Skip to content

Preload and fork ruby interpreter for fast script startup

Notifications You must be signed in to change notification settings

candlerb/snailgun

Repository files navigation

Snailgun

Snailgun accelerates the startup of Ruby applications which require large numbers of libraries. It does this by preparing a Ruby process with your chosen libraries preloaded, and then forking that process whenever a new command-line Ruby interpreter is required.

Installation

sudo gem install snailgun

Or for the latest code, git clone git://github.com/candlerb/snailgun.git and put the bin directory into your PATH.

Case 1: standalone

# WITHOUT SNAILGUN
$ time ruby -rubygems -e 'require "active_support"' -e 'puts "".blank?'
true

real	0m2.123s
user	0m1.424s
sys 	0m0.168s

# WITH SNAILGUN
$ snailgun -rubygems -ractive_support
Snailgun starting on /home/brian/.snailgun/14781 - 'exit' to end
$ time fruby -e 'puts "".blank?'
true

real	0m0.064s
user	0m0.020s
sys 	0m0.004s

$ exit
logout
Snailgun ended
$ 

Case 2: Rails app

When using Rails or Merb, snailgun will start a process preloaded for the test environment only unless told otherwise.

You need to edit config/environments/test.rb and set config.cache_classes = false. This is so that your application classes are loaded each time you run a test, rather than being preloaded into the test environment.

Snailgun will take several seconds to be ready to process requests. Start with snailgun -v if you wish to be notified when it is ready.

$ rails testapp
$ cd testapp
$ vi config/environments/test.rb
... set config.cache_classes = false
$ snailgun -I test
Now entering subshell. Use 'exit' to terminate snailgun

$ time RAILS_ENV=test fruby script/runner 'puts 1+2'
3

real	0m0.169s
user	0m0.040s
sys 	0m0.008s

# To run your test suite
$ frake test       # or frake spec

Your preloaded process will remain around until you type exit to terminate it.

Note that any attempt by fruby or frake to perform an action in an environment other than 'test' will fail. See below for how to run multiple snailgun environments.

Merb support has been contributed (using MERB_ENV), but it is untested by me.

Case 3: Rails with multiple environments

After reading the warnings below, you may choose to start multiple snailgun processes each configured for a different environment, as follows:

$ snailgun --rails test,development

This gives the potential for faster startup of rake tasks which involve the development environment (such as migrations) and the console. The utility fconsole is provided for this.

However, beware that frake and fruby need to decide which of the preloaded environments to dispatch the command to. The safest way is to force the correct one explicitly:

RAILS_ENV=test frake test:units
RAILS_ENV=development fruby script/server
RAILS_ENV=test fruby script/runner 'puts "".blank?'

If you do not specify the environment, then a simple heuristic is used:

  • fruby always defaults to the 'development' environment.

  • frake honours any RAILS_ENV=xxx setting on the command line. If missing, frake defaults to the 'test' environment if no args are given or if an arg containing the word 'test' or 'spec' is given; otherwise it falls back to the 'development' environment.

WARNING: The decision as to which of the preloaded environments to use is made before actually running the command. If the wrong choice is made, it can lead to problems.

In the worst case, you may have a 'test'-type task, but find that it is wrongly dispatched to your 'development' environment - and possibly ends up blowing away your development database. This actually happened to me while developing snailgun. SO IF YOUR DEVELOPMENT DATABASE CONTAINS USEFUL DATA, KEEP IT BACKED UP.

If you run test files individually, it is especially critical that you set the correct environment. e.g.

RAILS_ENV=test fruby -Ilib -Itest test/unit/some_test.rb

Case 4: Rails with cucumber

Cucumber creates its own Rails environment called "cucumber", so you can setup snailgun like this:

$ snailgun --rails test,cucumber

Then use frake cucumber to exercise the features. frake selects the "cucumber" environment if run with "cucumber" as an argument.

NOTE: to make your model classes be loaded on each run you need to set config.cache_classes = false in config/environments/cucumber.rb. Cucumber will give a big warning saying that this is known to be a problem with transactional fixtures. I don't use transactional fixtures so this isn't a problem for me.

For a substantial performance boost, remove :lib=>false lines from config/environments/cucumber.rb so that cucumber, webrat, nokogiri etc are preloaded.

Smaller performance boosts can be had from further preloading. For example, cucumber makes use of some rspec libraries for diffing even if you're not using rspec, so you can preload those. Add something like this to the end of config/environments/cucumber.rb

begin
  require 'spec/expectations'
  require 'spec/runner/differs/default'
rescue LoadError
end
require 'test_help'
require 'test/unit/testresult'
require 'active_support/secure_random'
require 'active_support/time_with_zone'

autotest

There is some simple support for autotest (from the ZenTest package). Just type fautotest instead of autotest after snailgun has been started. This hasn't been tested for a while.

Bypassing rubygems

You can get noticeably faster startup if you don't use rubygems to invoke the programs. To do this, you can add the binary directory directly into the front of your PATH, e.g. for Ubuntu

PATH=/var/lib/gems/1.8/gems/snailgun-1.0.3/bin:$PATH

Alternatively, create a file called fruby somewhere early on in your PATH (e.g. under $HOME/bin), like this:

#!/usr/bin/env ruby
load '/path/to/the/real/fruby'

Repeat for frake etc.

Other bugs and limitations

Only works with Linux/BSD systems, due to use of passing open file descriptors across a socket.

Ctrl-C doesn't terminate frake processes.

fruby script/console doesn't give any speedup, because script/console uses exec to invoke irb. Use the supplied fconsole instead.

The environment is not currently passed across the socket to the ruby process. This means it's not usable as a fast CGI replacement.

In Rails, you need to beware that any changes to your config/environment* will not be reflected until you stop and restart snailgun.

Licence

This code is released under the same licence as Ruby itself.

Author

Brian Candler [email protected]

Credits:

About

Preload and fork ruby interpreter for fast script startup

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages