Skip to content

Latest commit

 

History

History
1534 lines (1420 loc) · 63.9 KB

monitoring.md

File metadata and controls

1534 lines (1420 loc) · 63.9 KB
Raw
layout title description license
global
Monitoring and Instrumentation
Monitoring, metrics, and instrumentation guide for Spark SPARK_VERSION_SHORT
Licensed to the Apache Software Foundation (ASF) under one or more contributor license agreements. See the NOTICE file distributed with this work for additional information regarding copyright ownership. The ASF licenses this file to You under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with the License. You may obtain a copy of the License at http://www.apache.org/licenses/LICENSE-2.0 Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License.
  • This will become a table of contents (this text will be scraped). {:toc}

There are several ways to monitor Spark applications: web UIs, metrics, and external instrumentation.

Web Interfaces

Every SparkContext launches a Web UI, by default on port 4040, that displays useful information about the application. This includes:

  • A list of scheduler stages and tasks
  • A summary of RDD sizes and memory usage
  • Environmental information.
  • Information about the running executors

You can access this interface by simply opening http://<driver-node>:4040 in a web browser. If multiple SparkContexts are running on the same host, they will bind to successive ports beginning with 4040 (4041, 4042, etc).

Note that this information is only available for the duration of the application by default. To view the web UI after the fact, set spark.eventLog.enabled to true before starting the application. This configures Spark to log Spark events that encode the information displayed in the UI to persisted storage.

Viewing After the Fact

It is still possible to construct the UI of an application through Spark's history server, provided that the application's event logs exist. You can start the history server by executing:

./sbin/start-history-server.sh

This creates a web interface at http://<server-url>:18080 by default, listing incomplete and completed applications and attempts.

When using the file-system provider class (see spark.history.provider below), the base logging directory must be supplied in the spark.history.fs.logDirectory configuration option, and should contain sub-directories that each represents an application's event logs.

The spark jobs themselves must be configured to log events, and to log them to the same shared, writable directory. For example, if the server was configured with a log directory of hdfs://namenode/shared/spark-logs, then the client-side options would be:

spark.eventLog.enabled true
spark.eventLog.dir hdfs://namenode/shared/spark-logs

The history server can be configured as follows:

Environment Variables

Environment VariableMeaning
SPARK_DAEMON_MEMORY Memory to allocate to the history server (default: 1g).
SPARK_DAEMON_JAVA_OPTS JVM options for the history server (default: none).
SPARK_DAEMON_CLASSPATH Classpath for the history server (default: none).
SPARK_PUBLIC_DNS The public address for the history server. If this is not set, links to application history may use the internal address of the server, resulting in broken links (default: none).
SPARK_HISTORY_OPTS spark.history.* configuration options for the history server (default: none).

Applying compaction on rolling event log files

A long-running application (e.g. streaming) can bring a huge single event log file which may cost a lot to maintain and also requires a bunch of resource to replay per each update in Spark History Server.

Enabling spark.eventLog.rolling.enabled and spark.eventLog.rolling.maxFileSize would let you have rolling event log files instead of single huge event log file which may help some scenarios on its own, but it still doesn't help you reducing the overall size of logs.

Spark History Server can apply compaction on the rolling event log files to reduce the overall size of logs, via setting the configuration spark.history.fs.eventLog.rolling.maxFilesToRetain on the Spark History Server.

Details will be described below, but please note in prior that compaction is LOSSY operation. Compaction will discard some events which will be no longer seen on UI - you may want to check which events will be discarded before enabling the option.

When the compaction happens, the History Server lists all the available event log files for the application, and considers the event log files having less index than the file with smallest index which will be retained as target of compaction. For example, if the application A has 5 event log files and spark.history.fs.eventLog.rolling.maxFilesToRetain is set to 2, then first 3 log files will be selected to be compacted.

Once it selects the target, it analyzes them to figure out which events can be excluded, and rewrites them into one compact file with discarding events which are decided to exclude.

The compaction tries to exclude the events which point to the outdated data. As of now, below describes the candidates of events to be excluded:

  • Events for the job which is finished, and related stage/tasks events
  • Events for the executor which is terminated
  • Events for the SQL execution which is finished, and related job/stage/tasks events

Once rewriting is done, original log files will be deleted, via best-effort manner. The History Server may not be able to delete the original log files, but it will not affect the operation of the History Server.

Please note that Spark History Server may not compact the old event log files if figures out not a lot of space would be reduced during compaction. For streaming query we normally expect compaction will run as each micro-batch will trigger one or more jobs which will be finished shortly, but compaction won't run in many cases for batch query.

Please also note that this is a new feature introduced in Spark 3.0, and may not be completely stable. Under some circumstances, the compaction may exclude more events than you expect, leading some UI issues on History Server for the application. Use it with caution.

Spark History Server Configuration Options

Security options for the Spark History Server are covered more detail in the Security page.

