Skip to content


Executor is a process that is used for executing[tasks].

Executor typically runs for the entire lifetime of a Spark application which is called static allocation of executors (but you could also opt in for dynamic allocation).

Executors are managed by[executor backends].

Executors <> to the <> on the driver.

HeartbeatReceiver's Heartbeat Message Handler

Executors provide in-memory storage for RDDs that are cached in Spark applications (via[]).

When started, an executor first registers itself with the driver that establishes a communication channel directly to the driver to accept tasks for execution.

Launching tasks on executor using TaskRunners../images/executor/executor-taskrunner-executorbackend.png)

Executor offers are described by executor id and the host on which an executor runs (see <> in this document).

Executors can run multiple tasks over its lifetime, both in parallel and sequentially. They track[running tasks] (by their task ids in <> internal registry). Consult <> section.

Executors use a <> for <>.

Executors send <> (and heartbeats) using the <>.

It is recommended to have as many executors as data nodes and as many cores as you can get from the cluster.

Executors are described by their id, hostname, environment (as SparkEnv), and classpath (and, less importantly, and more for internal optimization, whether they run in[local] or[cluster] mode).

Creating Instance

Executor takes the following to be created:

  • Executor ID
  • Host name
  • SparkEnv
  • User-defined jars (default: empty)
  • isLocal flag (default: false)
  • Java's UncaughtExceptionHandler (default: SparkUncaughtExceptionHandler)
  • Resources (Map[String, ResourceInformation])

Executor is created when:

When Created

When created, Executor prints out the following INFO messages to the logs:

Starting executor ID [executorId] on host [executorHostname]

(only for non-local modes) Executor sets SparkUncaughtExceptionHandler as the default handler invoked when a thread abruptly terminates due to an uncaught exception.

(only for non-local modes) Executor requests the BlockManager to initialize (with the Spark application id of the SparkConf).

(only for non-local modes) Executor requests the MetricsSystem to register the ExecutorSource and shuffleMetricsSource of the BlockManager.

Executor uses SparkEnv to access the MetricsSystem and BlockManager.

Executor creates a task class loader (optionally with REPL support) and requests the system Serializer to use as the default classloader (for deserializing tasks).

Executor starts sending heartbeats with the metrics of active tasks.

Fetching File and Jar Dependencies

  newFiles: Map[String, Long],
  newJars: Map[String, Long]): Unit

updateDependencies fetches missing or outdated extra files (in the given newFiles). For every name-timestamp pair that...FIXME..., updateDependencies prints out the following INFO message to the logs:

Fetching [name] with timestamp [timestamp]

updateDependencies fetches missing or outdated extra jars (in the given newJars). For every name-timestamp pair that...FIXME..., updateDependencies prints out the following INFO message to the logs:

Fetching [name] with timestamp [timestamp]

updateDependencies fetches the file to the SparkFiles root directory.


updateDependencies is used when:

  • TaskRunner is requested to start (and run a task)


Executor uses the spark.driver.maxResultSize for TaskRunner when requested to run a task (and decide on a serialized task result).

Maximum Size of Direct Results

Executor uses the minimum of spark.task.maxDirectResultSize and spark.rpc.message.maxSize when TaskRunner is requested to run a task (and decide on the type of a serialized task result).


Enable ALL logging level for org.apache.spark.executor.Executor logger to see what happens inside.

Add the following line to conf/

Refer to Logging.

Review Me

== [[isLocal]] isLocal Flag

Executor is given a isLocal flag when created. This is how the executor knows whether it runs in local or cluster mode. It is disabled by default.

The flag is turned on for[Spark local] (via[LocalEndpoint]).

== [[userClassPath]] User-Defined Jars

Executor is given user-defined jars when created. There are no jars defined by default.

The jars are specified using[spark.executor.extraClassPath] configuration property (via[--user-class-path] command-line option of CoarseGrainedExecutorBackend).

Running Tasks Registry

runningTasks: Map[Long, TaskRunner]

Executor tracks TaskRunners by task IDs.

HeartbeatReceiver RPC Endpoint Reference

RPC endpoint reference to HeartbeatReceiver on the driver.

Set when Executor <>.

Used exclusively when Executor <> (that happens every <> interval).

== [[launchTask]] Launching Task

[source, scala]

launchTask( context: ExecutorBackend, taskDescription: TaskDescription): Unit

launchTask simply creates a[] (with the given[] and the TaskDescription) and adds it to the <> internal registry.

In the end, launchTask requests the <> to execute the TaskRunner (sometime in the future).

.Launching tasks on executor using TaskRunners image::executor-taskrunner-executorbackend.png[align="center"]

launchTask is used when:

  • CoarseGrainedExecutorBackend is requested to[handle a LaunchTask message]

  • LocalEndpoint RPC endpoint (of[LocalSchedulerBackend]) is requested to[reviveOffers]

  • MesosExecutorBackend is requested to[launchTask]

== [[heartbeater]] Heartbeat Sender Thread

heartbeater is a daemon {java-javadoc-url}/java/util/concurrent/ScheduledThreadPoolExecutor.html[ScheduledThreadPoolExecutor] with a single thread.

The name of the thread pool is driver-heartbeater.

== [[coarse-grained-executor]] Coarse-Grained Executors

Coarse-grained executors are executors that use[] for task scheduling.

== [[resource-offers]] Resource Offers

Read[resourceOffers] in TaskSchedulerImpl and[resourceOffer] in TaskSetManager.

== [[threadPool]] Executor task launch worker Thread Pool

Executor uses threadPool daemon cached thread pool with the name Executor task launch worker-[ID] (with ID being the task id) for <>.

threadPool is created when <> and shut down when <>.

