Ozone On Premise Installation

If you are feeling adventurous, you can setup ozone in a real cluster. Setting up a real cluster requires us to understand the components of Ozone. Ozone is designed to work concurrently with HDFS. However, Ozone is also capable of running independently. The components of ozone are the same in both approaches.

Ozone Components

  1. Ozone Manager - Is the server that is in charge of the namespace of Ozone. Ozone Manager is responsible for all volume, bucket and key operations.
  2. Storage Container Manager - Acts as the block manager. Ozone Manager requests blocks from SCM, to which clients can write data.
  3. Datanodes - Ozone data node code runs inside the HDFS datanode or in the independent deployment case runs an ozone datanode daemon.

Setting up an Ozone only cluster

  • Please untar the ozone-<version> to the directory where you are going to run Ozone from. We need Ozone jars on all machines in the cluster. So you need to do this on all machines in the cluster.

  • Ozone relies on a configuration file called ozone-site.xml. To generate a template that you can replace with proper values, please run the following command. This will generate a template called ozone-site.xml at the specified path (directory).

  1. ozone genconf <path>

Let us look at the settings inside the generated file (ozone-site.xml) and how they control ozone. Once the right values are defined, this file needs to be copied to ozone directory/etc/hadoop.

  • ozone.metadata.dirs Allows Administrators to specify where the metadata must reside. Usually you pick your fastest disk (SSD if you have them on your nodes). OzoneManager, SCM and datanode will write the metadata to this path. This is a required setting, if this is missing Ozone will fail to come up.

    Here is an example,

  1. <property>
  2. <name>ozone.metadata.dirs</name>
  3. <value>/data/disk1/meta</value>
  4. </property>
  • ozone.scm.names Storage container manager(SCM) is a distributed block service which is used by ozone. This property allows data nodes to discover SCM’s address. Data nodes send heartbeat to SCM. Until HA feature is complete, we configure ozone.scm.names to be a single machine.

Here is an example,

  1. <property>
  2. <name>ozone.scm.names</name>
  3. <value>scm.hadoop.apache.org</value>
  4. </property>
  • ozone.scm.datanode.id.dir Data nodes generate a Unique ID called Datanode ID. This identity is written to the file datanode.id in a directory specified by this path. Data nodes will create this path if it doesn’t exist already.

Here is an example,

  1. <property>
  2. <name>ozone.scm.datanode.id.dir</name>
  3. <value>/data/disk1/meta/node</value>
  4. </property>
  • ozone.om.address OM server address. This is used by OzoneClient and Ozone File System.

Here is an example,

  1. <property>
  2. <name>ozone.om.address</name>
  3. <value>ozonemanager.hadoop.apache.org</value>
  4. </property>

Ozone Settings Summary

SettingValueComment
ozone.metadata.dirsfile pathThe metadata will be stored here.
ozone.scm.namesSCM server nameHostname:port or IP:port address of SCM.
ozone.scm.block.client.addressSCM server name and portUsed by services like OM
ozone.scm.client.addressSCM server name and portUsed by client-side
ozone.scm.datanode.addressSCM server name and portUsed by datanode to talk to SCM
ozone.om.addressOM server nameUsed by Ozone handler and Ozone file system.

Startup the cluster

Before we boot up the Ozone cluster, we need to initialize both SCM and Ozone Manager.

  1. ozone scm --init

This allows SCM to create the cluster Identity and initialize its state. The init command is similar to Namenode format. Init command is executed only once, that allows SCM to create all the required on-disk structures to work correctly.

  1. ozone --daemon start scm

Once we know SCM is up and running, we can create an Object Store for our use. This is done by running the following command.

  1. ozone om --init

Once Ozone manager is initialized, we are ready to run the name service.

  1. ozone --daemon start om

At this point Ozone’s name services, the Ozone manager, and the block service SCM is both running.
Please note: If SCM is not running om --init command will fail. SCM start will fail if on-disk data structures are missing. So please make sure you have done both scm --init and om --init commands.

Now we need to start the data nodes. Please run the following command on each datanode.

  1. ozone --daemon start datanode

At this point SCM, Ozone Manager and data nodes are up and running.

Congratulations!, You have set up a functional ozone cluster.

Shortcut

If you want to make your life simpler, you can just run

  1. ozone scm --init
  2. ozone om --init
  3. start-ozone.sh

This assumes that you have set up the workers file correctly and ssh configuration that allows ssh-ing to all data nodes. This is the same as the HDFS configuration, so please refer to HDFS documentation on how to set this up.

Next >>