FLAME
-----
A super simple, distributed load tester. It acts to drive loads to a website and monitor the performance.
Think of it as a simple Apache Benchmark, distributed to mimic real world requests more accurately.
Flame consists of three components:
+ Flame Worker, which actually performs the job of pushing load to a server
+ Flame Dispatch, which dispatches load tasks to the workers
+ Flame Results, which collates results from the workers
To run Flame, you'll need a fee Amazon Web Services account, and access to the Simple Queue Service, SQS and SimpleDB.
Put your key and secret key in a file called 'config.yml', as show in the 'config.default' example.
+ Quick start
Whilst Flame is designed to work in a distributed cloud environment, you can just as easily run smaller tests from your laptop.
* You'll need the RightScale AWS gem:
sudo gem install right_aws
* Add your AWS credentials to the config.yml file.
key: 'EXAMPLE_KEY'
secret: 'EXAMPLE_SECRET'
* Then start a single Flame worker, and leave it running:
ruby flame.rb
* Start the Flame results monitor:
ruby flame_results.rb
* Then dispatch a simple load test for the worker to perform:
ruby flame_dispatch 1 50 http://your-ec2-instance-public-dns.amazonaws.com
This will drive a single worker to request the above URL 50 times. After a while, you should see output from the results monitor:
Result count: 50.0
Average load time: 806.381204 ms
+ Distributing load
Flame scales to many workers, and many levels of concurrency. Install Flame on an EC2 instance, and configure it to run at startup. Create an AMI, and instantiate the load testing fleet size of your choice. Spot instances are a good choice here.
* Dispatch load test to workers
Just as above, we can dispatch load tasks to our fleet of worker nodes. For example, if you have 50 instances each running 5 workers, you can start a full test with:
ruby flame_dispatch 250 500 http://your-ec2-instance-public-dns.amazonaws.com
This will start all 250 worker processes to drive 500 requests to the load test target.
* Collecting results
You can still run the result monitor locally to view the results, which are stored in SimpleDB. Customise the select statement for more fine grained statistics. Don't forget to empty the domains before re-running your load tests.
ruby flame_results.rb
While the test is running you can monitor your application and instance metrics with tools such as NewRelic and CloudWatch.
All times in Flame are in milliseconds.