背景
本文基于delta 0.7.0
spark 3.0.1
我们之前的spark delta写操作ACID事务前传–写文件基础类FileFormat/FileCommitProtocol分析分析了delta写数据的流程,但是还没分析deltalog 写数据的流程,这部分也是实现ACID的核心部分。
背景
本文基于delta 0.7.0
spark 3.0.1
我们之前的spark delta写操作ACID事务前传–写文件基础类FileFormat/FileCommitProtocol分析分析了delta写数据的流程,但是还没分析deltalog 写数据的流程,这部分也是实现ACID的核心部分。
##分析
直接到WriteIntoDelta.run
override def run(sparkSession: SparkSession): Seq[Row] = { deltaLog.withNewTransaction { txn => val actions = write(txn, sparkSession) val operation = DeltaOperations.Write(mode, Option(partitionColumns), options.replaceWhere, options.userMetadata) txn.commit(actions, operation) } Seq.empty }
我们来看一下deltaLog.withNewTrancation
方法 :
def withNewTransaction[T](thunk: OptimisticTransaction => T): T = { try { update() val txn = new OptimisticTransaction(this) OptimisticTransaction.setActive(txn) thunk(txn) } finally { OptimisticTransaction.clearActive() } }
- 首先update方法直接同步调用updateInternal用来更新当前deltalog的snapshot,具体的updateInternal如下:
val segment = getLogSegmentForVersion(currentSnapshot.logSegment.checkpointVersion) if (segment.version == currentSnapshot.version) { // Exit early if there is no new file lastUpdateTimestamp = clock.getTimeMillis() return currentSnapshot } logInfo(s"Loading version ${segment.version}" + segment.checkpointVersion.map(v => s"starting from checkpoint version $v.")) val newSnapshot = createSnapshot( segment, minFileRetentionTimestamp, segment.lastCommitTimestamp) ... currentSnapshot.uncache() currentSnapshot = newSnapshot
首先先通过getLogSegmentForVersion方法获取当前最新的snapshot,之后更新到内存
- 设置OptimisticTransaction,并在当前事务中执行当前语句
val actions = write(txn, sparkSession) val operation = DeltaOperations.Write(mode, Option(partitionColumns), options.replaceWhere, options.userMetadata) txn.commit(actions, operation)
val atcions = write(txn, sparksession)我们已经在spark delta写操作ACID事务前传–写文件基础类FileFormat/FileCommitProtocol分析分析了,即会返回Seq[AddAction],而实际的数据文件已经存储到了文件目录下
val operation = DeltaOperations.Write(mode, Option(partitionColumns)记录了这是一个delta write Operation
txn.commit(actions, operation) 是该提交delta log的关键:
def commit(actions: Seq[Action], op: DeltaOperations.Operation): Long = recordDeltaOperation( deltaLog, "delta.commit") { commitStartNano = System.nanoTime() val version = try { // Try to commit at the next version. var finalActions = prepareCommit(actions, op) // Find the isolation level to use for this commit val noDataChanged = actions.collect { case f: FileAction => f.dataChange }.forall(_ == false) val isolationLevelToUse = if (noDataChanged) { // If no data has changed (i.e. its is only being rearranged), then SnapshotIsolation // provides Serializable guarantee. Hence, allow reduced conflict detection by using // SnapshotIsolation of what the table isolation level is. SnapshotIsolation } else { Serializable } val isBlindAppend = { val dependsOnFiles = readPredicates.nonEmpty || readFiles.nonEmpty val onlyAddFiles = finalActions.collect { case f: FileAction => f }.forall(_.isInstanceOf[AddFile]) onlyAddFiles && !dependsOnFiles } if (spark.sessionState.conf.getConf(DeltaSQLConf.DELTA_COMMIT_INFO_ENABLED)) { commitInfo = CommitInfo( clock.getTimeMillis(), op.name, op.jsonEncodedValues, Map.empty, Some(readVersion).filter(_ >= 0), None, Some(isBlindAppend), getOperationMetrics(op), getUserMetadata(op)) finalActions = commitInfo +: finalActions } // Register post-commit hooks if any lazy val hasFileActions = finalActions.collect { case f: FileAction => f }.nonEmpty if (DeltaConfigs.SYMLINK_FORMAT_MANIFEST_ENABLED.fromMetaData(metadata) && hasFileActions) { registerPostCommitHook(GenerateSymlinkManifest) } val commitVersion = doCommit(snapshot.version + 1, finalActions, 0, isolationLevelToUse) logInfo(s"Committed delta #$commitVersion to ${deltaLog.logPath}") postCommit(commitVersion, finalActions) commitVersion } catch { case e: DeltaConcurrentModificationException => recordDeltaEvent(deltaLog, "delta.commit.conflict." + e.conflictType) throw e case NonFatal(e) => recordDeltaEvent( deltaLog, "delta.commit.failure", data = Map("exception" -> Utils.exceptionString(e))) throw e } runPostCommitHooks(version, actions) version }
prepareCommit用来做一些提交前的检查,以及增加一些actions,
如果是第一次提交还得增加Protocol,如{"protocol":{"minReaderVersion":1,"minWriterVersion":2}}
如metadata变化了,还得增加newMetadata,如 {"metaData":{"id":"2b2457e3-ce74-4897-abbd-04a94692304a","format":{"provider":"parquet","options":{}},"schemaString":"{\"type\":\"struct\",\"fields\":[{\"name\":\"value\",\"type\":\"string\",\"nullable\":true,\"metadata\":{}}]}","partitionColumns":[],"configuration":{},"createdTime":1609398723678}}
