Database Reference
In-Depth Information
Figure 4.3
HDFS replication
During the course of the write the NameNode controls where replicated
blocks are created and HDFS will create a minimum number of replicated
blocks before the write is considered successful. Selecting the location for
replicas is important for both fault tolerance and availability. HDFS is
server, rack, and data center aware and is sophisticated in placement of
replicated data so that maximum fault tolerance is achieved.
NOTE
The
dfs.replication.min
property controls the minimum number
of replicas that must be created. If there is a difference between the
dfs.replication
and
dfs.replication.min
properties, the
additional replicas are created asynchronously.
