1 unstable release
Uses old Rust 2015
0.0.2 | Mar 3, 2017 |
---|
#596 in Filesystem
5.5MB
153K
SLoC
FasterPath
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 ;-)
Why
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 File#initialize
and 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 3.3.4.1 | bootstrap-sass 3.3.6 |
Status
- Rust compilation is working
- Methods are most likely stable
- Testers and developers are most welcome!
Installation
Ensure Rust is installed:
curl -sSf https://static.rust-lang.org/rustup.sh | sh
Add this line to your application's Gemfile:
gem 'faster_path', '~> 0.1.0'
And then execute:
$ bundle
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.
Visual Benchmarks
Benchmarks in Faster Path now produce visual graph charts of performance improvements.
When you run rake bench
the graph art will be placed in doc/graph/
. Here's the performance
improvement result for the chop_basename
method.
Usage
Current methods implemented:
FasterPath Rust Implementation | Ruby 2.3.1 Implementation | Performance Improvement |
---|---|---|
FasterPath.absolute? |
Pathname#absolute? |
93.9% |
FasterPath.chop_basename |
Pathname#chop_basename |
50.6% |
FasterPath.relative? |
Pathname#relative? |
93.2% |
FasterPath.blank? |
||
FasterPath.directory? |
Pathname#directory? |
25.5% |
FasterPath.add_trailing_separator |
Pathname#add_trailing_separator |
46.8% |
FasterPath.has_trailing_separator? |
Pathname#has_trailing_separator |
61.2% |
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.
Note: 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!
Unstable optional bits
Optional methods which have regressions. are unstable. These will not be included by default in monkey-patches. Be cautious when using the FasterPath::RefineFile
refinement. To try them anyways use the environment flag of WITH_REGRESSION
. These methods are here to be improved upon.
FasterPath Implementation | Ruby Implementation |
---|---|
FasterPath.dirname |
File.dirname |
FasterPath.basename |
File.basename |
FasterPath.extname |
File.extname |
It's been my observation (and some others) that the Rust implementation of the C code for File
has similar results but
performance seems to vary based on CPU cache on possibly 64bit/32bit system environmnets.
Developers for FasterPath: Most of these need to be rewritten, please use WITH_REGRESSION=true TEST_MONKEYPATCHES=true
in your testing. You can see the resulting failures currently on TravisCI under "Allow Failures".
Getting Started with Development
The primary methods to target are mostly listed in the Why section above. You may find the Ruby source code useful for Pathname's Ruby source, C source, tests, and checkout the documentation.
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
.
Run rake test
to run the tests, and rake bench
for benchmarks.
Learn and share performance tips on the wiki!
Building and running tests
First, bundle the gem's development dependencies by running bundle
.
Then, build the rust code:
rake build_src
FasterPath is tested with The Ruby Spec Suite to ensure it is compatible with the native implementation, and also has its own test suite testing its monkey-patching and refinements functionality.
To run all the tests at once, simply run rake
.
To run all the ruby spec tests, run mspec
.
To run an individual test or benchmark from FasterPath's own suite:
# An individual test file:
ruby -I lib:test test/benches/absolute_benchmark.rb
# All tests:
rake minitest
To run an individual ruby spec test, run mspec
with a path relative to spec/ruby_spec
, e.g.:
# A path to a file or a directory:
mspec core/file/basename_spec.rb
# Tests most relevant to FasterPath:
mspec core/file library/pathname
# All tests:
mspec
Contributing
Bug reports and pull requests are welcome on GitHub at https://github.com/danielpclark/faster_path.
License
MIT License or APACHE 2.0 at your pleasure.