Property Name Default Meaning Since Version
spark.history.provider org.apache.spark.deploy.history.FsHistoryProvider Name of the class implementing the application history backend. Currently there is only one implementation, provided by Spark, which looks for application logs stored in the file system. 1.1.0
spark.history.fs.logDirectory file:/tmp/spark-events For the filesystem history provider, the URL to the directory containing application event logs to load. This can be a local file:// path, an HDFS path hdfs://namenode/shared/spark-logs or that of an alternative filesystem supported by the Hadoop APIs. 1.1.0
spark.history.fs.update.interval 10s The period at which the filesystem history provider checks for new or updated logs in the log directory. A shorter interval detects new applications faster, at the expense of more server load re-reading updated applications. As soon as an update has completed, listings of the completed and incomplete applications will reflect the changes. 1.4.0
spark.history.retainedApplications 50 The number of applications to retain UI data for in the cache. If this cap is exceeded, then the oldest applications will be removed from the cache. If an application is not in the cache, it will have to be loaded from disk if it is accessed from the UI. 1.0.0
spark.history.ui.maxApplications Int.MaxValue The number of applications to display on the history summary page. Application UIs are still available by accessing their URLs directly even if they are not displayed on the history summary page. 2.0.1
spark.history.ui.port 18080 The port to which the web interface of the history server binds. 1.0.0
spark.history.kerberos.enabled false Indicates whether the history server should use kerberos to login. This is required if the history server is accessing HDFS files on a secure Hadoop cluster. 1.0.1
spark.history.kerberos.principal (none) When spark.history.kerberos.enabled=true, specifies kerberos principal name for the History Server. 1.0.1
spark.history.kerberos.keytab (none) When spark.history.kerberos.enabled=true, specifies location of the kerberos keytab file for the History Server. 1.0.1
spark.history.fs.cleaner.enabled false Specifies whether the History Server should periodically clean up event logs from storage. 1.4.0
spark.history.fs.cleaner.interval 1d When spark.history.fs.cleaner.enabled=true, specifies how often the filesystem job history cleaner checks for files to delete. Files are deleted if at least one of two conditions holds. First, they're deleted if they're older than spark.history.fs.cleaner.maxAge. They are also deleted if the number of files is more than spark.history.fs.cleaner.maxNum, Spark tries to clean up the completed attempts from the applications based on the order of their oldest attempt time. 1.4.0
spark.history.fs.cleaner.maxAge 7d When spark.history.fs.cleaner.enabled=true, job history files older than this will be deleted when the filesystem history cleaner runs. 1.4.0
spark.history.fs.cleaner.maxNum Int.MaxValue When spark.history.fs.cleaner.enabled=true, specifies the maximum number of files in the event log directory. Spark tries to clean up the completed attempt logs to maintain the log directory under this limit. This should be smaller than the underlying file system limit like `dfs.namenode.fs-limits.max-directory-items` in HDFS. 3.0.0
spark.history.fs.endEventReparseChunkSize 1m How many bytes to parse at the end of log files looking for the end event. This is used to speed up generation of application listings by skipping unnecessary parts of event log files. It can be disabled by setting this config to 0. 2.4.0
spark.history.fs.inProgressOptimization.enabled true Enable optimized handling of in-progress logs. This option may leave finished applications that fail to rename their event logs listed as in-progress. 2.4.0
spark.history.fs.driverlog.cleaner.enabled spark.history.fs.cleaner.enabled Specifies whether the History Server should periodically clean up driver logs from storage. 3.0.0
spark.history.fs.driverlog.cleaner.interval spark.history.fs.cleaner.interval When spark.history.fs.driverlog.cleaner.enabled=true, specifies how often the filesystem driver log cleaner checks for files to delete. Files are only deleted if they are older than spark.history.fs.driverlog.cleaner.maxAge 3.0.0
spark.history.fs.driverlog.cleaner.maxAge spark.history.fs.cleaner.maxAge When spark.history.fs.driverlog.cleaner.enabled=true, driver log files older than this will be deleted when the driver log cleaner runs. 3.0.0
spark.history.fs.numReplayThreads 25% of available cores Number of threads that will be used by history server to process event logs. 2.0.0
spark.history.store.maxDiskUsage 10g Maximum disk usage for the local directory where the cache application history information are stored. 2.3.0
spark.history.store.path (none) Local directory where to cache application history data. If set, the history server will store application data on disk instead of keeping it in memory. The data written to disk will be re-used in the event of a history server restart. 2.3.0
spark.history.store.serializer JSON Serializer for writing/reading in-memory UI objects to/from disk-based KV Store; JSON or PROTOBUF. JSON serializer is the only choice before Spark 3.4.0, thus it is the default value. PROTOBUF serializer is fast and compact, compared to the JSON serializer. 3.4.0
spark.history.custom.executor.log.url (none) Specifies custom spark executor log URL for supporting external log service instead of using cluster managers' application log URLs in the history server. Spark will support some path variables via patterns which can vary on cluster manager. Please check the documentation for your cluster manager to see which patterns are supported, if any. This configuration has no effect on a live application, it only affects the history server.

For now, only YARN mode supports this configuration

3.0.0
spark.history.custom.executor.log.url.applyIncompleteApplication true Specifies whether to apply custom spark executor log URL to incomplete applications as well. If executor logs for running applications should be provided as origin log URLs, set this to `false`. Please note that incomplete applications may include applications which didn't shutdown gracefully. Even this is set to `true`, this configuration has no effect on a live application, it only affects the history server. 3.0.0
spark.history.fs.eventLog.rolling.maxFilesToRetain Int.MaxValue The maximum number of event log files which will be retained as non-compacted. By default, all event log files will be retained. The lowest value is 1 for technical reason.
Please read the section of "Applying compaction of old event log files" for more details. 		3.0.0
spark.history.store.hybridStore.enabled false Whether to use HybridStore as the store when parsing event logs. HybridStore will first write data to an in-memory store and having a background thread that dumps data to a disk store after the writing to in-memory store is completed. 3.1.0
spark.history.store.hybridStore.maxMemoryUsage 2g Maximum memory space that can be used to create HybridStore. The HybridStore co-uses the heap memory, so the heap memory should be increased through the memory option for SHS if the HybridStore is enabled. 3.1.0
spark.history.store.hybridStore.diskBackend ROCKSDB Specifies a disk-based store used in hybrid store; ROCKSDB or LEVELDB (deprecated). 3.3.0
spark.history.fs.update.batchSize Int.MaxValue Specifies the batch size for updating new eventlog files. This controls each scan process to be completed within a reasonable time, and such prevent the initial scan from running too long and blocking new eventlog files to be scanned in time in large environments. 3.4.0

Note that in all of these UIs, the tables are sortable by clicking their headers, making it easy to identify slow tasks, data skew, etc.

Note

  1. The history server displays both completed and incomplete Spark jobs. If an application makes multiple attempts after failures, the failed attempts will be displayed, as well as any ongoing incomplete attempt or the final successful attempt.

  2. Incomplete applications are only updated intermittently. The time between updates is defined by the interval between checks for changed files (spark.history.fs.update.interval). On larger clusters, the update interval may be set to large values. The way to view a running application is actually to view its own web UI.