== [[memory]] Executor Memory

You can control the amount of memory per executor using[spark.executor.memory] configuration property. It sets the available memory equally for all executors per application.

The amount of memory per executor is looked up when[SparkContext is created].

You can change the assigned memory per executor per node in[standalone cluster] using[SPARK_EXECUTOR_MEMORY] environment variable.

You can find the value displayed as Memory per Node in[web UI for standalone Master] (as depicted in the figure below).

.Memory per Node in Spark Standalone's web UI image::spark-standalone-webui-memory-per-node.png[align="center"]

The above figure shows the result of running[Spark shell] with the amount of memory per executor defined explicitly (on command line), i.e.

./bin/spark-shell --master spark://localhost:7077 -c spark.executor.memory=2g


Every executor registers its own[] to report metrics.

== [[stop]] Stopping Executor

[source, scala]

stop(): Unit

stop requests[MetricsSystem] for a report.

stop shuts <> down (and waits at most 10 seconds).

stop shuts <> down.

(only when <>) stop[requests SparkEnv to stop].

stop is used when[CoarseGrainedExecutorBackend] and[LocalEndpoint] are requested to stop their managed executors.

== [[computeTotalGcTime]] computeTotalGcTime Method

[source, scala]

computeTotalGcTime(): Long


computeTotalGcTime is used when:

  • TaskRunner is requested to[collectAccumulatorsAndResetStatusOnFailure] and[run]

  • Executor is requested to <>

== [[createClassLoader]] createClassLoader Method

[source, scala]

createClassLoader(): MutableURLClassLoader


createClassLoader is used when...FIXME

== [[addReplClassLoaderIfNeeded]] addReplClassLoaderIfNeeded Method

[source, scala]

addReplClassLoaderIfNeeded( parent: ClassLoader): ClassLoader


addReplClassLoaderIfNeeded is used when...FIXME

== [[reportHeartBeat]] Heartbeating With Partial Metrics For Active Tasks To Driver

[source, scala]

reportHeartBeat(): Unit

reportHeartBeat collects[TaskRunners] for <> (aka active tasks) with their[tasks] deserialized (i.e. either ready for execution or already started).[] has[task] deserialized when it[runs the task].

For every running task, reportHeartBeat takes its[TaskMetrics] and:

  • Requests[ShuffleRead metrics to be merged]
  •[Sets jvmGCTime metrics]

reportHeartBeat then records the latest values of[internal and external accumulators] for every task.

NOTE: Internal accumulators are a task's metrics while external accumulators are a Spark application's accumulators that a user has created.

reportHeartBeat sends a blocking Heartbeat message to <HeartbeatReceiver endpoint>> (running on the driver). reportHeartBeat uses the value of[spark.executor.heartbeatInterval] configuration property for the RPC timeout.

NOTE: A Heartbeat message contains the executor identifier, the accumulator updates, and the identifier of the[].

If the response (from <HeartbeatReceiver endpoint>>) is to re-register the BlockManager, you should see the following INFO message in the logs and reportHeartBeat requests the BlockManager to[re-register] (which will register the blocks the BlockManager manages with the driver).


Told to re-register on heartbeat

HeartbeatResponse requests the BlockManager to re-register when either[TaskScheduler] or HeartbeatReceiver know nothing about the executor.

When posting the Heartbeat was successful, reportHeartBeat resets <> internal counter.

In case of a non-fatal exception, you should see the following WARN message in the logs (followed by the stack trace).

Issue communicating with driver in heartbeater

Every failure reportHeartBeat increments <> up to[spark.executor.heartbeat.maxFailures] configuration property. When the heartbeat failures reaches the maximum, you should see the following ERROR message in the logs and the executor terminates with the error code: 56.

Exit as unable to send heartbeats to driver more than [HEARTBEAT_MAX_FAILURES] times

reportHeartBeat is used when Executor is requested to <> (that happens every[spark.executor.heartbeatInterval]).

== [[startDriverHeartbeater]][[heartbeats-and-active-task-metrics]] Sending Heartbeats and Active Tasks Metrics

Executors keep sending <> to the driver every <> (defaults to 10s with some random initial delay so the heartbeats from different executors do not pile up on the driver).

.Executors use HeartbeatReceiver endpoint to report task metrics image::executor-heartbeatReceiver-endpoint.png[align="center"]

An executor sends heartbeats using the <>.

.HeartbeatReceiver's Heartbeat Message Handler image::HeartbeatReceiver-Heartbeat.png[align="center"]

For each[task] in[] (in <> internal registry), the task's metrics are computed (i.e. mergeShuffleReadMetrics and setJvmGCTime) that become part of the heartbeat (with accumulators).

NOTE: Executors track the[] that run[tasks]. A[task might not be assigned to a TaskRunner yet] when the executor sends a heartbeat.

A blocking Heartbeat message that holds the executor id, all accumulator updates (per task id), and[] is sent to HeartbeatReceiver RPC endpoint (with <> timeout).

If the response requests to reregister BlockManager, you should see the following INFO message in the logs:

Told to re-register on heartbeat

BlockManager is requested to[reregister].

The internal <> counter is reset (i.e. becomes 0).

If there are any issues with communicating with the driver, you should see the following WARN message in the logs:


Issue communicating with driver in heartbeater

The internal <> is incremented and checked to be less than the <> (i.e. spark.executor.heartbeat.maxFailures Spark property). If the number is greater, the following ERROR is printed out to the logs:

Exit as unable to send heartbeats to driver more than [HEARTBEAT_MAX_FAILURES] times

The executor exits (using System.exit and exit code 56).

== [[internal-properties]] Internal Properties

=== [[executorSource]] ExecutorSource[]

=== [[heartbeatFailures]] heartbeatFailures

Back to top