腾讯TXSQL赋能云分布式数据库CynosDB.pptx 26页

腾讯T X S Q L 赋能云分布式数据库C y n o s D BPART01架构介绍PART02机遇&挑战PART03实现&优化PART04产品演进01架构介绍TXSQLRedundancyMasterSlaveI/OMigrationDiDsikskDiDsikskUpgradeBackup&RecoveryDiskRTO/RPODiskCOSScalabilityDataBinlogRedoFileCynosDBCBSMasterReplicaFewer I/OLog is databaseTXStoreCellCellCellReliabilityAvailabilityDiskScalabilityDiskCOSThroughputDataBinlogRedoFileLog Traffic1、Send log recordInnoDB2、Persist log recordStorage Client3、Ack for engine③①④4、Apply log records into blockHot Log5、Advance base blockData Block②⑥6、Garbage collect old versionsCell⑤AsyncSync02机遇& 挑战Feature Shared storage I/O offload Stateless Computable storage Parallel & AsynchronousBenefit StorageSharded Log/DataMultiple CopiesRecoveryBackupStorage Pool Light EngineOptional BinlogFile SystemFlush PageDouble WriteCheckpointChange BufferChallenge I/O GC Consistency DDL03实现& 优化Asynchronous Group CommitWait for log flushedStartOperationsCommitFlush LogFinish………Race conditionContext switchStartOperationsCommitWait LogFinishThread poolFinishCommitRe turnStartOperationsBackground threadsBGThreads……FinishLock-free queueCommitRe turnStartOperationsFinishContext holdingLogging Parallelly Background logging threads Lock-free buffer Easily serializeUser ThreadUser ThreadUser ThreadIn progressDoneDoneLSNFlusherWriterLog FlowTYPESpace IDPage noContentHeaderBlockTailMTR1 MTR2 MTR3 Log flow Compact format Remove unnecessary log Transfer / Archive （Reduce 20%-30%）MTR4RIOSync R/WAsync R/WRIO QueueRIO QueueRIO Handle Threadcellcellcellcellcell Lock-free Parallel AsynchronousReplicaHash tableApply ThreadApply ThreadHash tableBuffer PoolScan ThreadApply ThreadHash tableApply ThreadHash tableLog flow Transaction Delay Transaction Consistency Structure ConsistencyBatch OperationParallelMillisecond LevelDelayedRC/SIBtreeSegmentUndo&DataGCMasterReplica Replica Master Replica ReplicaGC stateTXStorePurgeCheckpointDataLogCellCellCell Cluster oldest TRX Oldest TRX Combined-Purge Purge Advance page version Checkpoint Combined-CheckpointMetaData