  3. Applications which exited without registering themselves as completed will be listed as incomplete —even though they are no longer running. This can happen if an application crashes.

  4. One way to signal the completion of a Spark job is to stop the Spark Context explicitly (sc.stop()), or in Python using the with SparkContext() as sc: construct to handle the Spark Context setup and tear down.

REST API

In addition to viewing the metrics in the UI, they are also available as JSON. This gives developers an easy way to create new visualizations and monitoring tools for Spark. The JSON is available for both running applications, and in the history server. The endpoints are mounted at /api/v1. For example, for the history server, they would typically be accessible at http://<server-url>:18080/api/v1, and for a running application, at http://localhost:4040/api/v1.

In the API, an application is referenced by its application ID, [app-id]. When running on YARN, each application may have multiple attempts, but there are attempt IDs only for applications in cluster mode, not applications in client mode. Applications in YARN cluster mode can be identified by their [attempt-id]. In the API listed below, when running in YARN cluster mode, [app-id] will actually be [base-app-id]/[attempt-id], where [base-app-id] is the YARN application ID.

EndpointMeaning
/applications A list of all applications.
?status=[completed|running] list only applications in the chosen state.
?minDate=[date] earliest start date/time to list.
?maxDate=[date] latest start date/time to list.
?minEndDate=[date] earliest end date/time to list.
?maxEndDate=[date] latest end date/time to list.
?limit=[limit] limits the number of applications listed.
Examples:
?minDate=2015-02-10
?minDate=2015-02-03T16:42:40.000GMT
?maxDate=2015-02-11T20:41:30.000GMT
?minEndDate=2015-02-12
?minEndDate=2015-02-12T09:15:10.000GMT
?maxEndDate=2015-02-14T16:30:45.000GMT
?limit=10
/applications/[app-id]/jobs A list of all jobs for a given application.
?status=[running|succeeded|failed|unknown] list only jobs in the specific state.
/applications/[app-id]/jobs/[job-id] Details for the given job.
/applications/[app-id]/stages A list of all stages for a given application.
?status=[active|complete|pending|failed] list only stages in the given state.
?details=true lists all stages with the task data.
?taskStatus=[RUNNING|SUCCESS|FAILED|KILLED|PENDING] lists only those tasks with the specified task status. Query parameter taskStatus takes effect only when details=true. This also supports multiple taskStatus such as ?details=true&taskStatus=SUCCESS&taskStatus=FAILED which will return all tasks matching any of specified task status.
?withSummaries=true lists stages with task metrics distribution and executor metrics distribution.
?quantiles=0.0,0.25,0.5,0.75,1.0 summarize the metrics with the given quantiles. Query parameter quantiles takes effect only when withSummaries=true. Default value is 0.0,0.25,0.5,0.75,1.0.
/applications/[app-id]/stages/[stage-id] A list of all attempts for the given stage.
?details=true lists all attempts with the task data for the given stage.
?taskStatus=[RUNNING|SUCCESS|FAILED|KILLED|PENDING] lists only those tasks with the specified task status. Query parameter taskStatus takes effect only when details=true. This also supports multiple taskStatus such as ?details=true&taskStatus=SUCCESS&taskStatus=FAILED which will return all tasks matching any of specified task status.
?withSummaries=true lists task metrics distribution and executor metrics distribution of each attempt.
?quantiles=0.0,0.25,0.5,0.75,1.0 summarize the metrics with the given quantiles. Query parameter quantiles takes effect only when withSummaries=true. Default value is 0.0,0.25,0.5,0.75,1.0.
Example:
?details=true
?details=true&taskStatus=RUNNING
?withSummaries=true
?details=true&withSummaries=true&quantiles=0.01,0.5,0.99
/applications/[app-id]/stages/[stage-id]/[stage-attempt-id] Details for the given stage attempt.
?details=true lists all task data for the given stage attempt.
?taskStatus=[RUNNING|SUCCESS|FAILED|KILLED|PENDING] lists only those tasks with the specified task status. Query parameter taskStatus takes effect only when details=true. This also supports multiple taskStatus such as ?details=true&taskStatus=SUCCESS&taskStatus=FAILED which will return all tasks matching any of specified task status.
?withSummaries=true lists task metrics distribution and executor metrics distribution for the given stage attempt.
?quantiles=0.0,0.25,0.5,0.75,1.0 summarize the metrics with the given quantiles. Query parameter quantiles takes effect only when withSummaries=true. Default value is 0.0,0.25,0.5,0.75,1.0.
Example:
?details=true
?details=true&taskStatus=RUNNING
?withSummaries=true
?details=true&withSummaries=true&quantiles=0.01,0.5,0.99
/applications/[app-id]/stages/[stage-id]/[stage-attempt-id]/taskSummary Summary metrics of all tasks in the given stage attempt.
?quantiles summarize the metrics with the given quantiles.
Example: ?quantiles=0.01,0.5,0.99
/applications/[app-id]/stages/[stage-id]/[stage-attempt-id]/taskList A list of all tasks for the given stage attempt.
?offset=[offset]&length=[len] list tasks in the given range.
?sortBy=[runtime|-runtime] sort the tasks.
?status=[running|success|killed|failed|unknown] list only tasks in the state.
Example: ?offset=10&length=50&sortBy=runtime&status=running
/applications/[app-id]/executors A list of all active executors for the given application.
/applications/[app-id]/executors/[executor-id]/threads Stack traces of all the threads running within the given active executor. Not available via the history server.
/applications/[app-id]/allexecutors A list of all(active and dead) executors for the given application.
/applications/[app-id]/storage/rdd A list of stored RDDs for the given application.
/applications/[app-id]/storage/rdd/[rdd-id] Details for the storage status of a given RDD.
/applications/[base-app-id]/logs Download the event logs for all attempts of the given application as files within a zip file.
/applications/[base-app-id]/[attempt-id]/logs Download the event logs for a specific application attempt as a zip file.
/applications/[app-id]/streaming/statistics Statistics for the streaming context.
/applications/[app-id]/streaming/receivers A list of all streaming receivers.
/applications/[app-id]/streaming/receivers/[stream-id] Details of the given receiver.
/applications/[app-id]/streaming/batches A list of all retained batches.
/applications/[app-id]/streaming/batches/[batch-id] Details of the given batch.
/applications/[app-id]/streaming/batches/[batch-id]/operations A list of all output operations of the given batch.
/applications/[app-id]/streaming/batches/[batch-id]/operations/[outputOp-id] Details of the given operation and given batch.
/applications/[app-id]/sql A list of all queries for a given application.
?details=[true (default) | false] lists/hides details of Spark plan nodes.
?planDescription=[true (default) | false] enables/disables Physical planDescription on demand when Physical Plan size is high.
?offset=[offset]&length=[len] lists queries in the given range.
/applications/[app-id]/sql/[execution-id] Details for the given query.
?details=[true (default) | false] lists/hides metric details in addition to given query details.
?planDescription=[true (default) | false] enables/disables Physical planDescription on demand for the given query when Physical Plan size is high.
/applications/[app-id]/environment Environment details of the given application.
/version Get the current spark version.

