hosted by
Apache HBase Present Model
Accumulate data in Memory
o Sorted Map
o Cell data bytes in Local Allocation Buffers (LABs) in DRAM
o LABs with size 2 MB in RAM.
Also write to Write Ahead Log (WAL)
o Cell data in Volatile RAM.
o To recover from server crash
o HDFS interaction adding more latency
hsync vs hflush - HBASE-19024
Flushes as files to HDFS on reaching memstores size
o 128 MB default flush size
Replay WAL on server crash
o Data unavailable till replay completes
o Large Mean Time To Recover (MTTR)
o Takes several minutes on large cluster (Complaint from many
users like Alibaba – HBaseConAsia , Huawei)