Bigdata Dev Apache Spark

Channel Version Revision Published Runs on
latest/stable 83 83 18 Mar 2021
Ubuntu 16.04 Ubuntu 14.04
latest/edge 70 70 18 Mar 2021
Ubuntu 14.04
juju deploy bigdata-dev-apache-spark
Show information


16.04 14.04


Spark Cluster

Apache Spark™ is a fast and general purpose engine for large-scale data processing. Key features:

  • Speed

    Run programs up to 100x faster than Hadoop MapReduce in memory, or 10x faster on disk. Spark has an advanced DAG execution engine that supports cyclic data flow and in-memory computing.

  • Ease of Use

    Write applications quickly in Java, Scala or Python. Spark offers over 80 high-level operators that make it easy to build parallel apps, and you can use it interactively from the Scala and Python shells.

  • General Purpose Engine

    Combine SQL, streaming, and complex analytics. Spark powers a stack of high-level tools including Shark for SQL, MLlib for machine learning, GraphX, and Spark Streaming. You can combine these frameworks seamlessly in the same application.


This charm allows the deployment of Apache Spark in the modes described below:

  • Standalone

    In this mode Spark units form a cluster that you can scale to match your needs. Starting with a single node:

    juju deploy apache-spark spark

    You can scale the cluster by adding more spark units:

    juju add-unit spark

    When in standalone mode Juju ensures a single Spark master is appointed. The status of the unit acting as master reads "Ready (standalone - master)", while the rest of the units display a status of "Ready (standalone)". In case you remove the master unit Juju will appoint a new master to the cluster. However, should a master fail in this standalone mode the cluster will stop functioning properly. Master node failures is handled properly when Apache spark is setup in High Availability mode (Standalone HA).

  • Standalone HA

    To enable High Availability properties of a cluster you need to add a relation between spark and a zookeeper deployment. The suggested deployment method is to use the apache-spark-HA bundle.

    juju-quickstart apache-spark-ha

    In this mode again you can scale your cluster to match your needs by adding/removing units. Spark units report "Ready (standalone HA)" in their status so if you need to identify the node acting as master you need to query the Zookeeper deployment.

    juju ssh zk/0 get /spark/master_status

  • Yarn-client and Yarn-cluster

    This charm leverages our pluggable Hadoop model with the hadoop-plugin interface. This means that you can relate this charm to a base Apache Hadoop cluster to run Spark jobs there. The suggested deployment method is to use the apache-hadoop-spark bundle. This will deploy the Apache Hadoop platform with a single Apache Spark unit that communicates with the cluster by relating to the apache-hadoop-plugin subordinate charm:

    juju-quickstart apache-hadoop-spark

Note: To transition among execution modes you need to set the spark-execution-mode config variable:

juju set spark spark-execution-mode=<new_mode>


Once deployment is complete, you can manually load and run Spark batch or streaming jobs in a variety of ways:

  • Spark shell

Spark’s shell provides a simple way to learn the API, as well as a powerful tool to analyse data interactively. It is available in either Scala or Python and can be run from the Spark unit as follows:

   juju ssh spark/0
   spark-shell # for interaction using scala
   pyspark     # for interaction using python
  • Command line

SSH to the Spark unit and manually run a spark-submit job, for example:

   juju ssh spark/0
   spark-submit --class org.apache.spark.examples.SparkPi \
    --master yarn-client /usr/lib/spark/lib/spark-examples*.jar 10
  • Apache Zeppelin visual service

Deploy Apache Zeppelin and relate it to the Spark unit:

juju deploy apache-zeppelin zeppelin
juju add-relation spark zeppelin

Once the relation has been made, access the web interface at http://{spark_unit_ip_address}:9090

  • IPyNotebook for Spark

The IPython Notebook is an interactive computational environment, in which you can combine code execution, rich text, mathematics, plots and rich media. Deploy IPython Notebook for Spark and relate it to the Spark unit:

juju deploy apache-spark-notebook notebook
juju add-relation spark notebook

Once the relation has been made, access the web interface at http://{spark_unit_ip_address}:8880



Amount of memory Spark will request for the Master. Specify gigabytes (e.g. 1g) or megabytes (e.g. 1024m). If running in local or standalone mode, you may also specify a percentage of total system memory (e.g. 50%).