The number of jobs and stages which can be retrieved is constrained by the same retention mechanism of the standalone Spark UI; "spark.ui.retainedJobs" defines the threshold value triggering garbage collection on jobs, and spark.ui.retainedStages that for stages. Note that the garbage collection takes place on playback: it is possible to retrieve more entries by increasing these values and restarting the history server.

Executor Task Metrics

The REST API exposes the values of the Task Metrics collected by Spark executors with the granularity of task execution. The metrics can be used for performance troubleshooting and workload characterization. A list of the available metrics, with a short description:

Spark Executor Task Metric name Short description
executorRunTime Elapsed time the executor spent running this task. This includes time fetching shuffle data. The value is expressed in milliseconds.
executorCpuTime CPU time the executor spent running this task. This includes time fetching shuffle data. The value is expressed in nanoseconds.
executorDeserializeTime Elapsed time spent to deserialize this task. The value is expressed in milliseconds.
executorDeserializeCpuTime CPU time taken on the executor to deserialize this task. The value is expressed in nanoseconds.
resultSize The number of bytes this task transmitted back to the driver as the TaskResult.
jvmGCTime Elapsed time the JVM spent in garbage collection while executing this task. The value is expressed in milliseconds.
ConcurrentGCCount This metric returns the total number of collections that have occurred. It only applies when the Java Garbage collector is G1 Concurrent GC.
ConcurrentGCTime This metric returns the approximate accumulated collection elapsed time in milliseconds. It only applies when the Java Garbage collector is G1 Concurrent GC.
resultSerializationTime Elapsed time spent serializing the task result. The value is expressed in milliseconds.
memoryBytesSpilled The number of in-memory bytes spilled by this task.
diskBytesSpilled The number of on-disk bytes spilled by this task.
peakExecutionMemory Peak memory used by internal data structures created during shuffles, aggregations and joins. The value of this accumulator should be approximately the sum of the peak sizes across all such data structures created in this task. For SQL jobs, this only tracks all unsafe operators and ExternalSort.
inputMetrics.* Metrics related to reading data from org.apache.spark.rdd.HadoopRDD or from persisted data.
    .bytesRead Total number of bytes read.
    .recordsRead Total number of records read.
outputMetrics.* Metrics related to writing data externally (e.g. to a distributed filesystem), defined only in tasks with output.
    .bytesWritten Total number of bytes written
    .recordsWritten Total number of records written
shuffleReadMetrics.* Metrics related to shuffle read operations.
    .recordsRead Number of records read in shuffle operations
    .remoteBlocksFetched Number of remote blocks fetched in shuffle operations
    .localBlocksFetched Number of local (as opposed to read from a remote executor) blocks fetched in shuffle operations
    .totalBlocksFetched Number of blocks fetched in shuffle operations (both local and remote)
    .remoteBytesRead Number of remote bytes read in shuffle operations
    .localBytesRead Number of bytes read in shuffle operations from local disk (as opposed to read from a remote executor)
    .totalBytesRead Number of bytes read in shuffle operations (both local and remote)
    .remoteBytesReadToDisk Number of remote bytes read to disk in shuffle operations. Large blocks are fetched to disk in shuffle read operations, as opposed to being read into memory, which is the default behavior.
    .fetchWaitTime Time the task spent waiting for remote shuffle blocks. This only includes the time blocking on shuffle input data. For instance if block B is being fetched while the task is still not finished processing block A, it is not considered to be blocking on block B. The value is expressed in milliseconds.
shuffleWriteMetrics.* Metrics related to operations writing shuffle data.
    .bytesWritten Number of bytes written in shuffle operations
    .recordsWritten Number of records written in shuffle operations
    .writeTime Time spent blocking on writes to disk or buffer cache. The value is expressed in nanoseconds.

Executor Metrics

Executor-level metrics are sent from each executor to the driver as part of the Heartbeat to describe the performance metrics of Executor itself like JVM heap memory, GC information. Executor metric values and their measured memory peak values per executor are exposed via the REST API in JSON format and in Prometheus format. The JSON end point is exposed at: /applications/[app-id]/executors, and the Prometheus endpoint at: /metrics/executors/prometheus. In addition, aggregated per-stage peak values of the executor memory metrics are written to the event log if spark.eventLog.logStageExecutorMetrics is true. Executor memory metrics are also exposed via the Spark metrics system based on the Dropwizard metrics library. A list of the available metrics, with a short description:

