Apache Spark¶

Plugin currently supports cluster modes Standalone and Mesos

Metadata associated with the spark plugin for collectd can be found here. The relevant code for the plugin can be found here.

Description
Requirements and Dependencies
Installation
Configuration
Usage
Metrics

DESCRIPTION¶

This is the SignalFx Apache Spark plugin. Note we currently only support cluster modes Standalone and Mesos. Follow these instructions to install the Apache Spark plugin for collectd.

The spark-collectd plugin collects metrics from Spark cluster and instances hitting endpoints specified in Spark’s Monitoring and Instrumentation documentation under REST API and Metrics: Sparkdocumentation

FEATURES¶

Built-in dashboards¶

Spark Overview: Overview of Spark cluster.

Master Process: Overview of master process (data captured from Metrics sink).

Worker Process: Overview of worker process (data captured from Metrics sink).

Spark Application: Overview of data by Spark Application and user.

Running Jobs: Overview of running jobs by Spark Application and user.

Active Stages: Overview of active stages by Spark Application and user.

Driver: Overview of driver executor by Spark Application and user.

Executors: Overview of executors (excluding driver) by Spark Application and user.

Streaming Statistics: Overview of streaming applications by Spark Application and user.

REQUIREMENTS AND DEPENDENCIES¶

Version information¶

Software	Version
collectd	4.9 or later
python	2.6 or later
spark	2.2.0 or later
Python plugin for collectd	(included with SignalFx collectd agent)

INSTALLATION¶

If you are using the new Smart Agent, see the docs for thecollectd/sparkmonitor for more information. The configuration documentation below may be helpful as well, but consult the Smart Agent repo’s docs for the exact schema.

Download collectd-spark. Place the spark_plugin.py file in /usr/share/collectd/collectd-spark
Modify the sample configuration file for this plugin to /etc/collectd/managed_config
Modify the sample configuration file as described in Configuration, below
Install the Python requirements with sudo pip install -r requirements.txt
Restart collectd

CONFIGURATION¶

Using the example configuration file 10-spark.conf as a guide, provide values for the configuration options listed below that make sense for your environment and the metrics you want to be collected and reported.

configuration option	definition	example value
ModulePath	Path on disk where collectd can find this module.	“/usr/share/collectd/collectd-spark/”
MetricsURL	URL for master or worker node if Metrics source (and therefore, by default Metrics HTTP Servlet Sink) are enabled	“http://localhost”
MasterPort	Master (webui) port to query for metrics	8080
WorkerPorts	Space-separated worker (webui) ports to query for metrics	8081 8082
Applications	Boolean indicating whether to collect application level metrics	“False”
Master	URL for master application	“http://localhost:8080”
Cluster	Your Spark cluster mode - only standalone and mesos supported	“Standalone”
EnhancedMetrics	Flag to specify whether to include additional metrics	“False”
IncludeMetrics	Metrics from enhanced metrics that can be included individually	“metric_name_1,metric_name_2”
ExcludeMetrics	Metrics from enhanced metrics that can be excluded individually	“metric_name_1,metric_name2”
Dimension	Key-value pair for a user-defined dimension	“dimension_key=dimension_value”
Dimension	Comma-separated key-value pairs for user-defined dimensions	“dimension_key1=dimension_value1,dimension_key2=dimension_value2”

Example configuration:

LoadPlugin python
<Plugin python>
  ModulePath "/usr/share/collectd/collectd-spark"

  Import spark_plugin

  <Module spark_plugin>
  MetricsURL "http://127.0.0.1"
  MasterPort 8080
  WorkerPorts 8081 8082
  Applications "True"
  Master "http://127.0.0.1:8080"
  Cluster "Standalone"
  EnhancedMetrics "True"
  ExcludeMetrics "jvm.pools.Code-Cache.committed"
  </Module>
</Plugin>

The plugin can be configured to collect metrics from multiple instances in the following manner.

