This patch add the following features
+ Additional instrumentation/metrics for
+ Adds a utility layer for maintaining metrics (see hadoop.metrics.util.*)
- This makes it easier to add new metrics (fewer lines of code to add for each metrics
- Mains additional metrics for min and max for rate based metrics
- Consistency in naming the suffix and prefix for metrics (e.g. Foo_num_ops, Foo_ave_time etc)
- Provides an interface for monitoring systems like JMX to get read the metrics when one of JMX's
clients reads the metrics (without out having an update
thread that reads the metrics periodically)
+Publishes all the Name Node instrumentation/metrics via JMX, Java's standard monitoring
framework. This will allow one to connect JConsole to monitor the name node.
Note the cost is minimal - the mbeans are registered - there is not additonal cost unless someone
is actually viewing the metrics.
See NameNodeMgtMBean and RpcMgtMBean for the full list of instrumentation available for the name node as part of this patch