Compaction – hbase.hregion.majorcompaction = 604800000 (缺省 值为一周) – hbase.hregion.majorcompaction.jitter = 0.5 (缺省值 为半周) • 执行过程非常长，且非常耗资源 – 无法控制只在合适的时间执行 • 建议在生产环境禁用计划Major Compaction， 通过命令行手工触发或自己进行物理数据删除

Block into ByteBuffer directly. 1. Background For reducing the Java GC impact to p99/p999 RPC latency, HBase 2.x has made an offheap read and write path. The KV are allocated from the JVM offheap, and (5 Nodes) Picture.2 G1GC related metrics in 100% Get Case In above pictures, the p999 latency is almost the same as G1GC STW cost (~100ms). After HBASE-11425​ , almost all memory allocations decrease, when cacheHitRatio is 100%, they should have no much difference in both throughput and latency. Case.1 Disabled BlockCache Configuration​: here we choose the buffer.size=130KB, because

happen “slow tcp connection” ● Enable Short Circuit Read ○ P99~100ms, P999~250ms. ○ Both QPS and latency are more stable. ReadRawCells / ResponseCells 398039155 / 30077 = 13234 ReadRawCells / ResponseCells

Performance Test ○ Test Async RPC by CompletableFuture.get() (Based on XiaoMi HBase0.98) ○ Proof that latency of async client is at least the same as blocking hbase client. Part-2 HBase + G1GC Tuning CMS