LoadPlugin python
<Plugin python>
  ModulePath "/usr/share/collectd/collectd-spark"

  Import spark_plugin

  <Module spark_plugin>
    MetricsURL "http://master"
    MasterPort 8080
    Applications "True"
    Master "http://master:8080"
    Cluster "Standalone"
    Dimension "name=MASTERTEST"
    IncludeMetrics "jvm.pools.Code-Cache.committed"
  </Module>

  <Module spark_plugin>
    MetricsURL "http://worker"
    WorkerPorts 8081 8082
    Applications "False"
    Master "http://master:8080"
    Dimension "name=WORKER1TEST"
    IncludeMetrics "jvm.pools.Code-Cache.committed"
  </Module>
</Plugin>

USAGE¶

Sample of built-in dashboard in SignalFx:

Metrics corresponding to Metrics sink will contain the following dimension by default:

spark_process, either master or worker to differentiate master- and worker- specific metrics like master.apps and worker.coresFree

Metrics at the application level (endpoint /api/v1/applications) will contain the following dimension by default:

cluster, set to value corresponding to key “Cluster” in configuration file

Additional details:

plugin is always set to apache_spark

METRICS¶

To emphasize, metrics will only be collected if MetricsURL is provided and/or Applications is set to True in 10-spark.conf (please read the example configuration file for associated required keys). See usage for details.

The following are default metrics captured and sent if Metrics Sink and Applications are enabled:

jvm.total.used
jvm.total.committed
jvm.heap.used
jvm.heap.committed
jvm.non-heap.used
jvm.non-heap.committed
jvm.MarkSweepCompact.count
jvm.MarkSweepCompact.time
worker.coresFree
worker.coresUsed
worker.executors
worker.memFree_MB
worker.memUsed_MB
master.aliveWorkers
master.apps
master.waitingApps
master.workers
spark.job.num_tasks
spark.job.num_active_tasks
spark.job.num_completed_tasks
spark.job.num_skipped_tasks
spark.job.num_failed_tasks
spark.job.num_active_stages
spark.job.num_completed_stages
spark.job.num_skipped_stages
spark.job.num_failed_stages
spark.stage.executor_run_time
spark.stage.input_bytes
spark.stage.input_records
spark.stage.output_bytes
spark.stage.output_records
spark.stage.memory_bytes_spilled
spark.stage.disk_bytes_spilled
spark.driver.memory_used
spark.driver.disk_used
spark.driver.total_input_bytes
spark.driver.total_shuffle_read
spark.driver.total_shuffle_write
spark.driver.max_memory
spark.executor.memory_used
spark.executor.disk_used
spark.executor.total_input_bytes
spark.executor.total_shuffle_read
spark.executor.total_shuffle_write
spark.executor.max_memory
spark.streaming.avg_input_rate
spark.streaming.num_total_completed_batches
spark.streaming.num_active_batches
spark.streaming.num_inactive_receivers
spark.streaming.num_received_records
spark.streaming.num_processed_records
spark.streaming.avg_processing_time
spark.streaming.avg_scheduling_delay
spark.streaming.avg_total_delay

The following are metrics that can be collected and sent if EnhancedMetrics is set to “True” in configurations (see CONFIGURATION for more information):

jvm.pools.Code-Cache.used
jvm.pools.Code-Cache.committed
jvm.pools.Compressed-Class-Space.used
jvm.pools.Compressed-Class-Space.committed
jvm.pools.Metaspace.used
jvm.pools.Metaspace.committed
jvm.pools.Eden-Space.used
jvm.pools.Eden-Space.committed
jvm.pools.Survivor-Space.used
jvm.pools.Survivor-Space.committed
jvm.pools.Tenured-Gen.used
jvm.pools.Tenured-Gen.committed
HiveExternalCatalog.fileCacheHits
HiveExternalCatalog.filesDiscovered
HiveExternalCatalog.hiveClientCalls
HiveExternalCatalog.parallelListingJobCount
HiveExternalCatalog.partitionsFetched
spark.stage.shuffle_read_bytes
spark.stage.shuffle_read_records
spark.stage.shuffle_write_bytes
spark.stage.shuffle_write_records
spark.driver.rdd_blocks
spark.driver.active_tasks
spark.driver.failed_tasks
spark.driver.completed_tasks
spark.driver.total_tasks
spark.driver.total_duration
spark.executor.rdd_blocks
spark.executor.active_tasks
spark.executor.failed_tasks
spark.executor.completed_tasks
spark.executor.total_tasks
spark.executor.total_duration

Metric naming¶

<metric type>.spark.<name of metric> or <metric type><name of metric>. This is the format of default metric names reported by the plugin.

Below is a list of all metrics.

Metric Name	Brief	Type
counter.HiveExternalCatalog.counter.HiveClientCalls	Total number of client calls sent to Hive for query processing	counter
counter.HiveExternalCatalog.fileCacheHits	Total number of file level cache hits occurred	counter
counter.HiveExternalCatalog.filesDiscovered	Total number of files discovered	counter
counter.HiveExternalCatalog.parallelListingJobCount	Total number of Hive-specific jobs running in parallel	counter
counter.HiveExternalCatalog.partitionsFetched	Total number of partitions fetched	counter
counter.spark.driver.completed_tasks	Total number of completed tasks in driver mapped to a particular application	counter
counter.spark.driver.disk_used	Amount of disk used by driver mapped to a particular application	counter
counter.spark.driver.failed_tasks	Total number of failed tasks in driver mapped to a particular application	counter
counter.spark.driver.memory_used	Amount of memory used by driver mapped to a particular application	counter
counter.spark.driver.total_duration	Fraction of time spent by driver mapped to a particular application	counter
counter.spark.driver.total_input_bytes	Number of input bytes in driver mapped to a particular application	counter
counter.spark.driver.total_shuffle_read	Size read during a shuffle in driver mapped to a particular application	counter
counter.spark.driver.total_shuffle_write	Size written to during a shuffle in driver mapped to a particular application	counter
counter.spark.driver.total_tasks	Total number of tasks in driver mapped to a particular application	counter
counter.spark.executor.completed_tasks	Completed tasks across executors working for a particular application	counter
counter.spark.executor.disk_used	Amount of disk used across executors working for a particular application	counter
counter.spark.executor.failed_tasks	Failed tasks across executors working for a particular application	counter
counter.spark.executor.memory_used	Amount of memory used across executors working for a particular application	counter
counter.spark.executor.total_duration	Fraction of time spent across executors working for a particular application	counter
counter.spark.executor.total_input_bytes	Number of input bytes across executors working for a particular application	counter
counter.spark.executor.total_shuffle_read	Size read during a shuffle in a particular application’s executors	counter
counter.spark.executor.total_shuffle_write	Size written to during a shuffle in a particular application’s executors	counter
counter.spark.executor.total_tasks	Total tasks across executors working for a particular application	counter
counter.spark.streaming.num_processed_records	Number of processed records in a streaming application	counter
counter.spark.streaming.num_received_records	Number of received records in a streaming application	counter
counter.spark.streaming.num_total_completed_batches	Number of batches completed in a streaming application	counter
gauge.jvm.MarkSweepCompact.count	Garbage collection count	gauge
gauge.jvm.MarkSweepCompact.time	Garbage collection time	gauge
gauge.jvm.heap.committed	Amount of committed heap memory (in MB)	gauge
gauge.jvm.heap.used	Amount of used heap memory (in MB)	gauge
gauge.jvm.non-heap.committed	Amount of committed non-heap memory (in MB)	gauge
gauge.jvm.non-heap.used	Amount of used non-heap memory (in MB)	gauge
gauge.jvm.pools.Code-Cache.committed	Amount of memory committed for compilation and storage of native code	gauge
gauge.jvm.pools.Code-Cache.used	Amount of memory used to compile and store native code	gauge
gauge.jvm.pools.Compressed-Class-Space.committed	Amount of memory committed for compressing a class object	gauge
gauge.jvm.pools.Compressed-Class-Space.used	Amount of memory used to compress a class object	gauge
gauge.jvm.pools.Eden-Space.committed	Amount of memory committed for the initial allocation of objects	gauge
gauge.jvm.pools.Eden-Space.used	Amount of memory used for the initial allocation of objects	gauge
gauge.jvm.pools.Metaspace.committed	Amount of memory committed for storing classes and classloaders	gauge
gauge.jvm.pools.Metaspace.used	Amount of memory used to store classes and classloaders	gauge
gauge.jvm.pools.Survivor-Space.committed	Amount of memory committed specifically for objects that have survived GC of the Eden Space	gauge
gauge.jvm.pools.Survivor-Space.used	Amount of memory used for objects that have survived GC of the Eden Space	gauge
gauge.jvm.pools.Tenured-Gen.committed	Amount of memory committed to store objects that have lived in the survivor space for a given period of time	gauge
gauge.jvm.pools.Tenured-Gen.used	Amount of memory used for objects that have lived in the survivor space for a given period of time	gauge
