Closes #1443
7.2 KiB
title | weight |
---|---|
Ozone On Premise Installation | 20 |
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
- 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.
- Storage Container Manager - Acts as the block manager. Ozone Manager requests blocks from SCM, to which clients can write data.
- 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 calledozone-site.xml
at the specified path (directory).
{{< highlight bash >}} ozone genconf {{< /highlight >}}
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.enabled This is the most critical setting for ozone.
Ozone is a work in progress and users have to enable this service explicitly.
By default, Ozone is disabled. Setting this flag to
true
enables ozone in the HDFS or Ozone cluster.
Here is an example,
{{< highlight xml >}} ozone.enabled true {{< /highlight >}}
-
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,
{{< highlight xml >}} ozone.metadata.dirs /data/disk1/meta {{< /highlight >}}
- 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,
{{< highlight xml >}} ozone.scm.names scm.hadoop.apache.org {{< /highlight >}}
- 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, {{< highlight xml >}} ozone.scm.datanode.id.dir /data/disk1/meta/node {{< /highlight >}}
- ozone.om.address OM server address. This is used by OzoneClient and Ozone File System.
Here is an example, {{< highlight xml >}} ozone.om.address ozonemanager.hadoop.apache.org {{< /highlight >}}
Ozone Settings Summary
Setting | Value | Comment |
---|---|---|
ozone.enabled | true | This enables SCM and containers in HDFS cluster. |
ozone.metadata.dirs | file path | The metadata will be stored here. |
ozone.scm.names | SCM server name | Hostname:port or IP:port address of SCM. |
ozone.scm.block.client.address | SCM server name and port | Used by services like OM |
ozone.scm.client.address | SCM server name and port | Used by client-side |
ozone.scm.datanode.address | SCM server name and port | Used by datanode to talk to SCM |
ozone.om.address | OM server name | Used 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.
{{< highlight bash >}}
ozone scm --init
{{< /highlight >}}
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.
{{< highlight bash >}}
ozone --daemon start scm
{{< /highlight >}}
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.
{{< highlight bash >}} ozone om --init {{< /highlight >}}
Once Ozone manager is initialized, we are ready to run the name service.
{{< highlight bash >}} ozone --daemon start om {{< /highlight >}}
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. {{< highlight bash >}} ozone --daemon start datanode {{< /highlight >}}
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 {{< highlight bash >}} ozone scm --init ozone om --init start-ozone.sh {{< /highlight >}}
This assumes that you have set up the slaves 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.