This page describes how to pre-configure a bare metal node, configure Zeppelin and connect it to existing YARN cluster running Hortonworks flavour of Hadoop. It also describes steps to configure Spark interpreter of Zeppelin.

Prepare Node

Zeppelin user (Optional)

This step is optional, however its nice to run Zeppelin under its own user. In case you do not like to use Zeppelin (hope not) the user could be deleted along with all the packages that were installed for Zeppelin, Zeppelin binary itself and associated directories.

Create a zeppelin user and switch to zeppelin user or if zeppelin user is already created then login as zeppelin.

useradd zeppelin
su - zeppelin

Assuming a zeppelin user is created then running whoami command must return


Its assumed in the rest of the document that zeppelin user is indeed created and below installation instructions are performed as zeppelin user.

List of Prerequisites

  • CentOS 6.x, Mac OSX, Ubuntu 14.X
  • Java 1.7
  • Hadoop client
  • Spark
  • Internet connection is required.

It's assumed that the node has CentOS 6.x installed on it. Although any version of Linux distribution should work fine.

Hadoop client

Zeppelin can work with multiple versions & distributions of Hadoop. A complete list is available here. This document assumes Hadoop 2.7.x client libraries including configuration files are installed on Zeppelin node. It also assumes /etc/hadoop/conf contains various Hadoop configuration files. The location of Hadoop configuration files may vary, hence use appropriate location.

hadoop version
Subversion -r f66cf95e2e9367a74b0ec88b2df33458b6cff2d0
Compiled by jenkins on 2015-07-25T22:36Z
Compiled with protoc 2.5.0
From source with checksum 54f9bbb4492f92975e84e390599b881d
This command was run using /usr/hdp/


Spark is supported out of the box and to take advantage of this, you need to Download appropriate version of Spark binary packages from Spark Download page and unzip it. Zeppelin can work with multiple versions of Spark. A complete list is available here. This document assumes Spark 1.6.0 is installed at /usr/lib/spark.

Note: Spark should be installed on the same node as Zeppelin.

Note: Spark's pre-built package for CDH 4 doesn't support yarn.


Checkout source code from git:// or download binary package from Download page. You can refer Install page for the details. This document assumes that Zeppelin is located under /home/zeppelin/zeppelin.

Zeppelin Configuration

Zeppelin configuration needs to be modified to connect to YARN cluster. Create a copy of zeppelin environment shell script.

cp /home/zeppelin/zeppelin/conf/ /home/zeppelin/zeppelin/conf/

Set the following properties

export JAVA_HOME="/usr/java/jdk1.7.0_79"
export HADOOP_CONF_DIR="/etc/hadoop/conf"
export ZEPPELIN_JAVA_OPTS="-Dhdp.version="
export SPARK_HOME="/usr/lib/spark"

As /etc/hadoop/conf contains various configurations of YARN cluster, Zeppelin can now submit Spark/Hive jobs on YARN cluster form its web interface. The value of hdp.version is set to This can be obtained by running the following command

hdp-select status hadoop-client | sed 's/hadoop-client - \(.*\)/\1/'
# It returned


Start Zeppelin

cd /home/zeppelin/zeppelin
bin/ start

After successful start, visit http://[zeppelin-server-host-name]:8080 with your web browser.

Stop Zeppelin

bin/ stop


Zeppelin provides various distributed processing frameworks to process data that ranges from Spark, JDBC, Ignite and Lens to name a few. This document describes to configure JDBC & Spark interpreters.


Zeppelin supports Hive through JDBC interpreter. You might need the information to use Hive and can find in your hive-site.xml

Once Zeppelin server has started successfully, visit http://[zeppelin-server-host-name]:8080 with your web browser. Click on Interpreter tab next to Notebook dropdown. Look for Hive configurations and set them appropriately. Set them as per Hive installation on YARN cluster. Click on Save button. Once these configurations are updated, Zeppelin will prompt you to restart the interpreter. Accept the prompt and the interpreter will reload the configurations.


It was assumed that 1.6.0 version of Spark is installed at /usr/lib/spark. Look for Spark configurations and click edit button to add the following properties

Property Name Property Value Remarks
master 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.
spark.driver.extraJavaOptions -Dhdp.version= -Dhdp.version=

Click on Save button. Once these configurations are updated, Zeppelin will prompt you to restart the interpreter. Accept the prompt and the interpreter will reload the configurations.

Spark & Hive notebooks can be written with Zeppelin now. The resulting Spark & Hive jobs will run on configured YARN cluster.


Zeppelin does not emit any kind of error messages on web interface when notebook/paragraph is run. If a paragraph fails it only displays ERROR. The reason for failure needs to be looked into log files which is present in logs directory under zeppelin installation base directory. Zeppelin creates a log file for each kind of interpreter.

[zeppelin@zeppelin-3529 logs]$ pwd
[zeppelin@zeppelin-3529 logs]$ ls -l
total 844
-rw-rw-r-- 1 zeppelin zeppelin  14648 Aug  3 14:45 zeppelin-interpreter-hive-zeppelin-zeppelin-3529.log
-rw-rw-r-- 1 zeppelin zeppelin 625050 Aug  3 16:05 zeppelin-interpreter-spark-zeppelin-zeppelin-3529.log
-rw-rw-r-- 1 zeppelin zeppelin 200394 Aug  3 21:15 zeppelin-zeppelin-zeppelin-3529.log
-rw-rw-r-- 1 zeppelin zeppelin  16162 Aug  3 14:03 zeppelin-zeppelin-zeppelin-3529.out
[zeppelin@zeppelin-3529 logs]$