gauge.jvm.total.committed	Amount of committed JVM memory (in MB)	gauge
gauge.jvm.total.used	Amount of used JVM memory (in MB)	gauge
gauge.master.aliveWorkers	Total functioning workers	gauge
gauge.master.apps	Total number of active applications in the spark cluster	gauge
gauge.master.waitingApps	Total number of waiting applications in the spark cluster	gauge
gauge.master.workers	Total number of workers in spark cluster	gauge
gauge.spark.driver.active_tasks	Total number of active tasks in driver mapped to a particular application	gauge
gauge.spark.driver.max_memory	Maximum memory used by driver mapped to a particular application	gauge
gauge.spark.driver.rdd_blocks	Number of RDD blocks in the driver mapped to a particular application	gauge
gauge.spark.executor.active_tasks	Total number of active tasks across all executors working for a particular application	gauge
gauge.spark.executor.count	Total number of executors performing for an active application in the spark cluster	gauge
gauge.spark.executor.max_memory	Max memory across all executors working for a particular application	gauge
gauge.spark.executor.rdd_blocks	Number of RDD blocks across all executors working for a particular application	gauge
gauge.spark.job.num_active_stages	Total number of active stages for an active application in the spark cluster	gauge
gauge.spark.job.num_active_tasks	Total number of active tasks for an active application in the spark cluster	gauge
gauge.spark.job.num_completed_stages	Total number of completed stages for an active application in the spark cluster	gauge
gauge.spark.job.num_completed_tasks	Total number of completed tasks for an active application in the spark cluster	gauge
gauge.spark.job.num_failed_stages	Total number of failed stages for an active application in the spark cluster	gauge
gauge.spark.job.num_failed_tasks	Total number of failed tasks for an active application in the spark cluster	gauge
gauge.spark.job.num_skipped_stages	Total number of skipped stages for an active application in the spark cluster	gauge
gauge.spark.job.num_skipped_tasks	Total number of skipped tasks for an active application in the spark cluster	gauge
gauge.spark.job.num_tasks	Total number of tasks for an active application in the spark cluster	gauge
gauge.spark.num_active_stages	Total number of active stages for an active application in the spark cluster	gauge
gauge.spark.num_running_jobs	Total number of running jobs for an active application in the spark cluster	gauge
gauge.spark.stage.disk_bytes_spilled	Actual size written to disk for an active application in the spark cluster	gauge
gauge.spark.stage.executor_run_time	Fraction of time spent by (and averaged across) executors for a particular application	gauge
gauge.spark.stage.input_bytes	Input size for a particular application	gauge
gauge.spark.stage.input_records	Input records received for a particular application	gauge
gauge.spark.stage.memory_bytes_spilled	Size spilled to disk from memory for an active application in the spark cluster	gauge
gauge.spark.stage.output_bytes	Output size for a particular application	gauge
gauge.spark.stage.output_records	Output records written to for a particular application	gauge
gauge.spark.stage.shuffle_read_bytes	Read size during shuffle phase for a particular application	gauge
gauge.spark.stage.shuffle_read_records	Number of records read during shuffle phase for a particular application	gauge
gauge.spark.stage.shuffle_write_bytes	Size written during shuffle phase for a particular application	gauge
gauge.spark.stage.shuffle_write_records	Number of records written to during shuffle phase for a particular application	gauge
gauge.spark.streaming.avg_input_rate	Average input rate of records across retained batches in a streaming application	gauge
gauge.spark.streaming.avg_processing_time	Average processing time in a streaming application	gauge
gauge.spark.streaming.avg_scheduling_delay	Average scheduling delay in a streaming application	gauge
gauge.spark.streaming.avg_total_delay	Average total delay in a streaming application	gauge
gauge.spark.streaming.num_active_batches	Number of active batches in a streaming application	gauge
gauge.spark.streaming.num_inactive_receivers	Number of inactive receivers in a streaming application	gauge
gauge.worker.coresFree	Total cores free for a particular worker process	gauge
gauge.worker.coresUsed	Total cores used by a particular worker process	gauge
gauge.worker.executors	Total number of executors for a particular worker process	gauge
gauge.worker.memFree_MB	Total memory free for a particular worker process	gauge
gauge.worker.memUsed_MB	Memory used by a particular worker process	gauge

