMap<>(); + + private StreamStateHandle metaStateHandle = null; + + private RocksDBIncrementalSnapshotOperation( + RocksDBKeyedStateBackend<?> stateBackend, + CheckpointStreamFactory checkpointStreamFactory, + long checkpointId, + long checkpointTimestamp) { + + this.stateBackend = stateBackend; + this.checkpointStreamFactory = checkpointStreamFactory; + this.checkpointId = checkpointId; + this.checkpointTimestamp = checkpointTimestamp; + } + + private StreamStateHandle materializeStateData(Path filePath) throws Exception { + try { + final byte[] buffer = new byte[1024]; + + FileSystem backupFileSystem = backupPath.getFileSystem(); + inputStream = backupFileSystem.open(filePath); + stateBackend.cancelStreamRegistry.registerClosable(inputStream); + + outputStream = checkpointStreamFactory + .createCheckpointStateOutputStream(checkpointId, checkpointTimestamp); + stateBackend.cancelStreamRegistry.registerClosable(outputStream); + + while (true) { + int numBytes = inputStream.read(buffer); + + if (numBytes == -1) { + break; + } + + outputStream.write(buffer, 0, numBytes); + } + + return outputStream.closeAndGetHandle(); + } finally { + if (inputStream != null) { + stateBackend.cancelStreamRegistry.unregisterClosable(inputStream); + inputStream.close(); + inputStream = null; + } + + if (outputStream != null) { + stateBackend.cancelStreamRegistry.unregisterClosable(outputStream); + outputStream.close(); + outputStream = null; + } + } + } + + private StreamStateHandle materializeMetaData() throws Exception { + try { + outputStream = checkpointStreamFactory + .createCheckpointStateOutputStream(checkpointId, checkpointTimestamp); + stateBackend.cancelStreamRegistry.registerClosable(outputStream); + + KeyedBackendSerializationProxy serializationProxy = + new KeyedBackendSerializationProxy(stateBackend.keySerializer, stateMetaInfos); + DataOutputView out = new DataOutputViewStreamWrapper(outputStream); + + serializationProxy.write(out); + + return outputStream.closeAndGetHandle(); + } finally { + if (outputStream != null) { + stateBackend.cancelStreamRegistry.unregisterClosable(outputStream); + outputStream.close(); + outputStream = null; + } + } + } + + void takeSnapshot() throws Exception { + // use the last completed checkpoint as the comparison base. + baseSstFiles = stateBackend.materializedSstFiles.get(stateBackend.lastCompletedCheckpointId); + + // save meta data + for (Map.Entry<String, Tuple2<ColumnFamilyHandle, RegisteredBackendStateMetaInfo<?, ?>>> stateMetaInfoEntry : stateBackend.kvStateInformation.entrySet()) { + + RegisteredBackendStateMetaInfo<?, ?> metaInfo = stateMetaInfoEntry.getValue().f1; + + KeyedBackendSerializationProxy.StateMetaInfo<?, ?> metaInfoProxy = + new KeyedBackendSerializationProxy.StateMetaInfo<>( + metaInfo.getStateType(), + metaInfo.getName(), + metaInfo.getNamespaceSerializer(), + metaInfo.getStateSerializer()); + + stateMetaInfos.add(metaInfoProxy); + } + + // save state data + backupPath = new Path(stateBackend.instanceBasePath.getAbsolutePath(), "chk-" + checkpointId); + backupFileSystem = backupPath.getFileSystem(); + if (backupFileSystem.exists(backupPath)) { + LOG.warn("Deleting an existing local checkpoint directory " + + backupPath + "."); + + backupFileSystem.delete(backupPath, true); + } + + // create hard links of living files in the checkpoint path + Checkpoint checkpoint = Checkpoint.create(stateBackend.db); + checkpoint.createCheckpoint(backupPath.getPath()); + } + + KeyedStateHandle materializeSnapshot() throws Exception { + // write meta data + metaStateHandle = materializeMetaData(); + + // write state data + Preconditions.checkState(backupFileSystem.exists(backupPath)); + + FileStatus[] fileStatuses = backupFileSystem.listStatus(backupPath); + if (fileStatuses != null ) { + for (FileStatus fileStatus : fileStatuses) { + Path filePath = fileStatus.getPath(); + String fileName = filePath.getName(); + + if (fileName.endsWith(SST_FILE_SUFFIX)) { + StreamStateHandle fileHandle = + baseSstFiles == null ? null : baseSstFiles.get(fileName); + + if (fileHandle == null) { + newSstFileNames.add(fileName); + fileHandle = materializeStateData(filePath); + } + + sstFiles.put(fileName, fileHandle); + } else { + StreamStateHandle fileHandle = materializeStateData(filePath); + miscFiles.put(fileName, fileHandle); + } + } + } + + stateBackend.materializedSstFiles.put(checkpointId, sstFiles); + + return new RocksDBKeyedStateHandle(stateBackend.jobId, + stateBackend.operatorIdentifier, stateBackend.keyGroupRange, + newSstFileNames, sstFiles, miscFiles, metaStateHandle); + } + + void releaseResources(boolean canceled) { + + if (inputStream != null) { + stateBackend.cancelStreamRegistry.unregisterClosable(inputStream); + try { + inputStream.close(); + } catch (Exception e) { + LOG.warn("Could not properly close the input stream.", e); + } + inputStream = null; + } + + if (outputStream != null) { + stateBackend.cancelStreamRegistry.unregisterClosable(outputStream); + try { + outputStream.close(); + } catch (Exception e) { + LOG.warn("Could not properly close the output stream.", e); + } + outputStream = null; + } + + if (backupPath != null) { + try { + if (backupFileSystem.exists(backupPath)) { + backupFileSystem.delete(backupPath, true); + } + } catch (Exception e) { + LOG.warn("Could not properly delete the checkpoint directory.", e); + } + } + + if (canceled) { + List<StateObject> statesToDiscard = new ArrayList<>(); + + if (metaStateHandle != null) { + statesToDiscard.add(metaStateHandle); + } + + statesToDiscard.addAll(miscFiles.values()); + + for (String newSstFileName : newSstFileNames) { + StreamStateHandle fileHandle = sstFiles.get(newSstFileName); + if (fileHandle != null) { + statesToDiscard.add(fileHandle); + } + } + + try { + StateUtil.bestEffortDiscardAllStateObjects(statesToDiscard); --- End diff -- This looks like we should also have some safety net mechanism or global cleanup hook in case an exception happens or the node goes down between cancelation and triggering of the discard. In this case, all knowledge about the created files is lost. I wonder if the shared state registry should perform cleanups at startup (maybe later also after incremental recoveries) for unreferenced files in the shared directory. However, this sounds like it could be an expensive operation on some file systems. > Implement incremental checkpointing in RocksDBStateBackend > ---------------------------------------------------------- > > Key: FLINK-6364 > URL: https://issues.apache.org/jira/browse/FLINK-6364 > Project: Flink > Issue Type: Sub-task > Components: State Backends, Checkpointing > Reporter: Xiaogang Shi > Assignee: Xiaogang Shi > > {{RocksDBStateBackend}} is well suited for incremental checkpointing because > RocksDB is base on LSM trees, which record updates in new sst files and all > sst files are immutable. By only materializing those new sst files, we can > significantly improve the performance of checkpointing. -- This message was sent by Atlassian JIRA (v6.3.15#6346)