Enabling Spark Optimization through Cross-stack Monitoring and Visualization
Clone or download
Latest commit 2bc64a2 Aug 23, 2017
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
.github
R
assembly
bin [SPARK-1267][SPARK-18129] Allow PySpark to be pip installed Nov 17, 2016
build [SPARK-14279][BUILD] Pick the spark version from pom Jun 6, 2016
common
conf
core
data
dev
docs Preparing Spark release v2.1.0-rc5 Dec 16, 2016
examples Preparing Spark release v2.1.0-rc5 Dec 16, 2016
external Preparing Spark release v2.1.0-rc5 Dec 16, 2016
graphx
launcher Preparing Spark release v2.1.0-rc5 Dec 16, 2016
licenses
mesos Preparing Spark release v2.1.0-rc5 Dec 16, 2016
mllib-local Preparing Spark release v2.1.0-rc5 Dec 16, 2016
mllib
project [SPARK-18754][SS] Rename recentProgresses to recentProgress Dec 7, 2016
python
repl Preparing Spark release v2.1.0-rc5 Dec 16, 2016
sbin [SPARK-18645][DEPLOY] Fix spark-daemon.sh arguments error lead to thr… Dec 1, 2016
sql
streaming
tools
yarn Preparing Spark release v2.1.0-rc5 Dec 16, 2016
.gitattributes [SPARK-3870] EOL character enforcement Oct 31, 2014
.gitignore
.travis.yml
CONTRIBUTING.md
LICENSE
NOTICE
README.md instructions updated for yarn, closes #18 Aug 23, 2017
appveyor.yml
pom.xml Preparing Spark release v2.1.0-rc5 Dec 16, 2016
recompileAndStart.sh
runTest.sh
runTestStreaming.sh
runTestYarn.sh add more helper scripts Jul 31, 2017
scalastyle-config.xml

README.md

SparkOscope

Getting started

To get started easily with Sparkoscope you can use the docker image provided here

Starts a one-node hdfs cluster, starts Sparkoscope and gives bash shell to the user to execute the spark examples.

Run it by giving:

docker run --rm -it -p 4040:4040 -p 8080:8080 -p 18080:18080 -p 8888:8888 yiannisgkoufas/sparkoscope

Then once on the bash shell give:

bin/run-example SparkPi 10000

Check http://localhost:8080 and http://localhost:18080 on your host for Spark Master and History Server respectively

Installation/Configuration

Configure Sigar metrics source

In all the nodes of the cluster Hyperic Sigar library must be installed. Download from http://sourceforge.net/projects/sigar/files/sigar/1.6/hyperic-sigar-1.6.4.zip/download. Extract the zip in any location.

In spark-env.sh you need to add to LD_LIBRARY_PATH variable the directory of the native libraries of Sigar. For instance:

LD_LIBRARY_PATH=/path/to/hyperic-sigar-1.6.4/sigar-bin/lib/:$LD_LIBRARY_PATH
#In case you plan to use Sparkoscope on Yarn
SPARK_YARN_USER_ENV="LD_LIBRARY_PATH=$LD_LIBRARY_PATH"

Add the source definition to metrics.properties

executor.source.jvm.class=org.apache.spark.metrics.source.SigarSource

Configure hadoop

In spark-env.sh you need to set the HADOOP_CONF_DIR variable to the configuration directory of your hadoop installation. For instance:

HADOOP_CONF_DIR=/path/to/hadoop/etc/hadoop

Configure hdfs metrics sink

In order for the executor metrics to be stored in HDFS and therefore be retrieved by the UI, you need to have the following in the metrics.properties file:

executor.sink.hdfs.class=org.apache.spark.metrics.sink.HDFSSink
executor.sink.hdfs.pollPeriod = 20
executor.sink.hdfs.dir = hdfs://localhost:9000/custom-metrics
executor.sink.hdfs.unit = seconds

Realtime Plots configuration

The same metrics that are exposed in the history server, now they can be viewed in real time on the application page.

