collectd-spark.md

collectd/spark

Monitor Type: collectd/spark (Source)

Accepts Endpoints: Yes

Multiple Instances Allowed: Yes

Overview

This integration collects metrics about a Spark cluster using the collectd Spark Python plugin. That plugin collects metrics from Spark cluster and instances by hitting endpoints specified in Spark's Monitoring and Instrumentation documentation under REST API and Metrics.

The following cluster modes are supported only through HTTP endpoints:

• Standalone
• Mesos
• Hadoop YARN

You must specify distinct monitor configurations and discovery rules for master and worker processes. For the master configuration, set isMaster to true.

When running Spark on Apache Hadoop / YARN, this integration is only capable of reporting application metrics from the master node. Use the collectd/hadoop monitor to report on the health of the cluster.

Example config:

An example configuration for monitoring applications on YARN

monitors:
  - type: collectd/spark
    host: 000.000.000.000
    port: 8088
    clusterType: Yarn
    isMaster: true
    collectApplicationMetrics: true

Configuration

To activate this monitor in the Smart Agent, add the following to your agent config:

monitors:  # All monitor config goes under this key
 - type: collectd/spark
   ...  # Additional config

For a list of monitor options that are common to all monitors, see Common Configuration.

Config option	Required	Type	Description
pythonBinary	no	string	Path to a python binary that should be used to execute the Python code. If not set, a built-in runtime will be used. Can include arguments to the binary as well.
host	yes	string
port	yes	integer
isMaster	no	bool	Set to true when monitoring a master Spark node (default: false)
clusterType	yes	string	Should be one of Standalone or Mesos or Yarn. Cluster metrics will not be collected on Yarn. Please use the collectd/hadoop monitor to gain insights to your cluster's health.
collectApplicationMetrics	no	bool	(default: false)
enhancedMetrics	no	bool	(default: false)

Metrics

These are the metrics available for this monitor. Metrics that are categorized as container/host (default) are in bold and italics in the list below.

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

Non-default metrics (version 4.7.0+)

To emit metrics that are not default, you can add those metrics in the generic monitor-level extraMetrics config option. Metrics that are derived from specific configuration options that do not appear in the above list of metrics do not need to be added to extraMetrics.

To see a list of metrics that will be emitted you can run agent-status monitors after configuring this monitor in a running agent instance.

Dimensions

The following dimensions may occur on metrics emitted by this monitor. Some dimensions may be specific to certain metrics.

Name	Description
cluster	set to value corresponding to key cluster in configuration file
spark_process	Either master or worker to differentiate master- and worker- specific metrics like master.apps and worker.coresFree