An HBase edit will first be written to a region server's Write Ahead Log (WAL). After the log is written successfully,
MemStore of the region server will be updated. As WAL is a sequence file on HDFS, it will be automatically replicated to the two other DataNode servers by default, so that a single region server crash will not cause a loss of the data stored on it.
As WAL is shared by all regions deployed on the region server, the WAL needs to first be split so that it can be replayed on each relative region, in order to recover from a region server crash. HBase handles region server failover automatically by using this algorithm.
HBase has a WAL tool providing manual WAL splitting and dumping facilities. ...