Executor Level Metric name Short description
rddBlocks RDD blocks in the block manager of this executor.
memoryUsed Storage memory used by this executor.
diskUsed Disk space used for RDD storage by this executor.
totalCores Number of cores available in this executor.
maxTasks Maximum number of tasks that can run concurrently in this executor.
activeTasks Number of tasks currently executing.
failedTasks Number of tasks that have failed in this executor.
completedTasks Number of tasks that have completed in this executor.
totalTasks Total number of tasks (running, failed and completed) in this executor.
totalDuration Elapsed time the JVM spent executing tasks in this executor. The value is expressed in milliseconds.
totalGCTime Elapsed time the JVM spent in garbage collection summed in this executor. The value is expressed in milliseconds.
totalInputBytes Total input bytes summed in this executor.
totalShuffleRead Total shuffle read bytes summed in this executor.
totalShuffleWrite Total shuffle write bytes summed in this executor.
maxMemory Total amount of memory available for storage, in bytes.
memoryMetrics.* Current value of memory metrics:
    .usedOnHeapStorageMemory Used on heap memory currently for storage, in bytes.
    .usedOffHeapStorageMemory Used off heap memory currently for storage, in bytes.
    .totalOnHeapStorageMemory Total available on heap memory for storage, in bytes. This amount can vary over time, on the MemoryManager implementation.
    .totalOffHeapStorageMemory Total available off heap memory for storage, in bytes. This amount can vary over time, depending on the MemoryManager implementation.
peakMemoryMetrics.* Peak value of memory (and GC) metrics:
    .JVMHeapMemory Peak memory usage of the heap that is used for object allocation. The heap consists of one or more memory pools. The used and committed size of the returned memory usage is the sum of those values of all heap memory pools whereas the init and max size of the returned memory usage represents the setting of the heap memory which may not be the sum of those of all heap memory pools. The amount of used memory in the returned memory usage is the amount of memory occupied by both live objects and garbage objects that have not been collected, if any.
    .JVMOffHeapMemory Peak memory usage of non-heap memory that is used by the Java virtual machine. The non-heap memory consists of one or more memory pools. The used and committed size of the returned memory usage is the sum of those values of all non-heap memory pools whereas the init and max size of the returned memory usage represents the setting of the non-heap memory which may not be the sum of those of all non-heap memory pools.
    .OnHeapExecutionMemory Peak on heap execution memory in use, in bytes.
    .OffHeapExecutionMemory Peak off heap execution memory in use, in bytes.
    .OnHeapStorageMemory Peak on heap storage memory in use, in bytes.
    .OffHeapStorageMemory Peak off heap storage memory in use, in bytes.
    .OnHeapUnifiedMemory Peak on heap memory (execution and storage).
    .OffHeapUnifiedMemory Peak off heap memory (execution and storage).
    .DirectPoolMemory Peak memory that the JVM is using for direct buffer pool (java.lang.management.BufferPoolMXBean)
    .MappedPoolMemory Peak memory that the JVM is using for mapped buffer pool (java.lang.management.BufferPoolMXBean)
    .ProcessTreeJVMVMemory Virtual memory size in bytes. Enabled if spark.executor.processTreeMetrics.enabled is true.
    .ProcessTreeJVMRSSMemory Resident Set Size: number of pages the process has in real memory. This is just the pages which count toward text, data, or stack space. This does not include pages which have not been demand-loaded in, or which are swapped out. Enabled if spark.executor.processTreeMetrics.enabled is true.
    .ProcessTreePythonVMemory Virtual memory size for Python in bytes. Enabled if spark.executor.processTreeMetrics.enabled is true.
    .ProcessTreePythonRSSMemory Resident Set Size for Python. Enabled if spark.executor.processTreeMetrics.enabled is true.
    .ProcessTreeOtherVMemory Virtual memory size for other kind of process in bytes. Enabled if spark.executor.processTreeMetrics.enabled is true.
    .ProcessTreeOtherRSSMemory Resident Set Size for other kind of process. Enabled if spark.executor.processTreeMetrics.enabled is true.
    .MinorGCCount Total minor GC count. For example, the garbage collector is one of Copy, PS Scavenge, ParNew, G1 Young Generation and so on.
    .MinorGCTime Elapsed total minor GC time. The value is expressed in milliseconds.
    .MajorGCCount Total major GC count. For example, the garbage collector is one of MarkSweepCompact, PS MarkSweep, ConcurrentMarkSweep, G1 Old Generation and so on.
    .MajorGCTime Elapsed total major GC time. The value is expressed in milliseconds.
The computation of RSS and Vmem are based on [proc(5)](http://man7.org/linux/man-pages/man5/proc.5.html)

API Versioning Policy

These endpoints have been strongly versioned to make it easier to develop applications on top. In particular, Spark guarantees:

  • Endpoints will never be removed from one version
  • Individual fields will never be removed for any given endpoint
  • New endpoints may be added
  • New fields may be added to existing endpoints
  • New versions of the api may be added in the future as a separate endpoint (e.g., api/v2). New versions are not required to be backwards compatible.
  • Api versions may be dropped, but only after at least one minor release of co-existing with a new api version.

Note that even when examining the UI of running applications, the applications/[app-id] portion is still required, though there is only one application available. E.g. to see the list of jobs for the running app, you would go to http://localhost:4040/api/v1/applications/[app-id]/jobs. This is to keep the paths consistent in both modes.

Metrics

Spark has a configurable metrics system based on the Dropwizard Metrics Library. This allows users to report Spark metrics to a variety of sinks including HTTP, JMX, and CSV files. The metrics are generated by sources embedded in the Spark code base. They provide instrumentation for specific activities and Spark components. The metrics system is configured via a configuration file that Spark expects to be present at $SPARK_HOME/conf/metrics.properties. A custom file location can be specified via the spark.metrics.conf configuration property. Instead of using the configuration file, a set of configuration parameters with prefix spark.metrics.conf. can be used. By default, the root namespace used for driver or executor metrics is the value of spark.app.id. However, often times, users want to be able to track the metrics across apps for driver and executors, which is hard to do with application ID (i.e. spark.app.id) since it changes with every invocation of the app. For such use cases, a custom namespace can be specified for metrics reporting using spark.metrics.namespace configuration property. If, say, users wanted to set the metrics namespace to the name of the application, they can set the spark.metrics.namespace property to a value like ${spark.app.name}. This value is then expanded appropriately by Spark and is used as the root namespace of the metrics system. Non-driver and executor metrics are never prefixed with spark.app.id, nor does the spark.metrics.namespace property have any such affect on such metrics.

