Modern Big Data Processing with Hadoop

HBase architecture in a nutshell

• The HBase cluster is comprised of one active master and one or more backup master servers
• The cluster has multiple RegionServers
• The HBase table is always large and rows are divided into partitions/shards called regions
• Each RegionServer hosts one or many regions
• The HBase catalog is known as META table, which stores the locations of table regions
• ZooKeeper stores the locations of the META table
• During a write, the client sends the put request to the HRegionServer
• Data is written to WAL
• Then data is pushed into MemStore and an acknowledgement is sent to the client
• Once enough data is accumulated in MemStore, it flushes data to the Hfile on HDFS
• The HBase compaction process activates periodically to merge ...