Elasticsearch on YARN allows through the command-line to provision, start, monitor and stop an Elasticsearch cluster inside a YARN environment.

Simply download elasticsearch-yarn-<version>.jar in a location of choice and make sure to have Hadoop/YARN available and configured in your classpath; double check through the hadoop version command:

$ hadoop version

Hadoop 2.4.1
Subversion -r 1604318
Compiled by jenkins on 2014-06-21T05:43Z
Compiled with protoc 2.5.0
From source with checksum bb7ac0a3c73dc131f4844b873c74b630
This command was run using /opt/share/hadoop/common/hadoop-common-2.4.1.jar

Once you have confirmed Hadoop is properly configured, do a basic sanity check for Elasticsearch on YARN by invoking hadoop jar:

Some Hadoop distros might display a warning to use yarn jar instead; the two commands can be used interchangeably

$ hadoop jar elasticsearch-yarn-<version>.jar
No command specified
     -download-es  : Downloads
     -install      : Installs/Provisions Elasticsearch-YARN into HDFS
     -install-es   : Installs/Provisions Elasticsearch into HDFS
     -start        : Starts provisioned Elasticsearch in YARN
     -status       : Reports status of Elasticsearch in YARN
     -stop         : Stops Elasticsearch in YARN
     -help         : Prints this help

Configuration options can be specified _after_ each command; see the documentation for more information.

Each command should be self-explanatory. The typical usage scenario is:

Download the Elasticsearch version needededit

This is a one-time action; if you already have Elasticsearch at hand, deploy it under downloads sub-folder. To wit:

$ hadoop jar elasticsearch-yarn-<version> -download-es
Downloading Elasticsearch <version>
Downloading ...........................................................................DONE

If you want to use a different version of Elasticsearch, you can specify so through the es.version parameter (see the Configuration section.

(Optional) Configuring Elasticsearchedit

If the default Elasticsearch is not suitable, for example certain plugins need to be installed or the storage path needs to be defined as mentioned here, one should do so by modifying the Elasticsearch archive as it will serve as the blueprint of all nodes installed within YARN.

Provision Elasticsearch into HDFSedit

Now that we have downloaded Elasticsearch, let us upload it into HDFS so it becomes available to the Hadoop nodes. Make sure that the proper user (typically hdfs) is used for writing data to HDFS; otherwise a permission denied exception will be thrown. This is another one-time action (as long as your HDFS cluster and the target location remain in place):

$ hadoop jar elasticsearch-yarn-<version>.jar -install-es

Uploaded /opt/es-yarn/downloads/elasticsearch-<version>.zip to HDFS at hdfs://<version>.zip

This command uploads the elasticsearch-<version>.zip (that we just downloaded) to HDFS (based on the Hadoop configuration detected in the classpath) under /apps/elasticsearch folder. Again the location can be changed if needed.

Note that the uploaded ZIP can be configured accordingly to your setup - for example, one can include her own configuration (such as using a certain storage location) to override the defaults or for example certain plugins. The zip acts as a template so be sure to include everything that need in it.

Provision Elasticsearch-YARN into HDFSedit

Let us do the same one-time command with the Elasticsearch-YARN jar:

$ hadoop jar elasticsearch-yarn-<version>.jar -install

Uploaded opt/es-yarn/elasticsearch-yarn-<version>.jar to HDFS at hdfs://<version>.jar

You can verify the provisioning by interrogating HDFS either through the web console or hadoop CLI:

$ hadoop dfs -ls /apps/elasticsearch
Found 2 items
-rw-r--r--   1 hdfs hdfs   30901787 2014-11-13 10:17 /apps/elasticsearch/elasticsearch-<version>.zip
-rw-r--r--   1 hdfs hdfs      52754 2014-11-13 16:16 /apps/elasticsearch/elasticsearch-yarn-<version>.jar

Start Elasticsearch on YARNedit

Once the necessary artifacts are in HDFS, one can start Elasticsearch:

$ hadoop jar elasticsearch-yarn-<version>.jar -start
Launched a 1 node Elasticsearch-YARN cluster [application_1415813090693_0001@http://hadoop:8088/proxy/application_1415921358606_0001/] at Wed Nov 14 19:24:53 EET 2014

By default only a single node is created; to start multiple nodes use the containers parameter (see Configuration for more information):

$ hadoop jar elasticsearch-yarn-<version>.jar -start containers=2
Launched a 2 nodes Elasticsearch-YARN cluster [application_1415921358606_0006@http://hadoop:8088/proxy/application_1415921358606_0006/] at Wed Nov 14 19:28:46 EET 2014

That’s it!