counter.HiveExternalCatalog.counter.HiveClientCalls¶

counter

The total number of client calls sent to Hive for query processing. This metric is reported with the dimension spark_process to indicate whether it corresponds to a master or worker process.

counter.HiveExternalCatalog.fileCacheHits¶

counter

The total number of file level cache hits occurred. This metric is reported with the dimension spark_process to indicate whether it corresponds to a master or worker process.

counter.HiveExternalCatalog.filesDiscovered¶

counter

The total number of files discovered. This metric is reported with the dimension spark_process to indicate whether it corresponds to a master or worker process.

counter.HiveExternalCatalog.parallelListingJobCount¶

counter

The total number of Hive-specific jobs running in parallel. This metric is reported with the dimension spark_process to indicate whether it corresponds to a master or worker process.

counter.HiveExternalCatalog.partitionsFetched¶

counter

The total number of partitions fetched in Hive. This metric is reported with the dimension spark_process to indicate whether it corresponds to a master or worker process.

counter.spark.driver.completed_tasks¶

counter

Total number of completed tasks in driver mapped to a particular application. This metric is reported with the dimension cluster to specify the cluster Spark is running on (e.g. Mesos)

counter.spark.driver.disk_used¶

counter

Amount of disk used by driver mapped to a particular application (expressed in bytes). This metric is reported with the dimension cluster to specify the cluster Spark is running on (e.g. Mesos)

counter.spark.driver.failed_tasks¶

counter

Total number of failed tasks in driver mapped to a particular application. This metric is reported with the dimension cluster to specify the cluster Spark is running on (e.g. Mesos)

counter.spark.driver.memory_used¶

counter

Amount of memory used by driver mapped to a particular application (expressed in bytes). This metric is reported with the dimension cluster to specify the cluster Spark is running on (e.g. Mesos)

counter.spark.driver.total_duration¶

counter

Fraction of time spent by driver mapped to a particular application (expressed in ms/s). This metric is reported with the dimension cluster to specify the cluster Spark is running on (e.g. Mesos)

counter.spark.driver.total_input_bytes¶

counter

The number of input bytes in driver mapped to a particular application (expressed in bytes). This metric is reported with the dimension cluster to specify the cluster Spark is running on (e.g. Mesos)

counter.spark.driver.total_shuffle_read¶

counter

The size read during a shuffle in driver mapped to a particular application(expressed in bytes). This metric is reported with the dimension cluster to specify the cluster Spark is running on (e.g. Mesos)

counter.spark.driver.total_shuffle_write¶

counter

The size written to during a shuffle in driver mapped to a particular application (expressed in bytes). This metric is reported with the dimension cluster to specify the cluster Spark is running on (e.g. Mesos)

counter.spark.driver.total_tasks¶

counter

Total number of tasks in driver mapped to a particular application. This metric is reported with the dimension cluster to specify the cluster Spark is running on (e.g. Mesos)

counter.spark.executor.completed_tasks¶

counter

The number of completed tasks across executors working for a particular application. This metric is reported with the dimension cluster to specify the cluster Spark is running on (e.g. Mesos)

counter.spark.executor.disk_used¶

counter

Amount of disk used across executors working for a particular application (expressed in bytes). This metric is reported with the dimension cluster to specify the cluster Spark is running on (e.g. Mesos)

counter.spark.executor.failed_tasks¶

counter

The number of failed tasks across executors working for a particular application. This metric is reported with the dimension cluster to specify the cluster Spark is running on (e.g. Mesos)

counter.spark.executor.memory_used¶

counter

Amount of memory used across executors working for a particular application (expressed in bytes). This metric is reported with the dimension cluster to specify the cluster Spark is running on (e.g. Mesos)

counter.spark.executor.total_duration¶

counter

Fraction of time spent across executors working for a particular application (expressed in ms/s). This metric is reported with the dimension cluster to specify the cluster Spark is running on (e.g. Mesos)

counter.spark.executor.total_input_bytes¶

counter

The number of input bytes across executors working for a particular application (expressed in bytes). This metric is reported with the dimension cluster to specify the cluster Spark is running on (e.g. Mesos)

counter.spark.executor.total_shuffle_read¶

counter

The size read during a shuffle in a particular application’s executors (expressed in bytes) - aggregated across executors. This metric is reported with the dimension cluster to specify the cluster Spark is running on (e.g. Mesos)

counter.spark.executor.total_shuffle_write¶

counter

The size written to during a shuffle in a particular application’s executors (expressed in bytes) - aggregated across executors. This metric is reported with the dimension cluster to specify the cluster Spark is running on (e.g. Mesos)

counter.spark.executor.total_tasks¶

counter

The total number of tasks across executors working for a particular application (expressed in bytes). This metric is reported with the dimension cluster to specify the cluster Spark is running on (e.g. Mesos)

counter.spark.streaming.num_processed_records¶

counter

The number of processed records in a streaming application. This metric is reported with the dimension cluster to specify the cluster Spark is running on (e.g. Mesos)

counter.spark.streaming.num_received_records¶

counter

The number of received records in a streaming application. This metric is reported with the dimension cluster to specify the cluster Spark is running on (e.g. Mesos)

counter.spark.streaming.num_total_completed_batches¶

counter

The number of batches completed in a streaming application. This metric is reported with the dimension cluster to specify the cluster Spark is running on (e.g. Mesos)

gauge.jvm.MarkSweepCompact.count¶

gauge

The number of times garbage collection have occurred in the Marksweep GC. This metric is reported with the dimension spark_process to indicate whether it corresponds to a master or worker process.

