Deploying MapReduce v2 (YARN) on a Cluster

The default installation in CDH 5 is MapReduce 2.x (MRv2) built on the YARN framework. In this document we usually refer to this new version as YARN. The fundamental idea of MRv2's YARN architecture is to split up the two primary responsibilities of the JobTracker — resource management and job scheduling/monitoring — into separate daemons: a global ResourceManager (RM) and per-application ApplicationMasters (AM). With MRv2, the ResourceManager (RM) and per-node NodeManagers (NM), form the data-computation framework. The ResourceManager service effectively replaces the functions of the JobTracker, and NodeManagers run on worker nodes instead of TaskTracker daemons. The per-application ApplicationMaster is, in effect, a framework specific library and is tasked with negotiating resources from the ResourceManager and working with the NodeManager(s) to execute and monitor the tasks. For details of the new architecture, see Apache Hadoop NextGen MapReduce (YARN).

See also Selecting Appropriate JAR files for Your Jobs.

Sample Configuration:

mapred-site.xml:

<property>
 <name>mapreduce.framework.name</name>
 <value>yarn</value>
</property>

Configure the following services: ResourceManager (on a dedicated host) and NodeManager (on every host where you plan to run MapReduce v2 jobs).

The following table shows the most important properties that you must configure for your cluster in yarn-site.xml

yarn.nodemanager.aux-services

yarn.resourcemanager.hostname

yarn.resourcemanager.
address,
yarn.resourcemanager.
admin.address, 
yarn.resourcemanager.
scheduler.address,
yarn.resourcemanager.
resource-tracker.address,
yarn.resourcemanager.
webapp.address

yarn.application.classpath

yarn.log.aggregation-enable

Next, you need to specify, create, and assign the correct permissions to the local directories where you want the YARN daemons to store data.

You specify the directories by configuring the following two properties in the yarn-site.xml file on all cluster nodes:

Here is an example configuration:

yarn-site.xml:

  <property>
    <name>yarn.resourcemanager.hostname</name>
    <value>resourcemanager.company.com</value>
  </property>   
  <property>
    <description>Classpath for typical applications.</description>
    <name>yarn.application.classpath</name>
    <value>
        $HADOOP_CONF_DIR,
        $HADOOP_COMMON_HOME/*,$HADOOP_COMMON_HOME/lib/*,
        $HADOOP_HDFS_HOME/*,$HADOOP_HDFS_HOME/lib/*,
        $HADOOP_MAPRED_HOME/*,$HADOOP_MAPRED_HOME/lib/*,
        $HADOOP_YARN_HOME/*,$HADOOP_YARN_HOME/lib/*
    </value>
  </property>
  <property>
    <name>yarn.nodemanager.aux-services</name>
    <value>mapreduce_shuffle</value>
  </property>
  <property>
    <name>yarn.nodemanager.local-dirs</name>
    <value>file:///data/1/yarn/local,file:///data/2/yarn/local,file:///data/3/yarn/local</value>
  </property>
  <property>
    <name>yarn.nodemanager.log-dirs</name>
    <value>file:///data/1/yarn/logs,file:///data/2/yarn/logs,file:///data/3/yarn/logs</value>
  </property>
  <property>
    <name>yarn.log.aggregation-enable</name>
    <value>true</value> 
  </property>  
  <property>
    <description>Where to aggregate logs</description>
    <name>yarn.nodemanager.remote-app-log-dir</name>
    <value>hdfs://<namenode-host.company.com>:8020/var/log/hadoop-yarn/apps</value>
  </property>

After specifying these directories in the yarn-site.xml file, you must create the directories and assign the correct file permissions to them on each node in your cluster.

In the following instructions, local path examples are used to represent Hadoop parameters. Change the path examples to match your configuration.