Spark's metrics are decoupled into different instances corresponding to Spark components. Within each instance, you can configure a set of sinks to which metrics are reported. The following instances are currently supported:

  • master: The Spark standalone master process.
  • applications: A component within the master which reports on various applications.
  • worker: A Spark standalone worker process.
  • executor: A Spark executor.
  • driver: The Spark driver process (the process in which your SparkContext is created).
  • shuffleService: The Spark shuffle service.
  • applicationMaster: The Spark ApplicationMaster when running on YARN.

Each instance can report to zero or more sinks. Sinks are contained in the org.apache.spark.metrics.sink package:

  • ConsoleSink: Logs metrics information to the console.
  • CSVSink: Exports metrics data to CSV files at regular intervals.
  • JmxSink: Registers metrics for viewing in a JMX console.
  • MetricsServlet: Adds a servlet within the existing Spark UI to serve metrics data as JSON data.
  • PrometheusServlet: (Experimental) Adds a servlet within the existing Spark UI to serve metrics data in Prometheus format.
  • GraphiteSink: Sends metrics to a Graphite node.
  • Slf4jSink: Sends metrics to slf4j as log entries.
  • StatsdSink: Sends metrics to a StatsD node.

The Prometheus Servlet mirrors the JSON data exposed by the Metrics Servlet and the REST API, but in a time-series format. The following are the equivalent Prometheus Servlet endpoints.

Component Port JSON End Point Prometheus End Point
Master 8080 /metrics/master/json/ /metrics/master/prometheus/
Master 8080 /metrics/applications/json/ /metrics/applications/prometheus/
Worker 8081 /metrics/json/ /metrics/prometheus/
Driver 4040 /metrics/json/ /metrics/prometheus/
Driver 4040 /api/v1/applications/{id}/executors/ /metrics/executors/prometheus/

Spark also supports a Ganglia sink which is not included in the default build due to licensing restrictions:

  • GangliaSink: Sends metrics to a Ganglia node or multicast group.

To install the GangliaSink you'll need to perform a custom build of Spark. Note that by embedding this library you will include LGPL-licensed code in your Spark package. For sbt users, set the SPARK_GANGLIA_LGPL environment variable before building. For Maven users, enable the -Pspark-ganglia-lgpl profile. In addition to modifying the cluster's Spark build user applications will need to link to the spark-ganglia-lgpl artifact.

The syntax of the metrics configuration file and the parameters available for each sink are defined in an example configuration file, $SPARK_HOME/conf/metrics.properties.template.

When using Spark configuration parameters instead of the metrics configuration file, the relevant parameter names are composed by the prefix spark.metrics.conf. followed by the configuration details, i.e. the parameters take the following form: spark.metrics.conf.[instance|*].sink.[sink_name].[parameter_name]. This example shows a list of Spark configuration parameters for a Graphite sink:

"spark.metrics.conf.*.sink.graphite.class"="org.apache.spark.metrics.sink.GraphiteSink"
"spark.metrics.conf.*.sink.graphite.host"="graphiteEndPoint_hostName>"
"spark.metrics.conf.*.sink.graphite.port"=<graphite_listening_port>
"spark.metrics.conf.*.sink.graphite.period"=10
"spark.metrics.conf.*.sink.graphite.unit"=seconds
"spark.metrics.conf.*.sink.graphite.prefix"="optional_prefix"
"spark.metrics.conf.*.sink.graphite.regex"="optional_regex_to_send_matching_metrics"

Default values of the Spark metrics configuration are as follows:

"*.sink.servlet.class" = "org.apache.spark.metrics.sink.MetricsServlet"
"*.sink.servlet.path" = "/metrics/json"
"master.sink.servlet.path" = "/metrics/master/json"
"applications.sink.servlet.path" = "/metrics/applications/json"

Additional sources can be configured using the metrics configuration file or the configuration parameter spark.metrics.conf.[component_name].source.jvm.class=[source_name]. At present the JVM source is the only available optional source. For example the following configuration parameter activates the JVM source: "spark.metrics.conf.*.source.jvm.class"="org.apache.spark.metrics.source.JvmSource"

List of available metrics providers

Metrics used by Spark are of multiple types: gauge, counter, histogram, meter and timer, see Dropwizard library documentation for details. The following list of components and metrics reports the name and some details about the available metrics, grouped per component instance and source namespace. The most common time of metrics used in Spark instrumentation are gauges and counters. Counters can be recognized as they have the .count suffix. Timers, meters and histograms are annotated in the list, the rest of the list elements are metrics of type gauge. The large majority of metrics are active as soon as their parent component instance is configured, some metrics require also to be enabled via an additional configuration parameter, the details are reported in the list.

Component instance = Driver

This is the component with the largest amount of instrumented metrics

  • namespace=BlockManager

    • disk.diskSpaceUsed_MB
    • memory.maxMem_MB
    • memory.maxOffHeapMem_MB
    • memory.maxOnHeapMem_MB
    • memory.memUsed_MB
    • memory.offHeapMemUsed_MB
    • memory.onHeapMemUsed_MB
    • memory.remainingMem_MB
    • memory.remainingOffHeapMem_MB
    • memory.remainingOnHeapMem_MB

  • namespace=HiveExternalCatalog

    • note: these metrics are conditional to a configuration parameter: spark.metrics.staticSources.enabled (default is true)
    • fileCacheHits.count
    • filesDiscovered.count
    • hiveClientCalls.count
    • parallelListingJobCount.count
    • partitionsFetched.count

  • namespace=CodeGenerator

    • note: these metrics are conditional to a configuration parameter: spark.metrics.staticSources.enabled (default is true)
    • compilationTime (histogram)
    • generatedClassSize (histogram)
    • generatedMethodSize (histogram)
    • sourceCodeSize (histogram)

  • namespace=DAGScheduler

    • job.activeJobs
    • job.allJobs
    • messageProcessingTime (timer)
    • stage.failedStages
    • stage.runningStages
    • stage.waitingStages

