Creating a Presto Cluster

[posts] March 25, 2016

Creating a Presto Cluster

I first came across Presto when researching data virtualization - the idea that all of your data can be integrated regardless of its format or storage location. One can use scripts or periodic jobs to mashup data or create regular reports from several independent sources. However, these methods don’t scale well, especially when the queries change frequently or the data is ingested in realtime. Presto allows one to query a variety of data sources using SQL and presents the data in a standard table format, where it can be manipulated and JOINed like traditional relational data.

In Facebook’s words:

Presto is an open source distributed SQL query engine for running interactive analytic queries against data sources of all sizes ranging from gigabytes to petabytes.

There is a guide for how to download and setup Presto here. Additionally, there is a well documented list of connectors for querying a wide variety of data sources (including Redis and Kafka!). However, the documentation on how to setup a cluster (coordinator and workers) is a little sparse.

This guide assumes you have machines runnning Java 8 that can interact with each other over port 8080. Some of the documentation for installation from Facebook is repeated for the sake of having all the instructions in one place. Additionally, you’ll need to have data sources to which Presto can connect. See the full list on the connectors page.

Install Presto

Download Presto. These instructions are for version 0.142. You can find the most recent version and more deployment information here.

$ wget https://repo1.maven.org/maven2/com/facebook/presto/presto-server/0.142/presto-server-0.142.tar.gz
$ tar zxvf presto-server-0.142.tar.gz

Download the CLI for the coordinator and name it presto

$ wget https://repo1.maven.org/maven2/com/facebook/presto/presto-cli/0.142/presto-cli-0.142-executable.jar
$ mv presto-cli-0.142-executable.jar presto
$ chmod +x presto

Create configure files

Create etc folder in presto-server-0.142 directory Create config.properties, jvm.config, log.properties, and node.properties files.

Install uuid to generate a node.id

$ sudo apt-get install uuid
$ uuid
7fd6390a-f2dc-11e5-a834-0eb3775013cf

node.properties

node.environment=production
node.id=<generated uuid>
node.data-dir=/home/ubuntu/data

jvm.config

-server
-Xmx16G
-XX:+UseG1GC
-XX:G1HeapRegionSize=32M
-XX:+UseGCOverheadLimit
-XX:+ExplicitGCInvokesConcurrent
-XX:+HeapDumpOnOutOfMemoryError
-XX:OnOutOfMemoryError=kill -9 %p

This is the configuration for the coordinator:

Coordinator config.properties

coordinator=true
node-scheduler.include-coordinator=false
http-server.http.port=8080
query.max-memory=50GB
query.max-memory-per-node=2GB
discovery-server.enabled=true
discovery.uri=<coordinator_ip>:8080

The above files can be configured as needed for your specific cluster. The options node-scheduler.include-coordinator=false and coordinator=true indicate that the node is the coordinator and tells the coordinator not to do any of the computation work itself and to use the workers. After configuring the coordinator, you need to configure workers to add some resources to the cluster. If you were to try and run a query now, you would see something like:

Query 20160324_223423_00000_pdnyh failed: No worker nodes available

For the workers, you need to create all the same files as above, but the config.properties file needs to look similar to the one below.

Worker config.properties

coordinator=false
http-server.http.port=8080
query.max-memory=50GB
query.max-memory-per-node=2GB
discovery.uri=<coordinator_ip>:8080

Next we will create the catalog folder which tells Presto how to connect to various data sources. This folder goes inside the etc folder on all nodes of the cluster including the coordinator. An example folder structure might look like:

etc/
├── catalog
│   ├── cassandra.properties
│   ├── hive.properties
│   ├── kafka.properties
│   ├── mysql.properties
│   ├── postgresql.properties
│   └── redshift.properties
├── config.properties
├── jvm.config
├── log.properties
└── node.properties

Consult the connectors page on how to write theses properties files. It may help to version control this folder to make it more easily distributable.

Start Presto

Now that we’ve set up our coordinator and worker node(s), we can start the cluster. First SSH into and start the coordinator node:

$ ./presto-server-0.142/bin/launcher.py start
Running as 4292

SSH into and start the worker node(s):

$ ./presto-server-0.142/bin/launcher.py start
Running as 3000

Once you start the workers, you can use the Presto CLI on the coordinator to ensure the workers are connected:

$ ./presto
presto> select * from system.runtime.nodes;
                node_id                |        http_uri        | node_version |
--------------------------------------+------------------------+--------------+-
    ffb969e8-f049-11e5-a8dd-0e144badbcb1 | http://10.0.0.216:8080 | 0.142        |
    fc991558-ec5a-11e5-a15a-0eb3775013cf | http://10.0.0.70:8080  | 0.142        |
    46001b10-fdcc-11e5-9797-0e75f1fc6277 | http://10.0.0.15:8080  | 0.142        |
    c2334f0c-f04f-11e5-9e77-0e144badbcb1 | http://10.0.0.211:8080 | 0.142        |

Now you can use the Presto CLI on the coordinator to query data sources in the catalog using the Presto workers.