As the application is running, the plots can be viewed in http://masterIP:8080/app/?appId=app-XXXXXXXXX-XXX You should modify the spark-defaults.conf the following way:

spark.moquette.port 1883
spark.moquette.websocket_port 8888

In the metrics.properties of every worker you should add the following entries:

executor.sink.mqtt.class=org.apache.spark.metrics.sink.MQTTSink
executor.sink.mqtt.pollPeriod = 1
executor.sink.mqtt.host = masterIP
executor.sink.mqtt.port = 1883
executor.sink.mqtt.unit = seconds

Where executor.sink.mqtt.port is the same as spark.moquette.port and masterIP is the host where Spark Master is running

Event and UI configuration

Start history server

sbin/start-history-server.sh

Event logging must be enabled in spark-defaults.conf:

spark.eventLog.enabled           true
spark.eventLog.dir               hdfs://127.0.0.1:9000/spark-logs
spark.history.fs.logDirectory    hdfs://127.0.0.1:9000/spark-logs

Also in spark-defaults.conf you should specify the folder from which the UI will read the metrics:

spark.hdfs.metrics.dir           hdfs://127.0.0.1:9000/custom-metrics

Access history server in http://localhost:18080/

Click the specific application and view the plots

Notes about the metrics

The metrics are grouped per application and the user can access the plots by selecting the Name entry under the Completed Applications table. The URL on the browser should look similar to http://ip-of-spark-master:port/history/app-201511XXXXXX-XXX Under the dropdown menu Executor Metrics the user can plot any of the metrics provided per executor but also metrics of the operating system of the host (physical or virtual):

sigar.ram

Percentage of RAM utilization

sigar.cpu

Percentage of CPU utilization

sigar.kBytesRxPerSecond / sigar.kBytesTxPerSecond

Number of Kilobytes received/transmitted from/to the network per second

sigar.kBytesReadPerSecond / sigar.kBytesWrittenPerSecond

Number of Kilobytes read/written from/to the disk per second

Important:

The folders spark.eventLog.dir, executor.sink.hdfs.dir and spark.hdfs.metrics.dir must already exist in the HDFS.

You should increase the limit for open files on the operating systems of the Master and the Workers.

Be sure to build spark according to the version of hadoop you are using.

Apache Spark

Spark is a fast and general cluster computing system for Big Data. It provides high-level APIs in Scala, Java, Python, and R, and an optimized engine that supports general computation graphs for data analysis. It also supports a rich set of higher-level tools including Spark SQL for SQL and DataFrames, MLlib for machine learning, GraphX for graph processing, and Spark Streaming for stream processing.

http://spark.apache.org/

Online Documentation

You can find the latest Spark documentation, including a programming guide, on the project web page. This README file only contains basic setup instructions.

Building Spark

Spark is built using Apache Maven. To build Spark and its example programs, run:

build/mvn -DskipTests clean package

(You do not need to do this if you downloaded a pre-built package.)

You can build Spark using more than one thread by using the -T option with Maven, see "Parallel builds in Maven 3". More detailed documentation is available from the project site, at "Building Spark".

For general development tips, including info on developing Spark using an IDE, see [http://spark.apache.org/developer-tools.html](the Useful Developer Tools page).

Interactive Scala Shell

The easiest way to start using Spark is through the Scala shell:

./bin/spark-shell

Try the following command, which should return 1000:

scala> sc.parallelize(1 to 1000).count()

Interactive Python Shell

Alternatively, if you prefer Python, you can use the Python shell:

./bin/pyspark

And run the following command, which should also return 1000:

>>> sc.parallelize(range(1000)).count()

Example Programs

Spark also comes with several sample programs in the examples directory. To run one of them, use ./bin/run-example <class> [params]. For example:

./bin/run-example SparkPi

will run the Pi example locally.

You can set the MASTER environment variable when running examples to submit examples to a cluster. This can be a mesos:// or spark:// URL, "yarn" to run on YARN, and "local" to run locally with one thread, or "local[N]" to run locally with N threads. You can also use an abbreviated class name if the class is in the examples package. For instance:

MASTER=spark://host:7077 ./bin/run-example SparkPi

Many of the example programs print usage help if no params are given.

Running Tests

Testing first requires building Spark. Once Spark is built, tests can be run using:

./dev/run-tests

Please see the guidance on how to run tests for a module, or individual tests.

A Note About Hadoop Versions

Spark uses the Hadoop core library to talk to HDFS and other Hadoop-supported storage systems. Because the protocols have changed in different versions of Hadoop, you must build Spark against the same version that your cluster runs.

Please refer to the build documentation at "Specifying the Hadoop Version" for detailed guidance on building for a particular distribution of Hadoop, including building for particular Hive and Hive Thriftserver distributions.

Configuration

Please refer to the Configuration Guide in the online documentation for an overview on how to configure Spark.

Contributing

Please review the Contribution to Spark guide for information on how to get started contributing to the project.