Upgrade to Pro
— share decks privately, control downloads, hide ads and more …
Speaker Deck
Features
Speaker Deck
PRO
Sign in
Sign up for free
Search
Search
高度なコード
Search
Aaron Patterson
September 22, 2014
Technology
13
1.2k
高度なコード
Aaron Patterson
September 22, 2014
Tweet
Share
More Decks by Aaron Patterson
See All by Aaron Patterson
Speeding up Instance Variables in Ruby 3.3
tenderlove
2
350
[Feature #20425] Speeding up delegate methods
tenderlove
3
220
RailsConf 2023
tenderlove
29
1k
Don't @ Me! Faster Instance Variables with Object Shapes
tenderlove
1
420
RailsConf 2022 Keynote
tenderlove
2
510
Some Assembly Required
tenderlove
1
520
HexDevs 2021
tenderlove
1
400
Compacting GC for MRI
tenderlove
60
4.5k
But At What Cost?
tenderlove
9
14k
Other Decks in Technology
See All in Technology
現場で役立つAPIデザイン
nagix
33
12k
アジャイル開発とスクラム
araihara
0
170
データ資産をシームレスに伝達するためのイベント駆動型アーキテクチャ
kakehashi
PRO
2
510
Swiftの “private” を テストする / Testing Swift "private"
yutailang0119
0
130
7日間でハッキングをはじめる本をはじめてみませんか?_ITエンジニア本大賞2025
nomizone
2
1.8k
MC906491 を見据えた Microsoft Entra Connect アップグレード対応
tamaiyutaro
1
540
技術負債の「予兆検知」と「状況異変」のススメ / Technology Dept
i35_267
1
1.1k
Platform Engineeringは自由のめまい
nwiizo
4
2.1k
エンジニアが加速させるプロダクトディスカバリー 〜最速で価値ある機能を見つける方法〜 / product discovery accelerated by engineers
rince
2
190
人はなぜISUCONに夢中になるのか
kakehashi
PRO
6
1.6k
モノレポ開発のエラー、誰が見る?Datadog で実現する適切なトリアージとエスカレーション
biwashi
6
800
Culture Deck
optfit
0
410
Featured
See All Featured
Fantastic passwords and where to find them - at NoRuKo
philnash
51
3k
Design and Strategy: How to Deal with People Who Don’t "Get" Design
morganepeng
129
19k
No one is an island. Learnings from fostering a developers community.
thoeni
21
3.1k
Building Applications with DynamoDB
mza
93
6.2k
jQuery: Nuts, Bolts and Bling
dougneiner
63
7.6k
Dealing with People You Can't Stand - Big Design 2015
cassininazir
366
25k
Practical Tips for Bootstrapping Information Extraction Pipelines
honnibal
PRO
12
960
Easily Structure & Communicate Ideas using Wireframe
afnizarnur
193
16k
How to Create Impact in a Changing Tech Landscape [PerfNow 2023]
tammyeverts
49
2.3k
Mobile First: as difficult as doing things right
swwweet
223
9.3k
Docker and Python
trallard
44
3.3k
ReactJS: Keep Simple. Everything can be a component!
pedronauck
666
120k
Transcript
( ͡ ° ͜ ʖ ͡ °) .oO(Hi)
ຖ߃ྫ$ ͷ$ ຊޠೳྗࢼݧ
Thanks!!!
Aaron Patterson
Ϩου ϋτ
None
None
Ruby Core Rails Core
ί ϛ ο τ ͠ ͯ! ϙ Π ϯ τ
Β ͓ ͏ ʂ
Revert Commits Count Too!
More mistakes == more points!!!!
