HBase Administration - HBase Essentials - page 123

HBase cluster (another distributed system running on top of Hadoop) administrators
need to continuously ensure that the cluster is up and operating as expected. For
such dificult tasks, HBase provides a large number of metrics that provide details
regarding their current status.
There are different solutions provided that can be further grouped into graphing and
monitoring solutions or both. Here, graphing solutions, such as Ganglia ( http://
ganglia.sourceforge.net/ ), capture the exposed metrics of a system and display
them in visual charts on the basis of time ilters such as daily, monthly, and so on.
Monitoring-based solutions, such as Nagios ( http://www.nagios.org/ ), use the JMX-
based metrics API unfolded by the HBase processes. This solution also sends an alert
to the administrators in case of problems or failures of any process within the cluster.
The HBase metrics framework
HBase inherits its monitoring APIs from the Hadoop framework and is tightly
integrated with Hadoop. Therefore, the HBase metrics framework is another reason
that HBase depends on Hadoop. In HBase, every process, including the master and
region servers, exposes a speciic set of metrics. These metrics are further regrouped
for each kind of server and the subsystems within each server.
The HBase metrics framework works by exposing metrics. These metrics are based
on the context implementations that implement the MetricsContext interface.
Multiple context implementations are provided by HBase for the generation of
data points to monitor and graph, including JMX and Ganglia. Here is a list of the
available implementations:
• GangliaContext : This exposes metrics to Ganglia.
• FileContext : This writes the metrics to a ile on disk.
• TimeStampingFileContext : This is an extension of FileContext , as it writes
the metrics to a ile on disk with the added timestamp preix to each metric
emitted.
• CompositeContext : This context is used to emit metrics to more than one
context such as using Ganglia and ile context together.
• NullContext : This context is used to switch off the HBase metrics
framework. With NullContext , HBase processes neither expose any metric
nor collect any metrics data.
• NullContextWithUpdateThread : The only difference between NullContext
and NullContextWithUpdateThread is that it allows data aggregation so
that JMX-based solutions can use it.