We set out to answer this problem by combining the ruby gem from http://blitz.io
Heroku::Client which allows you to change the number of dynos for
your app instantly.
The syntax is pretty simple
$ ./blitzer.rb Usage: blitzer user pass heroku-app [target] user - Your heroku account name (email) pass - Heroku API key (from the accounts page) app - The name of your Heroku app target - How many users do you want to get to? default = 1000
The app is just the name of your Heroku app which resolves to http://.heroku.com. You can set the target concurrency to a specific value, with the default being 1,000 users. When you run this, we keep adding dynos (up to 32 dynos) until the percentage of errors during the load test is greater than 2%.
The output from running the blitzer looks something like this:
Blitzing http://dyno-capacity.heroku.com with a target of 1000 users starting 1 dyno(s) .done rushing ..322 hits | 0 errors | 1 timeouts @ 167 users starting 2 dyno(s) ....done rushing .......6259 hits | 1 errors | 0 timeouts @ 670 users starting 4 dyno(s) ...done rushing .......4784 hits | 1 errors | 0 timeouts @ 591 users starting 8 dyno(s) .......done rushing .......4781 hits | 1 errors | 0 timeouts @ 589 users starting 16 dyno(s) ...done rushing .............done >> You need 16 dyno(s) to handle 1,000 users
Integrating load and performance testing into your continuous integration process doesn't get simpler than this!