isabella232 / ec2-fleet

A distributed load test framework using Amazon EC2 instances.

ec2-fleet

Originally inspired by Node.js 1 million connections test and branched from https://github.com/ashtuchkin/ec2-fleet

A distributed load test framework using Amazon EC2 instances and Siege.

This project makes it simple to launch and control lots of Amazon EC2 instances to run load tests using siege. Please use responsibly :)

The idea is that you setup an AMI image that will be used for load testing in each region you want to distribute your tests. The AMI should have all the tools setup, in this case nodejs, git and siege. take a look at cloud-config.sh for steps to install siege, node, git etc on aws images.

Originally it was written to make a Node.js 1 million connections test using 40 EC2 Micro instances.

Installation

# First, install Node.js (see http://nodejs.org)
# The installation is tested on Ubuntu 12.04.

git clone git://github.com/npr/ec2-fleet.git
cd ec2-fleet

# Install all needed modules.
npm install

# Install client side libs
bower install

#if you don't have bower, install bower and rerun bower install
npm install -g bower

Configuration

# Create configuration file
cp config/aws-config.json.sample config/aws-config.json

# Important! Update the instanceTags key/value pair set if you are running in multiple environments
# InstanceTags is arbitrary.  It really just identifies the instances that you will be using for load tests
# the keyValue pair under instanceTags needs to be unique to your test.
"instanceTags":{
    "yourKey":"yes"
}

# Edit accessKeyId, accessKeySecret in config/aws-config.json
# https://portal.aws.amazon.com/gp/aws/securityCredentials
# Also, choose regions where you wish your instances to be launched.
# Full list: ["ap-northeast-1", "ap-southeast-1", "eu-west-1", "sa-east-1", "us-east-1", "us-west-1", "us-west-2"]
# Important! In all these regions you should edit the amazon ec2 Security Group 'default' to open control port 8889 for TCP 0.0.0.0/0 

Usage

# Start 10 instances evenly distributed across the regions.
# All instances are tagged according to 'instanceTags' field in config/aws-config.json.
bin/aws.js start 10

Status

# A socket based version of the status page is available at http://localhost:300
node server.js

Go to http://localhost:3000

Running load tests

# Before trying to push out urls to the servers, wait until you see all instances that you expect in the status browser.
# Edit the urls.txt file with the urls that you would like to test.  
# This is a siege concept. One url per line
vim ./urls.txt

# and push it out to all servers
bin/aws.js urls ./urls.txt

# Start siege on all regions for 30 seconds with 100 concurrent connections
# After running this command you should see the siege status set to running on the status page
bin/aws.js siege 30 100

# Retrieve the last set of log results from all running servers
# Output is currently in tab delimited format. Easy enough to copy and paste in to excel for analysis
# Log output is also visible on the status page after each siege run
bin/aws.js logs

# Restart node process in all instances (recommended to do between tests).
bin/aws.js set restart 1

# After completing all tests, terminate all instances that we started.  Since we used instance tags, only those instances started
# using this script should be terminated.  Ensure that you are not using the exact same tag on other ec2 instances as a precaution.
bin/aws.js stop all

How does it work?

The framework uses a 'Cloud Init' feature of Ubuntu AWS images. When an instance is to be started, a vanilla image of Ubuntu is used (by default Ubuntu 12.04 64bit EBS), which runs a script given in file cloud-config.sh. The script installs Node.js, pulls down ec2-fleet-client and sets up an Upstart job to launch it. After this, the client.js starts listening on control port (8889 by default) and obeys given commands (see the source).

License: MIT