To configure local storage directories for use by YARN:

1. Create the yarn.nodemanager.local-dirs local directories:

   $ sudo mkdir -p /data/1/yarn/local /data/2/yarn/local /data/3/yarn/local /data/4/yarn/local

2. Create the yarn.nodemanager.log-dirs local directories:

   $ sudo mkdir -p /data/1/yarn/logs /data/2/yarn/logs /data/3/yarn/logs /data/4/yarn/logs

3. Configure the owner of the yarn.nodemanager.local-dirs directory to be the yarn user:

   $ sudo chown -R yarn:yarn /data/1/yarn/local /data/2/yarn/local /data/3/yarn/local /data/4/yarn/local

4. Configure the owner of the yarn.nodemanager.log-dirs directory to be the yarn user:

   $ sudo chown -R yarn:yarn /data/1/yarn/logs /data/2/yarn/logs /data/3/yarn/logs /data/4/yarn/logs

Here is a summary of the correct owner and permissions of the local directories:

In addition, make sure proxying is enabled for the mapred user; configure the following properties in core-site.xml:

YARN requires a staging directory for temporary files created by running jobs. By default it creates /tmp/hadoop-yarn/staging with restrictive permissions that may prevent your users from running jobs. To forestall this, you should configure and create the staging directory yourself; in the example that follows we use /user:

1. Configure yarn.app.mapreduce.am.staging-dir in mapred-site.xml:

   <property>
       <name>yarn.app.mapreduce.am.staging-dir</name>
       <value>/user</value>
   </property>

2. Once HDFS is up and running, you will create this directory and a history subdirectory under it (see Step 8).

Alternatively, you can do the following:

1. Configure mapreduce.jobhistory.intermediate-done-dir and mapreduce.jobhistory.done-dir in mapred-site.xml.
2. Create these two directories.
3. Set permissions on mapreduce.jobhistory.intermediate-done-dir to 1777.
4. Set permissions on mapreduce.jobhistory.done-dir to 750.

If you configure mapreduce.jobhistory.intermediate-done-dir and mapreduce.jobhistory.done-dir as above, you can skip Step 8.

Deploy the configuration if you have not already done so.

Start HDFS if you have not already done so.

Create the /tmp directory if you have not already done so.

This is a subdirectory of the staging directory you configured in Step 4. In this example we're using /user/history. Create it and set permissions as follows:

sudo -u hdfs hadoop fs -mkdir -p /user/history
sudo -u hdfs hadoop fs -chmod -R 1777 /user/history
sudo -u hdfs hadoop fs -chown mapred:hadoop /user/history

Create the /var/log/hadoop-yarn directory and set ownership:

 
sudo -u hdfs hadoop fs -mkdir -p /var/log/hadoop-yarn
sudo -u hdfs hadoop fs -chown yarn:mapred /var/log/hadoop-yarn

$ sudo -u hdfs hadoop fs -ls -R /

You should see:

drwxrwxrwt   - hdfs supergroup          0 2012-04-19 14:31 /tmp
drwxr-xr-x   - hdfs supergroup          0 2012-05-31 10:26 /user
drwxrwxrwt   - mapred hadoop          0 2012-04-19 14:31 /user/history
drwxr-xr-x   - hdfs   supergroup        0 2012-05-31 15:31 /var
drwxr-xr-x   - hdfs   supergroup        0 2012-05-31 15:31 /var/log
drwxr-xr-x   - yarn   mapred            0 2012-05-31 15:31 /var/log/hadoop-yarn

To start YARN, start the ResourceManager and NodeManager services:

On the ResourceManager system:

$ sudo service hadoop-yarn-resourcemanager start

On each NodeManager system (typically the same ones where DataNode service runs):

$ sudo service hadoop-yarn-nodemanager start

To start the MapReduce JobHistory Server

On the MapReduce JobHistory Server system:

$ sudo service hadoop-mapreduce-historyserver start

Create a home directory for each MapReduce user. It is best to do this on the NameNode; for example:

$ sudo -u hdfs hadoop fs -mkdir  /user/<user>
$ sudo -u hdfs hadoop fs -chown <user> /user/<user>

where <user> is the Linux username of each user.

Alternatively, you can log in as each Linux user (or write a script to do so) and create the home directory as follows:

sudo -u hdfs hadoop fs -mkdir /user/$USER
sudo -u hdfs hadoop fs -chown $USER /user/$USER

See Configuring the Hadoop Daemons to Start at Boot Time.