Middleware for profiling Rack-compatible apps using perftools.rb (github.com/tmm1/perftools.rb)
You'll need graphviz to generate call graphs using dot (for the GIF printer):
sudo port install graphviz # OS X brew install graphviz # Homebrew sudo apt-get install graphviz # Debian/Ubuntu
You'll need ps2pdf to generate PDFs (On OS X, ps2pdf comes is installed as part of Ghostscript)
sudo port install ghostscript # OSX brew install ghostscript # Homebrew sudo apt-get install ps2pdf # Debian/Ubuntu
Install the gem
gem install rack-perftools_profiler
Include the middleware
For Rails 2, add the following to config/environment.rb
config.gem 'rack-perftools_profiler', :version => '~> 0.1', :lib => 'rack/perftools_profiler' require 'rack/perftools_profiler' config.middleware.use ::Rack::PerftoolsProfiler, :default_printer => 'gif'
For Rails 3, add the following to your Gemfile
gem 'rack-perftools_profiler', '~> 0.1', :require => 'rack/perftools_profiler'
and add the following to config/application.rb
config.middleware.use ::Rack::PerftoolsProfiler, :default_printer => 'gif', :bundler => true
For Sinatra, call 'use' inside a configure block, like so:
configure :profiling do use ::Rack::PerftoolsProfiler, :default_printer => 'gif' end
For Rack::Builder, call 'use' inside the Builder constructor block
Rack::Builder.new do use ::Rack::PerftoolsProfiler, :default_printer => 'gif' end
:default_printer - can be set to 'text', 'gif', or 'pdf'. Default is :text
:mode - can be set to 'cputime', 'objects', 'walltime'. Default is :cputime. See the 'Profiling Modes' section below.
:frequency - in :cputime mode, the number of times per second the app will be sampled. Default is 100 (times/sec)
:bundler - run the profiler binary using 'bundle' if set to true. Default is false
:gemfile_dir - directory with Gemfile. Default is the current directory.
There are two ways to profile your app: with a single request or with multiple requests.
To profile with a single request, visit the URL you want to profile, but add the 'profile' and (optionally) the 'times' GET params (which will rerun the action the specified number of times).
Note that this will change the status, body, and headers of the response (you'll get back the profiling data, NOT the original response).
You can also profile your application using multiple requests. When you profile using this method, all responses are normal. You must visit __stop__ to complete profiling and then you can view the profiling data by visiting __data__.
curl http://localhost:3000/__start__ curl http://localhost:3000/foobar curl http://localhost:3000/foobaz curl http://localhost:3000/__stop__ curl http://localhost:3000/__data__
Profiling Data Options
Regardless of how you profile your application, you can add additional params to change how the data is displayed. When using a single request, these params are just added to the URL being profiled. When using multiple requests, they are added to the __data__ URL.
printer - overrides the default_printer option (see above)
ignore - a regular expression of the area of code to ignore
focus - a regular expression of the area of code to solely focus on.
(for 'ignore' and 'focus', please see google-perftools.googlecode.com/svn/trunk/doc/cpuprofile.html#pprof for more details)
perftools.rb (and therefore, the Rack middleware) can be put into three different profiling modes.
CPU time mode - Reports how many CPU cycles are spent in each section of code. This is the default and can be enabled by setting ':mode => :cputime'
Wall time mode - Reports the amount of time (as in, wall clock time) spent in each section of code. Enable by setting ':mode => :walltime'
Object allocation mode - Reports the percentage of object allocations performed in each section of code. Enable by setting ':mode => :objects'
For example, consider the following Sinatra application:
require 'sinatra' require 'rack/perftools_profiler' configure do use ::Rack::PerftoolsProfiler, :default_printer => 'gif', :mode => :cputime end get "/slow" do sleep(3) "hello" end
In the default mode, there will be no profiling data for the 'slow' route, because it uses few CPU cycles (You'll see the message 'No nodes to print').
If you change the mode to ':walltime', you'll get profiling data, since the call to 'sleep' causes the code to spend several seconds of wall time in the block.
Changing behavior with environment variables
The mode and frequency settings are enabled by setting environment variables. Some of these environment variables must be set before 'perftools' is required. If you only require 'rack/perftools_profiler', it will do the right thing (require 'perftools' after setting the environment variables).
If you need to require 'perftools' before 'rack/perftools_profiler' (or you have other problems changing the mode or frequency), try using these environment variables yourself.
Setting the frequency:
CPUPROFILE_FREQUENCY=500 ruby your_app.rb
Setting the mode to 'wall time'
CPUPROFILE_REALTIME=1 ruby your_app.rb
Setting the mode to 'object allocation'
CPUPROFILE_OBJECTS=1 ruby your_app.rb
A huge thanks to Aman Gupta for the awesome perftools.rb gem.
The basic idea and initial implementation of the middleware was heavily influenced by Rack::Profiler from rack-contrib.
Note on Patches/Pull Requests
Fork the project.
Make your feature addition or bug fix.
Add tests for it. This is important so I don't break it in a future version unintentionally.
Commit, do not mess with rakefile, version, or history. (if you want to have your own version, that is fine but bump version in a commit by itself I can ignore when I pull)
Send me a pull request. Bonus points for topic branches.
Copyright © 2010 Ben Brinckerhoff. See LICENSE for details.