  • namespace=LiveListenerBus

    • listenerProcessingTime.org.apache.spark.HeartbeatReceiver (timer)
    • listenerProcessingTime.org.apache.spark.scheduler.EventLoggingListener (timer)
    • listenerProcessingTime.org.apache.spark.status.AppStatusListener (timer)
    • numEventsPosted.count
    • queue.appStatus.listenerProcessingTime (timer)
    • queue.appStatus.numDroppedEvents.count
    • queue.appStatus.size
    • queue.eventLog.listenerProcessingTime (timer)
    • queue.eventLog.numDroppedEvents.count
    • queue.eventLog.size
    • queue.executorManagement.listenerProcessingTime (timer)

  • namespace=appStatus (all metrics of type=counter)

    • note: Introduced in Spark 3.0. Conditional to a configuration parameter: spark.metrics.appStatusSource.enabled (default is true)
    • stages.failedStages.count
    • stages.skippedStages.count
    • stages.completedStages.count
    • tasks.blackListedExecutors.count // deprecated use excludedExecutors instead
    • tasks.excludedExecutors.count
    • tasks.completedTasks.count
    • tasks.failedTasks.count
    • tasks.killedTasks.count
    • tasks.skippedTasks.count
    • tasks.unblackListedExecutors.count // deprecated use unexcludedExecutors instead
    • tasks.unexcludedExecutors.count
    • jobs.succeededJobs
    • jobs.failedJobs
    • jobDuration

  • namespace=AccumulatorSource

    • note: User-configurable sources to attach accumulators to metric system
    • DoubleAccumulatorSource
    • LongAccumulatorSource

  • namespace=spark.streaming

    • note: This applies to Spark Structured Streaming only. Conditional to a configuration parameter: spark.sql.streaming.metricsEnabled=true (default is false)
    • eventTime-watermark
    • inputRate-total
    • latency
    • processingRate-total
    • states-rowsTotal
    • states-usedBytes

  • namespace=JVMCPU

    • jvmCpuTime

  • namespace=executor

    • note: These metrics are available in the driver in local mode only.
    • A full list of available metrics in this namespace can be found in the corresponding entry for the Executor component instance.

  • namespace=ExecutorMetrics

    • note: these metrics are conditional to a configuration parameter: spark.metrics.executorMetricsSource.enabled (default is true)
    • This source contains memory-related metrics. A full list of available metrics in this namespace can be found in the corresponding entry for the Executor component instance.

  • namespace=ExecutorAllocationManager

    • note: these metrics are only emitted when using dynamic allocation. Conditional to a configuration parameter spark.dynamicAllocation.enabled (default is false)
    • executors.numberExecutorsToAdd
    • executors.numberExecutorsPendingToRemove
    • executors.numberAllExecutors
    • executors.numberTargetExecutors
    • executors.numberMaxNeededExecutors
    • executors.numberDecommissioningExecutors
    • executors.numberExecutorsGracefullyDecommissioned.count
    • executors.numberExecutorsDecommissionUnfinished.count
    • executors.numberExecutorsExitedUnexpectedly.count
    • executors.numberExecutorsKilledByDriver.count

  • namespace=plugin.<Plugin Class Name>

    • Optional namespace(s). Metrics in this namespace are defined by user-supplied code, and configured using the Spark plugin API. See "Advanced Instrumentation" below for how to load custom plugins into Spark.

Component instance = Executor

These metrics are exposed by Spark executors.

  • namespace=executor (metrics are of type counter or gauge)

    • notes:
      • spark.executor.metrics.fileSystemSchemes (default: file,hdfs) determines the exposed file system metrics.
    • bytesRead.count
    • bytesWritten.count
    • cpuTime.count
    • deserializeCpuTime.count
    • deserializeTime.count
    • diskBytesSpilled.count
    • filesystem.file.largeRead_ops
    • filesystem.file.read_bytes
    • filesystem.file.read_ops
    • filesystem.file.write_bytes
    • filesystem.file.write_ops
    • filesystem.hdfs.largeRead_ops
    • filesystem.hdfs.read_bytes
    • filesystem.hdfs.read_ops
    • filesystem.hdfs.write_bytes
    • filesystem.hdfs.write_ops
    • jvmGCTime.count
    • memoryBytesSpilled.count
    • recordsRead.count
    • recordsWritten.count
    • resultSerializationTime.count
    • resultSize.count
    • runTime.count
    • shuffleBytesWritten.count
    • shuffleFetchWaitTime.count
    • shuffleLocalBlocksFetched.count
    • shuffleLocalBytesRead.count
    • shuffleRecordsRead.count
    • shuffleRecordsWritten.count
    • shuffleRemoteBlocksFetched.count
    • shuffleRemoteBytesRead.count
    • shuffleRemoteBytesReadToDisk.count
    • shuffleTotalBytesRead.count
    • shuffleWriteTime.count
    • Metrics related to push-based shuffle:
      • shuffleCorruptMergedBlockChunks
      • shuffleMergedFetchFallbackCount
      • shuffleMergedRemoteBlocksFetched
      • shuffleMergedLocalBlocksFetched
      • shuffleMergedRemoteChunksFetched
      • shuffleMergedLocalChunksFetched
      • shuffleMergedRemoteBytesRead
      • shuffleMergedLocalBytesRead
      • shuffleRemoteReqsDuration
      • shuffleMergedRemoteReqsDuration
    • succeededTasks.count
    • threadpool.activeTasks
    • threadpool.completeTasks
    • threadpool.currentPool_size
    • threadpool.maxPool_size
    • threadpool.startedTasks

  • namespace=ExecutorMetrics

    • notes:
      • These metrics are conditional to a configuration parameter: spark.metrics.executorMetricsSource.enabled (default value is true)
      • ExecutorMetrics are updated as part of heartbeat processes scheduled for the executors and for the driver at regular intervals: spark.executor.heartbeatInterval (default value is 10 seconds)
      • An optional faster polling mechanism is available for executor memory metrics, it can be activated by setting a polling interval (in milliseconds) using the configuration parameter spark.executor.metrics.pollingInterval
    • JVMHeapMemory
    • JVMOffHeapMemory
    • OnHeapExecutionMemory
    • OnHeapStorageMemory
    • OnHeapUnifiedMemory
    • OffHeapExecutionMemory
    • OffHeapStorageMemory
    • OffHeapUnifiedMemory
    • DirectPoolMemory
    • MappedPoolMemory
    • MinorGCCount
    • MinorGCTime
    • MajorGCCount
    • MajorGCTime
    • "ProcessTree*" metric counters:
      • ProcessTreeJVMVMemory
      • ProcessTreeJVMRSSMemory
      • ProcessTreePythonVMemory
      • ProcessTreePythonRSSMemory
      • ProcessTreeOtherVMemory
      • ProcessTreeOtherRSSMemory
      • note: "ProcessTree*" metrics are collected only under certain conditions. The conditions are the logical AND of the following: /proc filesystem exists, spark.executor.processTreeMetrics.enabled=true. "ProcessTree*" metrics report 0 when those conditions are not met.