ΦΨϫ͞Μ
ΰϒͪΌΜ
νϡʔνϡʔ
None
Ϗοάσʔλ
ϝλϧʹ$ ۙ͘ͳΔͨΊʹ
Node.JS
None
Speeding up Rails 4.2
ߟ͑ࣄ
Rack
͏ऴΘΓ
Rack ָ͡Όͳ͍
Rack 2.0 ൃද͠ͳ͍* * Rack 2.0 ൃද͠·͢ɻ
the_metal http://github.com/tenderlove/the_metal
def call(env) end άϩʔόϧม R ack 1.x
ετϦʔϛϯά class Stream def each loop { yield "hello world\n"
} end end $ def call(env) [200, {}, Stream.new] end R ack 1.x
def call(request, response) end IO IO the_m etal
request IOͱ ϦΩΣετͷใ! Λ͍࣋ͬͯ·͢ɻ
responseIOͱ Ϩεϙϯεͷใ! Λ͍࣋ͬͯ·͢ɻ
શ࣮ྫ TheMetal.create_server(->(req, res) { res.write_head 200, 'Content-Type' => 'text/plain' res.write
"Hello World\n" res.finish }).listen 9292, '0.0.0.0'
Great for HTTP/1.1 Extendable to HTTP/ 2.0
RackΛͬͯ ָΛ͠·͠ΐ͏
ߟ͑ࣄ͚ͩɻ
ಡΜͰԼ͍͞ʂ http://github.com/tenderlove/the_metal
Speeding up Rails 4.2
Adequate$ Record
Performance$ Tools
benchmark/ips
benchmark/ips require 'benchmark/ips' require 'set' $ list = ('a'..'zzzz').to_a set
= Set.new list $ Benchmark.ips do |x| x.report("set access") { set.include? "foo" } $ x.report("ary access") { list.include? "foo" } end G EM
ग़ྗ Calculating ------------------------------------- set access 68622 i/100ms ary access 395
i/100ms ------------------------------------------------- set access 3047175.3 (±12.7%) i/s - 14959596 in 5.018692s ary access 3899.2 (±7.1%) i/s - 19750 in 5.096118s ߹ ܭ IPS
Set Include: 3047175.3 / sec
Array Include: 3899.2 / sec
IPS: Higher Is Better
Blackbox Testing
Ωϟογϡͷςετ cache1 = Cache1.new cache2 = Cache2.new $ cache1["x"] =
Object.new cache2["x"] = Object.new $ Benchmark.ips do |x| x.report("cache1") { cache1["x"] } x.report("cache2") { cache2["x"] } end
࣮ߦͷάϥϑ 10,000 ܁Γฦ࣌ؒ͢ (seconds) 0.01 0.1 1 10 100 ΩϟογϡɹαΠζ
10 elements 100 elements 1000 elements 100000 elements Ωϟογϡ 1 Ωϟογϡ 2
Cacheͷ࣮ class Cache1 def initialize @cache = {} end def
[] k; @cache[k]; end def []= k,v; @cache[k] = v; end end $ class Cache2 def initialize @cache = [] end def [] k; x, = @cache.assoc(k); x; end def []= k,v; @cache << [k, v]; end end ఆ ઢܗ
۩ମతͳྫ Routes
Route දͷେ͖͞ link_to ʹؔ͋Δʁ
route ΛೖΕΔ class MyTest routes = ActionDispatch::Routing::RouteSet.new routes.draw { resources(:articles)
N.times do |num| resources num.to_s.to_sym end } end 10, 100, 1000
Sec /100k calls 9.5 9.7 9.9 10.1 10.3 1 2
3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 link_to
URL ͷ͞ ࣮ߦʹ ؔ͋Γ·͔͢ʁ
͞ΛมԽ class MyTest routes = ActionDispatch::Routing::RouteSet.new link = N.times.map(&:to_s).join '/'
$ routes.draw { get "/#{link}/:id", :as => :article, :controller => :articles, :action => :show } end 10, 100, 1000
ඵ / 10ສճͷݺͼग़͠ 9 10.5 12 13.5 15 1 2
3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 link_to
Object Allocations
GC.stat()
߹ܭͷׂΓͯ GC.stat(:total_allocated_object)
۩ମతͳྫɿ Views
Request Benchmark task :allocated_objects do app = Ko1TestApp::Application.instance app.app do_test_task(app)
env = rackenv "/books/new" do_test_task(app, env.dup) before = GC.stat :total_allocated_object TEST_CNT.times { do_test_task(app, env.dup) } after = GC.stat :total_allocated_object puts (after - before) / TEST_CNT end "/books/new"
Request Benchmark task :allocated_objects do app = Ko1TestApp::Application.instance app.app do_test_task(app)
env = rackenv "/books/new" do_test_task(app, env.dup) before = GC.stat :total_allocated_object TEST_CNT.times { do_test_task(app, env.dup) } after = GC.stat :total_allocated_object puts (after - before) / TEST_CNT end
Request Benchmark task :allocated_objects do app = Ko1TestApp::Application.instance app.app do_test_task(app)
env = rackenv "/books/new" do_test_task(app, env.dup) before = GC.stat :total_allocated_object TEST_CNT.times { do_test_task(app, env.dup) } after = GC.stat :total_allocated_object puts (after - before) / TEST_CNT end
ςετͷ݁Ռ Request ͷΦϒδΣΫτஔ 2000 2150 2300 2450 2600 4-0-stable 4-1-stable
master 2000
Ͳ͏ͬͯ$ ӕΛ͔ͭ͘
ςετͷ݁Ռ Request ͷΦϒδΣΫτஔ 0 650 1300 1950 2600 4-0-stable 4-1-stable
master
4-0-stable ͔Β ~19%ݮΓ·ͨ͠
4-1-stable ͔Β ~14%ݮΓ·ͨ͠
allocation_tracer https://github.com/ko1/allocation_tracer