Get status of Elasticsearch clusters in YARNedit

There are plenty of tools in Hadoop to check running YARN applications; with Elasticsearch YARN try the -status command:

$ hadoop jar elasticsearch-yarn-<version>.jar -status
Id                              State       Status     Start Time         Finish Time        Tracking URL
application_1415921358606_0007  RUNNING     UNDEFINED  11/14/14 19:34 PM  N/A                http://hadoop:8088/proxy/application_1415921358606_0007/A

If you prefer the web ui, point your browser to the cluster console (typically on port 8088 - http://hadoop:8088/cluster) and the newly created Elasticsearch cluster, or rather its ApplicationMaster will show up:

AppManager List

One can inspect the containers manually by checking the container list (typically by accessing the ApplicationManager UI):

Container List

You should be able to see the application and the associated containers marked as RUNNING ; if that is not the case, check out the logs to see what is wrong.

Stop Elasticsearch clusters in YARNedit

To shutdown your cluster, use the -stop command:

$ hadoop jar elasticsearch-yarn-<version>.jar -stop
Stopped Elasticsearch-YARN cluster with id application_1415921358606_0007


Elasticsearch on YARN offers various knobs for tweaking its behavior - all can be passed as options after each command, overriding the default configuration. Multiple options can be specified (if you specify the same parameter multiple times, the last one wins):

$ hadoop jar elasticsearch-yarn-<version>.jar [-command] []=[option.value] []=[option.value]

The following parameter are available:

download.local.dir (default ./downloads/)
Local folder where Elasticsearch on YARN downloads remote artifacts (like
hdfs.upload.dir (default /apps/elasticsearch/)
HDFS folder used for provisioning
es.version (default 5.3.3)
Elasticsearch version used for downloading, provisioning HDFS and running on the YARN cluster
containers (default 1)
The number of containers or Elasticsearch nodes for starting the cluster
container.mem (default 2048)
Memory requested for each YARN container
container.vcores (default 1)
CPU cores requested for each YARN container
container.priority (default -1)
YARN queue priority for each container
Pattern for setting up environment variables on each container - each property starting with env. prefix will be set remotely on each container:
$ hadoop jar elasticsearch-yarn-<version>.jar [-command] env.ES_USE_GC_LOGGING=true env.PROP=someValue

Sets up on each container two variables ES_USE_GC_LOGGING and PROP.

Added in 2.2.

Pattern for setting up the JVM system property on each container - each property starting with sys.prop. prefix will be set remotely on each container:
$ hadoop jar elasticsearch-yarn-<version>.jar [-command]"access,failure"

Sets up on each JVM security debugging - not something one wants in a production environment.

Property file to be loaded as configuration. That is, instead of specifying the options in each command, save them to a file and use that instead. For example the above example, the two options (for setting the container environment variables) can be moved into a properties file

and tell Elasticsearch on YARN to load it:

$ hadoop jar elasticsearch-yarn-<version>.jar [-command]

Added in 5.0.

Name of the kerberos prinicpal that the Application Master should use when connecting to HDFS to retrieve parcels for provisioning containers. This defaults to the current system user name. When YARN is in secure container mode, the Application Master is executed using the user name that submitted the application originally, and thus, that original user name will be used if nothing is specified for this config. If a user specifies a _HOST pattern in the principal, then the pattern will be replaced with the environment’s hostname. This allows administrators to provision a different keytab file per host. To wit:

This principal would be unmodified no matter which host the Application Master starts on.

If the Application Manager starts on a node called pd-yarn-01 then this principal would be modified into elasticsearch/pd-yarn-01@SECURITY.REALM.

Path to a readable keytab file that is accessible on every node that the Application Master may start on. The keytab must contain valid credentials for the principal configured using the am.kerberos.principal setting, either in the format of user@REALM or user/host@REALM.