就是伪装起来的KV存储 这⾥里里我想了了好久如何给⼤大家讲HBase数据模型的物理理视图 数据模型 物理理视图 数据模型 物理理视图 或者换⼀一个⽅方式来看 • 设备id作为rowkey • ⽇日志时间作为version时间 • 其余作为10列列 1个ID, 100条⾏行行为(100个版本), 10列列 = 1000个KV对: id存了了1000次, 时间存了了1000次, 所有列列 这个很类似关系数据库⾥里里我们说⽔水平/垂直分表的意思. • 读缓存: BlockCache • 写缓存: Memstore • 写操作⽇日志: WAL • 数据⽂文件: HFile 系统组成 RegionServer 深⼊入RegionServer内部. 有两个Cache和两种⽂文件 • 先写WAL做故障恢复⽤用 • 写到Memstore中 • ⾜足够⼤大时flush到HFile • Compaction

sequence id(s): ❏ Last Pushed Sequence ID: Update this id for the specific region after replicated a WAL entry to peer cluster, it means the progress of the serial peer. ❏ Barrier: Will store the region’s each region when RS open it. ❏ Last Pushed Sequence ID VS WAL Position: ❏ Last Pushed Sequence ID is a sequence id for a given region. ❏ WAL Position is the latest replicated offset for current W-A-L hbase:meta ○ WAL Entry only has an encoded region name, but rowkey in hbase:meta is a full region name, the two different rowkey format messed up the hbase:meta. ○ Inconsistent between WAL position(ZK)

client will be allocated offheap and retained until those key values are successfully written to the WAL log and Memstore. The ConcurrentSkipListSet in Memstore does not directly store the Cell data, but

典型用户案例：Data Storage • 场景 – 用于收集并存储非结构化以及半结构化数据 – 数据存储要求可靠 – 保证数据强一致性 – 数据可被排序以便提供低延时的随机查询 • 案例 – 原始日志查询系统 – 在线指标查询系统 • 主要组件 – HBase, Flume, Sqoop 10 Postgres Conference China 2016 中国用户大会 HBase适用场景