Installing-spark-standalone-to-a-cluster-manually

About

Next to running Spark on the YARN cluster manager, Spark itself provides a standalone deploy mode. We can launch a standalone cluster manually by starting a master and workers daemons by hand. This approach makes testing easier by running these daemons on a single machine. I'll walk you through launching the cluster, as well as connecting an app to the cluster, launching the app, where to view the monitoring and logging.

Start a cluster

First we need to start spark standalone master serve {apache spark directory/libexec}/sbin/start-master.sh

Now open http://localhost:8080/ in the browser, and copy the spark url (spark://HOSTNAME:PORT)

Second we can start one or more workers and connect them to the master {apache spark directory/libexec}/sbin/start-slave.sh <master-spark-URL> {apache spark directory/libexec}/sbin/start-slaves.sh <master-spark-URL>

If you look again at th spark master’s web UI (http://localhost:8080). You'll see the new node/s (worker/s) listed there.

Connecting an app to the cluster

Either run the application on the Spark cluster by passing the spark://HOSTNAME:PORT to the SparkContext constructor.

Or to run an interactive Spark shell against the cluster, run the following command:

spark-shell --master spark://HOSTNAME:PORT

Launching an app

To launch a compiled Spark application to a cluster is by using the spark-submit script.
spark-submit --class <main-class> \

--master <master-url> \

--deploy-mode <deploy-mode> \

--conf <key>=<value> \

... # other options

<application-jar> \

[application-arguments]

Explanations

--class: your application entry point

--master: The master URL for the cluster (e.g.local for one worker local[k] for k workers)

--deploy-mode: default: client

--conf: Spark configuration property in key=value format (e.g to enable history, --conf spark.eventLog.enabled=true).

application-jar: a bundled jar for all of your application's dependencies.

application-arguments: Arguments passed to the main method in the main class (e.g. could be files' path as input and output).

Monitoring and Logging

Spark’s standalone mode offers a user interface to monitor the cluster called 'spark server history'

setup server history

The Spark History server allows us to review Spark application metrics after the application has been completed.

1. Create ~/tmp/spark-events

2. Create a file spark-defaults.conf under the path [SPARK_PATH]/libexec/conf )

3. Add these lines in the spark-defaults.conf file:

spark.eventLog.enabled true

spark.eventLog.dir file:/Users/<USER_NAME>/tmp/spark-events

spark.history.fs.logDirectory file:/Users/<USER_NAME>/tmp/spark-events

4. Start the history server : [SPARK_PATH]/sbin/start-history-server.sh

5. You can see the history http://localhost:18080

Refernces:

https://spark.apache.org/docs/latest/spark-standalone.html#cluster-launch-scripts

https://spark.apache.org/docs/latest/submitting-applications.html

https://spark.apache.org/docs/latest/monitoring.html