gauge.jvm.MarkSweepCompact.time¶

gauge

The time taken for garbage collection that have occurred in the Marksweep GC. This metric is reported with the dimension spark_process to indicate whether it corresponds to a master or worker process.

gauge.jvm.heap.committed¶

gauge

The total amount of heap memory (expressed in MB) committed. This metric is reported with the dimension spark_process to indicate whether it corresponds to a master or worker process.

gauge.jvm.heap.used¶

gauge

The total amount of heap memory (expressed in MB) used. This metric is reported with the dimension spark_process to indicate whether it corresponds to a master or worker process.

gauge.jvm.non-heap.committed¶

gauge

The total amount of non-heap memory (expressed in MB) committed. This metric is reported with the dimension spark_process to indicate whether it corresponds to a master or worker process.

gauge.jvm.non-heap.used¶

gauge

The total amount of non-heap memory (expressed in MB) used. This metric is reported with the dimension spark_process to indicate whether it corresponds to a master or worker process.

gauge.jvm.pools.Code-Cache.committed¶

gauge

The amount of memory (expressed in MB) committed for compilation and storage of native code. This metric is reported with the dimension spark_process to indicate whether it corresponds to a master or worker process.

gauge.jvm.pools.Code-Cache.used¶

gauge

The amount of memory (expressed in MB) used to compile and store native code. This metric is reported with the dimension spark_process to indicate whether it corresponds to a master or worker process.

gauge.jvm.pools.Compressed-Class-Space.committed¶

gauge

The amount of memory (expressed in MB) committed for compressing a class object. This metric is reported with the dimension spark_process to indicate whether it corresponds to a master or worker process.

gauge.jvm.pools.Compressed-Class-Space.used¶

gauge

The amount of memory (expressed in MB) used to compress a class object. This metric is reported with the dimension spark_process to indicate whether it corresponds to a master or worker process.

gauge.jvm.pools.Eden-Space.committed¶

gauge

The amount of memory (expressed in MB) committed for the initial allocation of objects. This metric is reported with the dimension spark_process to indicate whether it corresponds to a master or worker process.

gauge.jvm.pools.Eden-Space.used¶

gauge

The amount of memory (expressed in MB) used for the initial allocation of objects. This metric is reported with the dimension spark_process to indicate whether it corresponds to a master or worker process.

gauge.jvm.pools.Metaspace.committed¶

gauge

The amount of memory (expressed in MB) committed for storing classes and classloaders. This metric is reported with the dimension spark_process to indicate whether it corresponds to a master or worker process.

gauge.jvm.pools.Metaspace.used¶

gauge

The amount of memory (expressed in MB) used to store classes and classloaders. This metric is reported with the dimension spark_process to indicate whether it corresponds to a master or worker process.

gauge.jvm.pools.Survivor-Space.committed¶

gauge

Amount of memory (expressed in MB) committed specifically for objects that have survived garbace collection of the Eden Space. This metric is reported with the dimension spark_process to indicate whether it corresponds to a master or worker process.

gauge.jvm.pools.Survivor-Space.used¶

gauge

Amount of memory (expressed in MB) used for objects that have survived garbace collection of the Eden Space. This metric is reported with the dimension spark_process to indicate whether it corresponds to a master or worker process.

gauge.jvm.pools.Tenured-Gen.committed¶

gauge

Amount of memory (expressed in MB) committed to store objects that have lived in the survivor space for a given period of time. This metric is reported with the dimension spark_process to indicate whether it corresponds to a master or worker process.

gauge.jvm.pools.Tenured-Gen.used¶

gauge

Amount of memory (expressed in MB) used for objects that have lived in the survivor space for a given period of time. This metric is reported with the dimension spark_process to indicate whether it corresponds to a master or worker process.

gauge.jvm.total.committed¶

gauge

The total amount of JVM memory (expressed in MB) committed. This metric is reported with the dimension spark_process to indicate whether it corresponds to a master or worker process.

gauge.jvm.total.used¶

gauge

The total amount of JVM memory (expressed in MB) used. This metric is reported with the dimension spark_process to indicate whether it corresponds to a master or worker process.

gauge.master.aliveWorkers¶

gauge

The total number of functioning workers reporting to the master process. This metric is reported with the dimension spark_process to indicate whether it corresponds to a master or worker process.

