prashantabmc / meter-plugin-spark

Boundary Spark Plugin

A Boundary Meter plugin that collects metrics from the Spark MetricsServlet sink.

Prerequisites

OS	Linux	Windows	SmartOS	OS X
Supported	v	-	-	-

This plugin is compatible with Spark 1.2.1 or later.

Boundary Meter versions v4.2 or later

• To install new meter go to Settings->Installation or see instructions.
• To upgrade the meter to the latest version - see instructions.

Plugin Setup

MetricsServlet

MetricsServlet is added by default as a sink in master, worker and client driver. See the /etc/conf/metrics.properties file on your Spark installation for more details.

JMV Source

You can also enable the jvm source for instance master, worker, driver and executor to get detailed metrics of the JVM uncommenting the following lines in your /etc/conf/metrics.properties

The plugin gathers metrics from the Master and an optional running application. So you need to configure the host and port for the WebUI of the master and application process.

WebUI Configuration

By default, the WebUI for the master runs on port 18080 and, for applications the its on 4040 port. These are the default values for this parameters but you can change them based on your configuration.

Plugin Configuration Fields

Property	Description
Host	Host of the metrics endpoint on the WebUI for Master or Application.
Port	Port of the metrics endpoint on the WebUI for Master or Application.
Instance Type	Type of the instance to monitor. It can be Application or Master.
Poll Time (ms)	How often to poll for metrics in milliseconds
Source	The source to display in the legend for this instance
Debug Level	If enabled it will show additional debug output in the Plugin Console.

Metrics Collected

Metric Name	Description	Context
SPARK_MASTER_WORKERS_COUNT	The number of active workers on the master.	Master
SPARK_MASTER_APPLICATIONS_RUNNING_COUNT	Running application count on the master.	Master
SPARK_MASTER_APPLICATIONS_WAITING_COUNT	Waiting application count on the master.	Master
SPARK_MASTER_JVM_MEMORY_USED	Memory used by the JVM on the master.	Master
SPARK_MASTER_JVM_MEMORY_COMMITTED	Memory committed by the JVM on the master.	Master
SPARK_MASTER_JVM_HEAP_MEMORY_USED	Heap memory used by the JVM on the master.	Master
SPARK_MASTER_JVM_HEAP_MEMORY_USAGE	Percentage of heap memory used by the JVM on the master.	Master
SPARK_MASTER_JVM_NONHEAP_MEMORY_COMMITTED	Non-heap memory committed by the JVM on the master.	Master
SPARK_MASTER_JVM_NONHEAP_MEMORY_USED	Non-heap memory used by the JVM on the master.	Master
SPARK_MASTER_JVM_NONHEAP_MEMORY_USAGE	Percentage of non-heap memory usage by the JVM on the master.	Master
SPARK_APP_JOBS_ACTIVE	Jobs running on the application	App
SPARK_APP_JOBS_ALL	All jobs created by the application.	App
SPARK_APP_STAGES_FAILED	Failed stages for the application.	App
SPARK_APP_STAGES_RUNNING	Running stages for the application.	App
SPARK_APP_STAGES_WAITING	Waiting stages for the application.	App
SPARK_APP_BLKMGR_DISK_SPACE_USED	Block manager disk space used	App
SPARK_APP_BLKMGR_MEMORY_USED	Block manager memory used	App
SPARK_APP_BLKMGR_MEMORY_FREE	Block manager remaining memory	App
SPARK_APP_JVM_MEMORY_COMMITTED	Memory committed by the JVM of the app	App
SPARK_APP_JVM_MEMORY_USED	Memory used by the JVM of the app	App
SPARK_APP_JVM_HEAP_MEMORY_COMMITTED	Heap memory committed by the JVM of the app	App
SPARK_APP_JVM_HEAP_MEMORY_USED	Heap memory used by the JVM of the app	App
SPARK_APP_JVM_HEAP_MEMORY_USAGE	Percentage of heap memory in use by the JVM of the app	App
SPARK_APP_JVM_NOHEAP_MEMORY_COMMITTED	Non-heap memory committed by the JVM of the app	App
SPARK_APP_JVM_NONHEAP_MEMORY_USED	Non-heap memory used by the JVM of the app	App
SPARK_APP_JVM_NONHEAP_MEMORY_USAGE	Percentage of non-heap memory in use by the JVM of the app	App

Dashboards

• Spark Master
• Spark Application

References

http://spark.apache.org/docs/1.2.1/monitoring.html