Amount of memory Spark will request for each executor. Specify gigabytes (e.g. 1g) or megabytes (e.g. 1024m). If running in local or standalone mode, you may also specify a percentage of total system memory (e.g. 50%). Take care when specifying percentages in local modes, as this value is for each executor. Your Spark job will fail if, for example, you set this value > 50% and attempt to run 2 or more executors.


Install the SparkBench benchmarking suite. If true (the default), this charm will download spark bench from the URL specified by spark_bench_ppc64le or spark_bench_x86_64, depending on the unit's architecture.


Spark has four modes of execution: local, standalone, yarn-client, and yarn-cluster. The default mode is yarn-client and can be changed by setting the spark_execution_mode config variable.

  • Local

    In Local mode, Spark processes jobs locally without any cluster resources. There are 3 ways to specify 'local' mode:

    • local

      Run Spark locally with one worker thread (i.e. no parallelism at all).

    • local[K]

      Run Spark locally with K worker threads (ideally, set this to the number of cores on your machine).

    • local[*]

      Run Spark locally with as many worker threads as logical cores on your machine.

  • Standalone

    In standalone mode, Spark launches a Master and Worker daemon on the Spark unit. This mode is useful for simulating a distributed cluster environment without actually setting up a cluster.

  • YARN-client

    In yarn-client mode, the driver runs in the client process, and the application master is only used for requesting resources from YARN.

  • YARN-cluster

    In yarn-cluster mode, the Spark driver runs inside an application master process which is managed by YARN on the cluster, and the client can go away after initiating the application.

Verify the deployment

Status and Smoke Test

The services provide extended status reporting to indicate when they are ready:

juju status --format=tabular

This is particularly useful when combined with watch to track the on-going progress of the deployment:

watch -n 0.5 juju status --format=tabular

The message for each unit will provide information about that unit's state. Once they all indicate that they are ready, you can perform a "smoke test" to verify that Spark is working as expected using the built-in smoke-test action:

juju action do spark/0 smoke-test

After a few seconds or so, you can check the results of the smoke test:

juju action status

You will see status: completed if the smoke test was successful, or status: failed if it was not. You can get more information on why it failed via:

juju action fetch <action-id>
Verify Job History

The Job History server shows all active and finished spark jobs submitted. To view the Job History server you need to expose spark (juju expose spark) and navigate to http://{spark_master_unit_ip_address}:18080 of the unit acting as master.


This charm provides several benchmarks, including the Spark Bench benchmarking suite (if enabled), to gauge the performance of your environment.

The easiest way to run the benchmarks on this service is to relate it to the Benchmark GUI. You will likely also want to relate it to the Benchmark Collector to have machine-level information collected during the benchmark, for a more complete picture of how the machine performed.

However, each benchmark is also an action that can be called manually:

$ juju action do spark/0 pagerank
Action queued with id: 88de9367-45a8-4a4b-835b-7660f467a45e
$ juju action fetch --wait 0 88de9367-45a8-4a4b-835b-7660f467a45e
      direction: asc
      units: secs
      value: "77.939000"
    raw: |
    start: 2015-12-10T23:41:34Z
    stop: 2015-12-10T23:43:16Z
      direction: asc
      units: secs
      value: "77.939000"
      direction: desc
      units: x/sec
      value: ".922363"
status: completed
  completed: 2015-12-10 23:43:59 +0000 UTC
  enqueued: 2015-12-10 23:42:10 +0000 UTC
  started: 2015-12-10 23:42:15 +0000 UTC

Valid action names at this time are:

  • logisticregression
  • matrixfactorization
  • pagerank
  • sql
  • streaming
  • svdplusplus
  • svm
  • trianglecount
  • sparkpi

Upgrading and Apache Bigtop Spark

This charm does not currently support upgrading, because it currently only supports a single version of Apache Spark. It is recommended to look to the Apache Bigtop Spark charm for support of future versions of Apache Spark, via the Apache Bigtop project.

Contact Information