Automated Installation on a Cluster

Pre-Requisites

1. Red Hat Enterprise Linux 6.x (64-bit) or CentOS equivalent.
2. You must have Ambari installed and thus transitively fulfill Ambari’s requirements.
3. Oracle Java JDK 1.8 (64-bit). Note that when installing Ambari you will be prompted to select a JDK. You can tell Ambari to download Oracle JDK 1.8 or point Ambari to an existing installation. Presto picks up the JDK Ambari was installed with so it is imperative that Ambari is running on Oracle JDK 1.8.
4. Disable requiretty. On RHEL 6.x this can be done by editing the /etc/sudoers file and commenting out Defaults    requiretty.
5. Install wget on all nodes that will run a Presto component.

Adding the Presto Service

This section and the following sections walk you through the integration steps needed to get Presto working with Ambari. By default, this integration code installs the latest Teradata Presto release.

To integrate the Presto service with Ambari:

1. Assuming HDP 2.3 was installed with Ambari, create the following directory on the node where the ambari-server is running:

   $ mkdir /var/lib/ambari-server/resources/stacks/HDP/2.3/services/PRESTO
   $ cd /var/lib/ambari-server/resources/stacks/HDP/2.3/services/PRESTO
   

2. Place the integration files within the newly created PRESTO directory by uploading the integration archive to your cluster and extracting it as follows:

   $ tar -xvf /path/to/integration/package/ambari-presto-0.1.0.tar.gz -C /var/lib/ambari-server/resources/stacks/HDP/2.3/services/PRESTO
   $ mv /var/lib/ambari-server/resources/stacks/HDP/2.3/services/PRESTO/ambari-presto-0.1.0/* /var/lib/ambari-server/resources/stacks/HDP/2.3/services/PRESTO
   $ rm -rf /var/lib/ambari-server/resources/stacks/HDP/2.3/services/PRESTO/ambari-presto-0.1.0
   

3. Make all integration files executable and restart the Ambari server:

   $ chmod -R +x /var/lib/ambari-server/resources/stacks/HDP/2.3/services/PRESTO/*
   $ ambari-server restart
   

4. After the server restarts, point your browser to it.

5. On the main Ambari Web UI page, click the Add Service button and follow the on screen wizard to add Presto.

   The following sections provide more details about the options and choices you need to make when adding Presto to the cluster.

Assigning Presto Processes to Nodes in a Cluster

Presto is composed of a coordinator and worker processes. The same code runs all nodes because the same Presto server RPM is installed for both workers and coordinator. The configuration on each node determines how a particular node behaves.

Presto can run in distributed mode or psuedo-distributed mode.

• In distributed mode, the Presto coordinator runs on one node and the Presto workers run on other nodes. Distributed mode is recommended.
• In pseudo-distributed mode, a single Presto process on one node acts as both coordinator and worker. Pseudo-distributed mode is not recommended for production nodes.

When you select a topology for Presto and finish the installation procedures as described below, you cannot modify that topology.

The client component of Presto is the presto-cli executable JAR. During or after installation, place the client component of Presto on any node all nodes where you expect to access the Presto server with this command line utility. The presto-cli executable JAR does not need to be co-located with either a worker or a coordinator. It can be installed on its own. The CLI can be found at /usr/lib/presto/bin/presto-cli.

After you add the Presto Service as described in the previous section, two screens on the main Ambari WebUI page allow you to assign the Presto processes among the nodes in your cluster.

1. Assign the Presto processes among the nodes in your cluster:
   1. Select a node for the Presto coordinator.
   2. Assign as many Presto workers to nodes as you need. Do not place a worker on the same node as a coordinator. Doing so causes the installation to fail because the integration software will attempt to install the RPM twice.
2. Place the client component on all nodes where you expect to access the Presto server with this command line utility.
3. If you would like to schedule work on the Presto coordinator, effectively turning the process into a dual worker/coordinator, enable the node-scheduler.include-coordinator toggle available in the configuration screen.

Configuring Presto

The one configuration property that does not have a default and requires input is discovery.uri. The expected value is http://<FQDN-of-node-hosting-coordinator>:8081. Note that it is http and not https and that the port is 8081. If you change the value of http-server.http.port, make sure you also change it in disovery.uri.

Some of the most popular properties are displayed in the Settings tab (open by default). In the Advanced tab, set custom properties by opening up the correct drop down and specifying a key and a value. Note that specifying a property that Presto does not recognize will cause the installation to finish with errors because some or all servers fail to start.

You can change the Presto configuration after installation by doing the following:

1. In Ambari, select the Presto service.
2. Select the Configs tab.
3. Change a configuration option.
4. Restart Presto for the changes to take effect.

If you are running a version of Ambari that is older than 2.1 (version less than 2.1), you must omit the memory suffix (GB) when setting the following memory-related configurations:

• query.max-memory-per-node
• query.max-memory

For these two properties, the memory suffix is automatically added by the integration software. For all other memory-related configurations that you add as custom properties, you must include the memory suffix when specifying the value.

{
    'hive': ['connector.name=hive-cdh4', 'hive.metastore.uri=thrift://example.net:9083'],
    'kafka': ['connector.name=kafka', 'kafka.table-names=table1,table2', 'kafka.nodes=host1:port,host2:port']
}