Preparing a Node for an HBase Upgrade from 0.94

When you upgrade from HBase 0.94 to HBase 0.98 or 1.1, you need a node in your existing cluster that is not running HBase 0.94 but has connectivity to ZooKeeper and MFS services on the cluster. The following instructions explain how to workaround this issue by preparing an existing node that is running HBase for the upgrade. You cannot add a node to the cluster to serve this purpose.

To prepare a node for an HBase upgrade:
  1. Pause the Hbase load balancer via the HBase shell. For example:
    hbase(main):001:0> balance_switch false
    true
    0 row(s) in 0.3590 seconds
  2. Identify the node in your cluster that serves the lowest number of HBase regions.
  3. Stop all HBase services on that node. See the HBase documentation for details about the graceful_stop.sh script, which you can use to gracefully stop regions after pausing the load balancer. This script gradually unloads regions, allowing the node to be shut down without affecting data availability: Alternatively, you can use the MCS or a maprcli command to shut down the RegionServer on that node without first unloading regions. However, this option results in a short window of data unavailability while HBase data-recovery operations are running.
    /opt/mapr/hbase/hbase-<Version>/bin/graceful_stop.sh   hostname
  4. Continue with the main HBase upgrade procedure, starting with an installation of mapr-hbase on that node.
If your upgrade plan includes both a new version of the MapR distribution and a new version of HBase, follow these steps before upgrading the MapR software.