Database Reference
In-Depth Information
Decommissioning a node
For a maintenance activity such as a hardware upgrade, node decommissioning is
required. However, at the same time, an administrator needs to ensure that there is
minimal impact on the cluster availability. To perform this decommissioning activity,
HBase provides the
graceful_stop.sh
script under the
$HBASE_HOME/bin
directory.
The preceding screenshot shows the usage of the
graceful-stop.sh
script.
The
graceful-stop.sh
script performs the following steps in order to stop
a RegionServer:
1.
Disable the region balancer.
2.
Randomly assign the regions from the RegionServer in relation to other
available servers in the cluster. It also veriies that the region is copied to
the new location before it moves the next region.
3.
Stop the RegionServer process.
Decommissioning nodes is an important management task and once the server
goes ofline, it should be removed from the region server list.
Upgrading a cluster
As Hadoop and HBase are open source, the community brings forth patches and
new releases very frequently that contain bug ixes, new features, and performance
improvements. For such cases, there is a possibility that new upgrades might or
might not be compatible with the existing versions and also might or might not
require cluster downtime. When the new upgrade is backward compatible with an
existing version, an administrator can upgrade one node at a time without bringing
down the whole cluster. This process is called rolling upgrades, as updates are rolled
on node by node. During the node upgrade, the cluster is available for use, except for
the node that is being upgraded.