  • namespace=JVMCPU

    • jvmCpuTime

  • namespace=NettyBlockTransfer

    • shuffle-client.usedDirectMemory
    • shuffle-client.usedHeapMemory
    • shuffle-server.usedDirectMemory
    • shuffle-server.usedHeapMemory

  • namespace=HiveExternalCatalog

    • note: these metrics are conditional to a configuration parameter: spark.metrics.staticSources.enabled (default is true)
    • fileCacheHits.count
    • filesDiscovered.count
    • hiveClientCalls.count
    • parallelListingJobCount.count
    • partitionsFetched.count

  • namespace=CodeGenerator

    • note: these metrics are conditional to a configuration parameter: spark.metrics.staticSources.enabled (default is true)
    • compilationTime (histogram)
    • generatedClassSize (histogram)
    • generatedMethodSize (histogram)
    • sourceCodeSize (histogram)

  • namespace=plugin.<Plugin Class Name>

    • Optional namespace(s). Metrics in this namespace are defined by user-supplied code, and configured using the Spark plugin API. See "Advanced Instrumentation" below for how to load custom plugins into Spark.

Source = JVM Source

Notes:

  • Activate this source by setting the relevant metrics.properties file entry or the configuration parameter:spark.metrics.conf.*.source.jvm.class=org.apache.spark.metrics.source.JvmSource
  • These metrics are conditional to a configuration parameter: spark.metrics.staticSources.enabled (default is true)
  • This source is available for driver and executor instances and is also available for other instances.
  • This source provides information on JVM metrics using the Dropwizard/Codahale Metric Sets for JVM instrumentation and in particular the metric sets BufferPoolMetricSet, GarbageCollectorMetricSet and MemoryUsageGaugeSet.

Component instance = applicationMaster

Note: applies when running on YARN

  • numContainersPendingAllocate
  • numExecutorsFailed
  • numExecutorsRunning
  • numLocalityAwareTasks
  • numReleasedContainers

Component instance = master

Note: applies when running in Spark standalone as master

  • workers
  • aliveWorkers
  • apps
  • waitingApps

Component instance = ApplicationSource

Note: applies when running in Spark standalone as master

  • status
  • runtime_ms
  • cores

Component instance = worker

Note: applies when running in Spark standalone as worker

  • executors
  • coresUsed
  • memUsed_MB
  • coresFree
  • memFree_MB

Component instance = shuffleService

Note: applies to the shuffle service

  • blockTransferRate (meter) - rate of blocks being transferred

  • blockTransferMessageRate (meter) - rate of block transfer messages, i.e. if batch fetches are enabled, this represents number of batches rather than number of blocks

  • blockTransferRateBytes (meter)

  • blockTransferAvgSize_1min (gauge - 1-minute moving average)

  • numActiveConnections.count

  • numRegisteredConnections.count

  • numCaughtExceptions.count

  • openBlockRequestLatencyMillis (timer)

  • registerExecutorRequestLatencyMillis (timer)

  • fetchMergedBlocksMetaLatencyMillis (timer)

  • finalizeShuffleMergeLatencyMillis (timer)

  • registeredExecutorsSize

  • shuffle-server.usedDirectMemory

  • shuffle-server.usedHeapMemory

  • note: the metrics below apply when the server side configuration spark.shuffle.push.server.mergedShuffleFileManagerImpl is set to org.apache.spark.network.shuffle.MergedShuffleFileManager for Push-Based Shuffle

  • blockBytesWritten - size of the pushed block data written to file in bytes

  • blockAppendCollisions - number of shuffle push blocks collided in shuffle services as another block for the same reduce partition were being written

  • lateBlockPushes - number of shuffle push blocks that are received in shuffle service after the specific shuffle merge has been finalized

  • deferredBlocks - number of the current deferred block parts buffered in memory

  • deferredBlockBytes - size of the current deferred block parts buffered in memory

  • staleBlockPushes - number of stale shuffle block push requests

  • ignoredBlockBytes - size of the pushed block data that was transferred to ESS, but ignored. The pushed block data are considered as ignored when: 1. it was received after the shuffle was finalized; 2. when a push request is for a duplicate block; 3. ESS was unable to write the block.

Advanced Instrumentation

Several external tools can be used to help profile the performance of Spark jobs:

  • Cluster-wide monitoring tools, such as Ganglia, can provide insight into overall cluster utilization and resource bottlenecks. For instance, a Ganglia dashboard can quickly reveal whether a particular workload is disk bound, network bound, or CPU bound.
  • OS profiling tools such as dstat, iostat, and iotop can provide fine-grained profiling on individual nodes.
  • JVM utilities such as jstack for providing stack traces, jmap for creating heap-dumps, jstat for reporting time-series statistics and jconsole for visually exploring various JVM properties are useful for those comfortable with JVM internals.

Spark also provides a plugin API so that custom instrumentation code can be added to Spark applications. There are two configuration keys available for loading plugins into Spark:

  • spark.plugins
  • spark.plugins.defaultList

Both take a comma-separated list of class names that implement the org.apache.spark.api.plugin.SparkPlugin interface. The two names exist so that it's possible for one list to be placed in the Spark default config file, allowing users to easily add other plugins from the command line without overwriting the config file's list. Duplicate plugins are ignored.