This gem shaves off more than 30% of my Rails application page load time.
The primary GOAL of this project is to improve performance in the most heavily used areas of Ruby as path relation and file lookup is currently a huge bottleneck in performance. As this is the case the path performance updates will likely not be limited to just changing the Pathname class but also will be offering changes in related methods and classes.
Users will have the option to write their apps directly for this library, or they can choose to either refine or monkeypatch the existing standard library. Refinements are narrowed to scope and monkeypatching will be a sledge hammer 😉
I read a blog post about the new Sprockets 3.0 series being faster than the 2.0 series so I tried it out. It was not faster but rather it made my website take 31.8% longer to load. So I reverted back to the 2.0 series and I did a check on Rails on what methods were being called the most and where the application spends most of its time. It turns out roughly 80% (as far as I can tell) of the time spent and calls made are file Path handling. This is shocking, but it only gets worse when handling assets. That is why we need to deal with these load heavy methods in the most efficient manner!
Here’s a snippet of a Rails stack profile with some of the most used and time expensive methods.
Booting: development Endpoint: "/" user system total real 100 requests 26.830000 1.780000 28.610000 ( 28.866952) Running `stackprof tmp/2016-06-09T00:42:10-04:00-stackprof-cpu-myapp.dump`. Execute `stackprof --help` for more info ================================== Mode: cpu(1000) Samples: 7184 (0.03% miss rate) GC: 1013 (14.10%) ================================== TOTAL (pct) SAMPLES (pct) FRAME 1894 (26.4%) 1894 (26.4%) Pathname#chop_basename 1466 (20.4%) 305 (4.2%) Pathname#plus 1628 (22.7%) 162 (2.3%) Pathname#+ 234 (3.3%) 117 (1.6%) ActionView::PathResolver#find_template_paths 2454 (34.2%) 62 (0.9%) Pathname#join 57 (0.8%) 52 (0.7%) ActiveSupport::FileUpdateChecker#watched 760 (10.6%) 47 (0.7%) Pathname#relative? 131 (1.8%) 25 (0.3%) ActiveSupport::FileUpdateChecker#max_mtime 88 (1.2%) 21 (0.3%) Sprockets::Asset#dependency_fresh? 18 (0.3%) 18 (0.3%) ActionView::Helpers::AssetUrlHelper#compute_asset_extname 108 (1.5%) 14 (0.2%) ActionView::Helpers::AssetUrlHelper#asset_path
Here are some addtional stats. From Rails loading to my home page, these methods are called (not directly, Rails & gems call them) this many times. And the home page has minimal content.
Pathname#to_s called 29172 times. Pathname#<=> called 24963 times. Pathname#chop_basename called 24456 times Pathname#initialize called 23103 times. File#initialize called 23102 times. Pathname#absolute? called 4840 times. Pathname#+ called 4606 times. Pathname#plus called 4606 times. Pathname#join called 4600 times. Pathname#extname called 4291 times. Pathname#hash called 4207 times. Pathname#to_path called 2706 times. Pathname#directory? called 2396 times. Pathname#entries called 966 times. Dir#each called 966 times. Pathname#basename called 424 times. Pathname#prepend_prefix called 392 times. Pathname#cleanpath called 392 times. Pathname#cleanpath_aggressive called 392 times. Pathname#split called 161 times. Pathname#open called 153 times. Pathname#exist? called 152 times. Pathname#sub called 142 times.
After digging further I’ve found that Pathname is heavily used in Sprockets 2 but in Sprockets 3 they switched to calling Ruby’s faster methods from
Dir#each . It appears they’ve written all of the path handling on top of these themselves in Ruby. They achieved some performance gain by switching to rawer code methods, but then they lost more than that in performance by the many method calls built on top of that.
If you want to see the best results in Rails with this gem you will likely need to be using the Sprockets 2.0 series. Otherwise this library would need to rewrite Sprockets itself.
I’ve said this about Sprockets but this required two other gems to be updated as well. These are the gems and versions I upgraded and consider group 1 (Sprockets 2) and group 2 (Sprockets 3). My data is based on method calls rather than source code.
|Sprockets 2 Group||Sprockets 3 Group|
|sprockets 2.12.4||sprockets 3.6|
|sass 3.2.19||sass 5.0.4|
|bootstrap-sass 188.8.131.52||bootstrap-sass 3.3.6|
- Rust compilation is working
- Methods are most likely stable
- Testers and developers are most welcome!
Ensure Rust is installed:
curl -sSf https://static.rust-lang.org/rustup.sh | sudo sh -s -- --channel=nightly
Add this line to your application’s Gemfile:
gem 'faster_path', '~> 0.1.0'
And then execute:
Or install it yourself as:
$ gem install faster_path
MAC USERS:At the moment Mac users need to install the extension manualy. Go to the gem directory and run
cargo build --release . There is an issue opened for this and I’m looking for people who have Macs to help on this.
Current methods implemented:
|FasterPath Rust Implementation||Ruby 2.3.1 Implementation||Performance Improvement|
| || ||1234.6%|
| || ||31.3%|
| || ||66.0%|
| || ||1262.3%|
You may choose to use the methods directly, or scope change to rewrite behavior on the standard library with the included refinements, or even call a method to monkeypatch everything everywhere.
Pathname#chop_basename in Ruby STDLIB has a bug with blank strings, that is the only difference in behavior against FasterPath’s implementation.
For the scoped refinements you will need to
require "faster_path/optional/refinements" using FasterPath::RefinePathname
And for the sledgehammer of monkey patching you can do
require "faster_path/optional/monkeypatches" FasterPath.sledgehammer_everything!
Getting Started with Development
Methods will be written as exclusively in Rust as possible. Even just writing a not in Ruby with a Rust method like
!absolute? (not absolute) drops 39% of the performance already gained in Rust. Whenever feasible implement it in Rust.
After checking out the repo, make sure you have Rust installed. Then run
bundle && rake build_lib . Then, run
rake test to run the tests, and
rake bench for benchmarks.
Bug reports and pull requests are welcome on GitHub at https://github.com/danielpclark/faster_path .
MIT License or APACHE 2.0 at your pleasure.