gauge.master.apps¶

gauge

The total number of applications still active/running in the spark cluster. This metric is reported with the dimension spark_process to indicate whether it corresponds to a master or worker process.

gauge.master.waitingApps¶

gauge

The total number of applications in queue waiting to be executed in the spark cluster. This metric is reported with the dimension spark_process to indicate whether it corresponds to a master or worker process.

gauge.master.workers¶

gauge

The total number of workers in spark cluster registered to the master process. This metric is reported with the dimension spark_process to indicate whether it corresponds to a master or worker process.

gauge.spark.driver.active_tasks¶

gauge

Total number of active tasks in driver mapped to a particular application. This metric is reported with the dimension cluster to specify the cluster Spark is running on (e.g. Mesos)

gauge.spark.driver.max_memory¶

gauge

Maximum memory used by driver mapped to a particular application (express in bytes). This metric is reported with the dimension cluster to specify the cluster Spark is running on (e.g. Mesos)

gauge.spark.driver.rdd_blocks¶

gauge

Number of RDD blocks in the driver mapped to a particular application. This metric is reported with the dimension cluster to specify the cluster Spark is running on (e.g. Mesos)

gauge.spark.executor.active_tasks¶

gauge

Total number of active tasks across all executors working for a particular application. This metric is reported with the dimension cluster to specify the cluster Spark is running on (e.g. Mesos)

gauge.spark.executor.count¶

gauge

Total number of executors performing for an active application in the spark cluster. This metric is reported with the dimension cluster to specify the cluster Spark is running on (e.g. Mesos)

gauge.spark.executor.max_memory¶

gauge

Max memory across all executors working for a particular application (expressed in bytes). This metric is reported with the dimension cluster to specify the cluster Spark is running on (e.g. Mesos)

gauge.spark.executor.rdd_blocks¶

gauge

Number of RDD blocks across all executors working for a particular application. This metric is reported with the dimension cluster to specify the cluster Spark is running on (e.g. Mesos)

gauge.spark.job.num_active_stages¶

gauge

The total number of active stages for an active application in the spark cluster - aggregated by jobs. This metric is reported with the dimension cluster to specify the cluster Spark is running on (e.g. Mesos)

gauge.spark.job.num_active_tasks¶

gauge

The total number of active tasks for an active application in the spark cluster - aggregated by jobs. This metric is reported with the dimension cluster to specify the cluster Spark is running on (e.g. Mesos)

gauge.spark.job.num_completed_stages¶

gauge

The total number of completed stages for an active application in the spark cluster - aggregated by jobs. This metric is reported with the dimension cluster to specify the cluster Spark is running on (e.g. Mesos)

gauge.spark.job.num_completed_tasks¶

gauge

The total number of completed tasks for an active application in the spark cluster - aggregated by jobs. This metric is reported with the dimension cluster to specify the cluster Spark is running on (e.g. Mesos)

gauge.spark.job.num_failed_stages¶

gauge

The total number of failed stages for an active application in the spark cluster - aggregated by jobs. This metric is reported with the dimension cluster to specify the cluster Spark is running on (e.g. Mesos)

gauge.spark.job.num_failed_tasks¶

gauge

The total number of failed tasks for an active application in the spark cluster - aggregated by jobs. This metric is reported with the dimension cluster to specify the cluster Spark is running on (e.g. Mesos)

gauge.spark.job.num_skipped_stages¶

gauge

The total number of skipped stages for an active application in the spark cluster - aggregated by jobs. This metric is reported with the dimension cluster to specify the cluster Spark is running on (e.g. Mesos)

gauge.spark.job.num_skipped_tasks¶

gauge

The total number of skipped tasks for an active application in the spark cluster - aggregated by jobs. This metric is reported with the dimension cluster to specify the cluster Spark is running on (e.g. Mesos)

gauge.spark.job.num_tasks¶

gauge

The total number of tasks for an active application in the spark cluster - aggregated by jobs. This metric is reported with the dimension cluster to specify the cluster Spark is running on (e.g. Mesos)

gauge.spark.num_active_stages¶

gauge

The total number of active stages for an active application in the spark cluster. This metric is reported with the dimension cluster to specify the cluster Spark is running on (e.g. Mesos)

gauge.spark.num_running_jobs¶

gauge

The total number of running jobs for an active application in the spark cluster. This metric is reported with the dimension cluster to specify the cluster Spark is running on (e.g. Mesos)

gauge.spark.stage.disk_bytes_spilled¶

gauge

Actual size written to disk (expressed in bytes) - aggregated by active stages. This metric is reported with the dimension cluster to specify the cluster Spark is running on (e.g. Mesos)

