Skip to content

Latest commit

 

History

History
124 lines (86 loc) · 4.14 KB

README.md

File metadata and controls

124 lines (86 loc) · 4.14 KB
Raw

elasticsearch*

(aka elasticsearch-star)

Starcluster plugin to install and configure an elasticsearch cluster.

Introduction

Starcluster is a command line tool for spinning up clusters of ec2 nodes quickly, great for some analytics / data chomping. A great introductory screencast is here: http://youtu.be/vC3lJcPq1FY

Elasticsearch is a distributed search engine with smart clustering capabilities: http://www.elasticsearch.org/videos/2010/02/07/es-introduction.html

The elasticsearch* plugin makes it dead easy to start a starcluster with elasticsearch on top in 2-3 minutes.

This can be great for testing all sorts of things like:

  • which instance size is optimal for my queries? (e.g. m1.xlarge or c1.xlarge)
  • what elasticsearch configuration parameters work best for my queries?
  • how does elasticsearch version x compare to version y?
  • how many shards should I have?
  • how much will it cost to meet performance requirement X?
  • training large scale machine learning on a spot instance priced elasticsearch cluster
  • and many more...

The elasticsearch cluster will be configured with the cloud-aws plugin to perform auto discovery.

Just add dataTM.

Installation

sudo pip install starcluster
git clone git://github.com/barnybug/elasticsearch-star.git
cp elasticsearch.py ~/.starcluster/plugins
cp config.example ~/.starcluster/config
emacs/vim ~/.starcluster/config # configure your EC2 keys and clusters
starcluster start -s 5 smallcluster # start a 5 node cluster

Once the cluster is provisioned, the http url of elasticsearch (the one on master) will be printed back to the console. This usually take about 2 minutes.

When you are done with the cluster:

starcluster terminate smallcluster

Configuration

A minimalist configuration is included in config.example - simply copy to ~/.starcluster/config and fill out your keys (as above).

If you already have an existing starcluster configuration, then you just need to:

Enable the plugin - add the following to your .starcluster/config:

PLUGINS = elasticsearchstar
PERMISSIONS = elasticsearch

to each [cluster ...] section you wish to use the plugin with.

At the end of config add:

[plugin elasticsearchstar]
SETUP_CLASS = elasticsearchstar.Elasticsearch

[permission elasticsearch]
PROTOCOL = tcp
FROM_PORT = 9200
TO_PORT = 9200
# CIDR_IP = <your_ip>/32
# add this in to secure elasticsearch access

In your cluster section as you'll not be using SGE (the Sun Grid Engine), you can set:

DISABLE_QUEUE=TRUE

Parameters:

The plugin can be configured by adding to the [plugin elasticsearchstar] section:

version = 0.90.0

(optional) Set the version of elasticsearch to use. Defaults to 0.90.1.

data_dir = /var/lib/elasticsearch

(optional) Set an alternative place to store data directory on the instance. Defaults to /mnt/elasticsearch - i.e. on ephemeral storage.

heap_size = 256

(optional) Set heap size in MB. The recommended amount of memory to allocate is about 50% of the instance memory. This will be calculated by default to match your instance size selected, but can be overridden with this parameter.

Reconfiguration

If you change any of these parameters, you can reconfigure elasticsearch on the cluster by running:

starcluster runplugin elasticsearchstar <clustername>

Elasticsearch Notes

  • Cluster name will match the starcluster name.
  • Each node will be named to match the starcluster allocated node name, eg master, node001, etc.
  • Elasticsearch will be allocated the suggested 50% of the instance memory, leaving the rest for disk buffers.
  • The data directory will be set to /mnt/elasticsearch by default - the ephemeral storage on an instance. This is plentiful in size, but comes with the usual caveats of being volatile, so stopping or terminating the instances will lose the data.
  • If you need your cluster data to survive stop/start, then change the data_dir parameter, you'll probably want to mount some EBS volumes too. But perhaps your use case is more intricate than transient throw-away clusters this plugin has in mind...