ྫ ObjectSpace::AllocationTracer.trace do 1000.times { ["foo", {}] } end $
ObjectSpace::AllocationTracer.allocated_count_table
Output {:T_NONE=>0, :T_OBJECT=>0, :T_CLASS=>0, :T_MODULE=>0, :T_FLOAT=>0, :T_STRING=>1000, :T_REGEXP=>0, :T_ARRAY=>1000, :T_HASH=>1000,
:T_ZOMBIE=>0}
Speeding up Helpers
Profile Request / Response
TOTAL (pct) SAMPLES (pct) FRAME 813 (9.5%) 813 (9.5%) ActiveSupport::SafeBuffer#initialize
699 (8.1%) 350 (4.1%) block in ActiveRecord::Read#read_attribute 486 (5.7%) 298 (3.5%) ActionController::UrlFor#url_options 670 (7.8%) 274 (3.2%) ActionDispatch::Journey::Format#evaluate 773 (9.0%) 253 (2.9%) ActionDispatch#parameterize_args 1172 (13.6%) 220 (2.6%) ActiveRecord::Persistence#instantiate 213 (2.5%) 213 (2.5%) block in SQLite3::Statement#each 208 (2.4%) 208 (2.4%) ActiveSupport::SafeBuffer#html_safe? 204 (2.4%) 204 (2.4%) ActionDispatch::UrlFor#routes_generation? 245 (2.9%) 191 (2.2%) block (2 levels) in Class#class_attribute
ActiveSupport::SafeBuffer #initialize
Ͳ͜Ͱ࡞ͬͨͷʁ
Tag Options def tag_option(key, value, escape) if value.is_a?(Array) value =
escape ? safe_join(value, " ") : value.join(" ") else value = escape ? ERB::Util.h(value) : value end %(#{key}="#{value}") end
HTML Sanitization in Rails
Ordinary String >> x = "foo" => "foo" >> x.class
=> String >> x.html_safe? => false
SafeBuffer >> x = "foo" => "foo" >> y =
x.html_safe => "foo" >> y.class => ActiveSupport::SafeBuffer >> y.html_safe? => true
ERB::Utils.h
ERB::Utils.h def html_escape(s) s = s.to_s if s.html_safe? s else
s.gsub(HTML_ESCAPE_REGEXP, HTML_ESCAPE).html_safe end end
Creates 2 Objects.
Tag Options def tag_option(key, value, escape) if value.is_a?(Array) value =
escape ? safe_join(value, " ") : value.join(" ") else value = escape ? ERB::Util.h(value) : value end %(#{key}="#{value}") end
String SafeBuffer String
Extract Method def unwrapped_html_escape(s) # :nodoc: s = s.to_s if
s.html_safe? s else s.gsub(HTML_ESCAPE_REGEXP, HTML_ESCAPE) end end $ def html_escape(s) unwrapped_html_escape(s).html_safe end
Update Callers def tag_option(key, value, escape) if value.is_a?(Array) value =
escape ? safe_join(value, " ") : value.join(" ") else value = escape ? ERB::Util.unwrapped_html_escape(value) : value end %(#{key}="#{value}") end
String String
~200 Allocations Per Request for /books/new
YMMV (Your Mileage (kilometers?) May Vary)
Speeding up Output
ERB Template <% books.each do |book| %> <tr><td> <%= book.name
%> </tr></td> <% end %>
Compiled Template @output_buffer = output_buffer || ActionView::OutputBuffer.new;@output_buffer.safe_append='<h1>Listing books</h1> $ <table>
<thead> <tr> <th>Name</th> <th colspan="3"></th> </tr> </thead> $ <tbody> '.freeze; @books.each do |book| @output_buffer.safe_append=' <tr> <td>'.freeze;@output_buffer.append=( book.name );@output_buffer.safe_append='</td> <td>'.freeze;@output_buffer.append=( link_to 'Show', book );@output_buffer.safe_append='</ td> </tr> '.freeze; end @output_buffer.safe_append=' </tbody> </table> $ <br>
Compiled Template @output_buffer = ActionView::OutputBuffer.new @output_buffer.safe_append=' <tr> <td>'.freeze @output_buffer.append=( book.name
) HTML Literal
safe_append= class OutputBuffer def safe_append=(value) return self if value.nil? super(value.to_s)
end end W hy?
Don’t accept nil
safe_append= class OutputBuffer def safe_append=(value) super(value.to_s) end end
safe_append= class OutputBuffer def safe_append=(value) super(value) end end
safe_append= class OutputBuffer def safe_append=(value) super end end
safe_append= class OutputBuffer end ߴͳίʔυʂ
Results
Allocations Per Request 0 275 550 825 1100 T_STRING T_ARRAY
T_HASH T_NODE T_DATA OTHER 4-0-stable 4-1-stable master
~19% reduction since 4-0-stable
~14% reduction since 4-1-stable
·ͱΊɻ
Eliminate Objects
Ұ൪͍ίʔυ ଘࡏ͠ͳ͍ίʔυ
Limit Types
Fewer Types = Less Code
Less Code = Faster Code
Measure, measure measure measure
MEASURE!
Rails 4.2 will be the fastest ever!
͋Γ͕ͱ͏ʂ