如果配置了spark.databricks.delta.commitInfo.enabled(默认是true)则还会增加commitInfo信息等,如{"commitInfo":{"timestamp":1609400013646,"operation":"WRITE","operationParameters":{"mode":"Append","partitionBy":"[]"},"readVersion":0,"isBlindAppend":true,"operationMetrics":{"numFiles":"1","numOutputBytes":"306","numOutputRows":"0"}}},
如果做了Presto / Athena兼容,还会注册GenerateSymlinkManifest postCommitHook,在commit成功后还会进行调用
doCommit(snapshot.version + 1, finalActions, 0, isolationLevelToUse) 同步提交最终的action到deltalog:
protected def doCommit( attemptVersion: Long, actions: Seq[Action], attemptNumber: Int, isolationLevel: IsolationLevel): Long = deltaLog.lockInterruptibly { try { ... deltaLog.store.write( deltaFile(deltaLog.logPath, attemptVersion), actions.map(_.json).toIterator) val commitTime = System.nanoTime() val postCommitSnapshot = deltaLog.update() if (postCommitSnapshot.version < attemptVersion) { recordDeltaEvent(deltaLog, "delta.commit.inconsistentList", data = Map( "committedVersion" -> attemptVersion, "currentVersion" -> postCommitSnapshot.version )) throw new IllegalStateException( s"The committed version is $attemptVersion " + s"but the current version is ${postCommitSnapshot.version}.") } // Post stats var numAbsolutePaths = 0 var pathHolder: Path = null val distinctPartitions = new mutable.HashSet[Map[String, String]] val adds = actions.collect { case a: AddFile => pathHolder = new Path(new URI(a.path)) if (pathHolder.isAbsolute) numAbsolutePaths += 1 distinctPartitions += a.partitionValues a } ... attemptVersion } catch { case e: java.nio.file.FileAlreadyExistsException => checkAndRetry(attemptVersion, actions, attemptNumber, isolationLevel) } }
deltaLog.store.write(
deltaFile(deltaLog.logPath, attemptVersion),
actions.map(_.json).toIterator)` 方法直接调用HDFSLogStore的write方法,而最终调用writeInternal方法,这里attemptVersion是当前的version+1
我们看一下writeInternal方法:
private def writeInternal(path: Path, actions: Iterator[String], overwrite: Boolean): Unit = { val fc: FileContext = try { getFileContext(path) } catch { case e: IOException if e.getMessage.contains(noAbstractFileSystemExceptionMessage) => val newException = DeltaErrors.incorrectLogStoreImplementationException(sparkConf, e) logError(newException.getMessage, newException.getCause) throw newException } if (!overwrite && fc.util.exists(path)) { // This is needed for the tests to throw error with local file system throw new FileAlreadyExistsException(path.toString) } val tempPath = createTempPath(path) var streamClosed = false // This flag is to avoid double close var renameDone = false // This flag is to save the delete operation in most of cases. val stream = fc.create( tempPath, EnumSet.of(CREATE), CreateOpts.checksumParam(ChecksumOpt.createDisabled())) try { actions.map(_ + "\n").map(_.getBytes(UTF_8)).foreach(stream.write) stream.close() streamClosed = true try { val renameOpt = if (overwrite) Options.Rename.OVERWRITE else Options.Rename.NONE fc.rename(tempPath, path, renameOpt) renameDone = true // TODO: this is a workaround of HADOOP-16255 - remove this when HADOOP-16255 is resolved tryRemoveCrcFile(fc, tempPath) } catch { case e: org.apache.hadoop.fs.FileAlreadyExistsException => throw new FileAlreadyExistsException(path.toString) } } finally { if (!streamClosed) { stream.close() } if (!renameDone) { fc.delete(tempPath, false) } } }
如果文件存在,则抛出异常。否则写入该log文件,这里注意:如果是本地文件的话,需要加同步进行rename操作,因为本地文件的rename操作,即使目标文件存在了也不会报异常,其他的文件类型,则不需要加同步。 如果deltaLog.store.write没有发生异常,则获取最新的snaphost,进行记录,返回传入的attemptVersion, 如果发生了异常,则执行checkAndRetry(attemptVersion, actions, attemptNumber, isolationLevel),进行重试,重试的时候,我们得查找以下从事务提交到失败这段时间的由其他程序提交的Actions,如果和当前actions没有冲突则继续提交,否则抛出异常
postCommit 进行checkpoint操作,每隔 10 个提交,Delta Lake 会在 _delta_log 子目录下自动生成一个 Parquet 格式的 checkpoint 文件,便于快速replays,且清除过期的deltalog,默认保存30天(这里会把addfile和removefile合并了,比如先ad
d A文件,之后remove A 合并完了A就没记录了):
def checkpoint(): Unit = recordDeltaOperation(this, "delta.checkpoint") { val checkpointMetaData = checkpoint(snapshot) val json = JsonUtils.toJson(checkpointMetaData) store.write(LAST_CHECKPOINT, Iterator(json), overwrite = true) doLogCleanup() }
- runPostCommitHooks 如果注册了postCommitHooks,就执行
至此整个delta写deltalog的流程就结束了,
整个流程如下:
update 获取最新的snapshot | v write()写入delta data | v commit 事务提交 -> prepareCommit 用来做一些提交前的检查,以及增加一些actions | v doCommit 真正写入deltalog,会一直重试直到冲突 | v postCommit 进行checkpoint操作,合并Addfile和RemoveFile,便于快速replays,且清除过期的delta log | v runPostCommitHooks 如果存在hook则执行对应的hook | v 结束