HBase Metrics

HBase emits Hadoop metrics.

First read up on Hadoop metrics. If you are using ganglia, the GangliaMetrics wiki page is useful read.

To have HBase emit metrics, edit $HBASE_HOME/conf/hadoop-metrics.properties and enable metric 'contexts' per plugin. As of this writing, hadoop supports file and ganglia plugins. Yes, the hbase metrics files is named hadoop-metrics rather than hbase-metrics because currently at least the hadoop metrics system has the properties filename hardcoded. Per metrics context, comment out the NullContext and enable one or more plugins instead.

If you enable the hbase context, on regionservers you'll see total requests since last metric emission, count of regions and storefiles as well as a count of memcache size. On the master, you'll see a count of the cluster's requests.

Enabling the rpc context is good if you are interested in seeing metrics on each hbase rpc method invocation (counts and time taken).

The jvm context is useful for long-term stats on running hbase jvms -- memory used, thread counts, etc. As of this writing, if more than one jvm is running emitting metrics, at least in ganglia, the stats are aggregated rather than reported per instance.