gauge.spark.stage.executor_run_time¶

gauge

Fraction of time spent by (and averaged across) executors for a particular application (expressed in ms/s) - aggregated by active stages. This metric is reported with the dimension cluster to specify the cluster Spark is running on (e.g. Mesos)

gauge.spark.stage.input_bytes¶

gauge

Input size for a particular application (expressed as bytes) - aggregated by active stages. This metric is reported with the dimension cluster to specify the cluster Spark is running on (e.g. Mesos)

gauge.spark.stage.input_records¶

gauge

Input records received for a particular application - aggregated by active stages. This metric is reported with the dimension cluster to specify the cluster Spark is running on (e.g. Mesos)

gauge.spark.stage.memory_bytes_spilled¶

gauge

Size spilled to disk from memory (expressed in bytes) - aggregated by active stages. This metric is reported with the dimension cluster to specify the cluster Spark is running on (e.g. Mesos)

gauge.spark.stage.output_bytes¶

gauge

Output size for a particular application (expressed as bytes) - aggregated by active stages. This metric is reported with the dimension cluster to specify the cluster Spark is running on (e.g. Mesos)

gauge.spark.stage.output_records¶

gauge

Output records written to for a particular application - aggregated by active stages. This metric is reported with the dimension cluster to specify the cluster Spark is running on (e.g. Mesos)

gauge.spark.stage.shuffle_read_bytes¶

gauge

Read size during shuffle phase for a particular application (expressed as bytes) - aggregated by active stages. This metric is reported with the dimension cluster to specify the cluster Spark is running on (e.g. Mesos)

gauge.spark.stage.shuffle_read_records¶

gauge

Number of records read during shuffle phase for a particular application - aggregated by active stages. This metric is reported with the dimension cluster to specify the cluster Spark is running on (e.g. Mesos)

gauge.spark.stage.shuffle_write_bytes¶

gauge

Size written during shuffle phase for a particular application (expressed as bytes) - aggregated by active stages. This metric is reported with the dimension cluster to specify the cluster Spark is running on (e.g. Mesos)

gauge.spark.stage.shuffle_write_records¶

gauge

Number of records written to during shuffle phase for a particular application - aggregated by active stages. This metric is reported with the dimension cluster to specify the cluster Spark is running on (e.g. Mesos)

gauge.spark.streaming.avg_input_rate¶

gauge

The average input rate of records across retained batches in a streaming application (expressed in ms/s). This metric is reported with the dimension cluster to specify the cluster Spark is running on (e.g. Mesos)

gauge.spark.streaming.avg_processing_time¶

gauge

The average processing time in a streaming application (expressed in ms). This metric is reported with the dimension cluster to specify the cluster Spark is running on (e.g. Mesos)

gauge.spark.streaming.avg_scheduling_delay¶

gauge

The average scheduling delay in a streaming application (expressed in ms). This metric is reported with the dimension cluster to specify the cluster Spark is running on (e.g. Mesos)

gauge.spark.streaming.avg_total_delay¶

gauge

The average total delay in a streaming application (expressed in ms). This metric is reported with the dimension cluster to specify the cluster Spark is running on (e.g. Mesos)

gauge.spark.streaming.num_active_batches¶

gauge

The number of active batches in a streaming application. This metric is reported with the dimension cluster to specify the cluster Spark is running on (e.g. Mesos)

gauge.spark.streaming.num_inactive_receivers¶

gauge

The number of receivers (e.g. Kafka, Flume, etc.) that have become inactive in a streaming application. This metric is reported with the dimension cluster to specify the cluster Spark is running on (e.g. Mesos)

gauge.worker.coresFree¶

gauge

The total number of cores free for a particular worker process. This metric is reported with the dimension spark_process to indicate whether it corresponds to a master or worker process.

gauge.worker.coresUsed¶

gauge

The total number of cores used by a particular worker process. This metric is reported with the dimension spark_process to indicate whether it corresponds to a master or worker process.

gauge.worker.executors¶

gauge

The total number of executors running jobs for a particular worker process. This metric is reported with the dimension spark_process to indicate whether it corresponds to a master or worker process.

gauge.worker.memFree_MB¶

gauge

The total amount of memory (expressed in MB) available to a particular worker process. This metric is reported with the dimension spark_process to indicate whether it corresponds to a master or worker process.

gauge.worker.memUsed_MB¶

gauge

The amount of memory (expressed in MB) currently used by a particular worker process. This metric is reported with the dimension spark_process to indicate whether it corresponds to a master or worker process.