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
pt&Goroutines
Search
monochromegane
May 31, 2014
Technology
20
15k
pt&Goroutines
pt(the_platinum_searcher) を高速化するために Goroutines まわりで試したことを発表しました。
http://connpass.com/event/6370/
monochromegane
May 31, 2014
Tweet
Share
More Decks by monochromegane
See All by monochromegane
Go言語での実装を通して学ぶLLMファインチューニングの仕組み / fukuokago22-llm-peft
monochromegane
0
67
不確実性下における目的と手段の統合的探索に向けた連続腕バンディットの応用 / iot70_gp_rff_mab
monochromegane
2
130
なめらかなシステムと運用維持の終わらぬ未来 / dicomo2025_coherently_fittable_system
monochromegane
0
2.3k
ベクトル検索システムの気持ち
monochromegane
36
11k
Go言語での実装を通して学ぶ、高速なベクトル検索を支えるクラスタリング技術/fukuokago-kmeans
monochromegane
1
200
Go言語でターミナルフレンドリーなAIコマンド、afaを作った/fukuokago20_afa
monochromegane
2
280
多様かつ継続的に変化する環境に適応する情報システム/thesis-defense-presentation
monochromegane
1
1k
Online Nonstationary and Nonlinear Bandits with Recursive Weighted Gaussian Process
monochromegane
0
610
AIを前提とした体験の実現に向けて/toward_ai_based_experiences
monochromegane
2
1k
Other Decks in Technology
See All in Technology
[CV勉強会@関東 CVPR2025 読み会] MegaSaM: Accurate, Fast, and Robust Structure and Motion from Casual Dynamic Videos (Li+, CVPR2025)
abemii
0
190
マイクロモビリティシェアサービスを支える プラットフォームアーキテクチャ
grimoh
1
200
7月のガバクラ利用料が高かったので調べてみた
techniczna
3
290
ドキュメントはAIの味方!スタートアップのアジャイルを加速するADR
kawauso
3
340
LLMエージェント時代に適応した開発フロー
hiragram
1
410
LLM時代の検索とコンテキストエンジニアリング
shibuiwilliam
2
1.1k
Understanding Go GC #coefl_go_jp
bengo4com
0
1.1k
[OCI Skill Mapping] AWSユーザーのためのOCI(2025年8月20日開催)
oracle4engineer
PRO
2
140
Oracle Base Database Service:サービス概要のご紹介
oracle4engineer
PRO
2
20k
[CVPR2025論文読み会] Linguistics-aware Masked Image Modelingfor Self-supervised Scene Text Recognition
s_aiueo32
0
210
TypeScript入門
recruitengineers
PRO
10
1.8k
Preferred Networks (PFN) とLLM Post-Training チームの紹介 / 第4回 関東Kaggler会 スポンサーセッション
pfn
PRO
1
180
Featured
See All Featured
Making the Leap to Tech Lead
cromwellryan
134
9.5k
Building Flexible Design Systems
yeseniaperezcruz
328
39k
Rebuilding a faster, lazier Slack
samanthasiow
83
9.1k
CSS Pre-Processors: Stylus, Less & Sass
bermonpainter
358
30k
CoffeeScript is Beautiful & I Never Want to Write Plain JavaScript Again
sstephenson
161
15k
Scaling GitHub
holman
462
140k
Easily Structure & Communicate Ideas using Wireframe
afnizarnur
194
16k
Balancing Empowerment & Direction
lara
2
590
Exploring the Power of Turbo Streams & Action Cable | RailsConf2023
kevinliebholz
34
6k
Sharpening the Axe: The Primacy of Toolmaking
bcantrill
44
2.4k
Let's Do A Bunch of Simple Stuff to Make Websites Faster
chriscoyier
507
140k
Design and Strategy: How to Deal with People Who Don’t "Get" Design
morganepeng
131
19k
Transcript
pt &Goroutine - GoCon 2014 spring -
MIYAKE Yusuke (@monochromegane)
GMO Pepabo, Inc.
grep ͯ͠·͔͢ʁ
grep?
ack?
ag?
pt The Platinum Searcher
Written in Golang
Mac OSX Linux Windows
UTF-8 EUC-JP Shift-JIS
AND
fast ! ack go 6.24s user 1.06s system 99% cpu
7.304 total # ack ag go 0.88s user 1.39s system 221% cpu 1.027 total # ag pt go 1.09s user 1.01s system 235% cpu 0.892 total # pt
How?
Goroutine & Channel
͍ͬ͠ΐʹߴԽͯ͠Έ·͠ΐ͏
1. ϑΝΠϧΛݕࡧͯ͠(find) 2. จࣈྻΛݕࡧͯ͠(grep) 3. ݁ՌΛදࣔ͢Δ(print) ύλʔϯݕࡧͱ
Approach-0 ! ॱ൪ʹ
find
find grep
find grep print
// find find := find.Find{Option: self.Option} find.Do(self.Root) ! // grep
grep := grep.Grep{ Files: find.Files, // result Pattern: self.Pattern, Option: self.Option} grep.Do() ! // print print := print.Print{ Matches: grep.Matches, // result Pattern: self.Pattern, Option: self.Option} print.Do()
> the_simple_searcher go $GOROOT > /dev/null
0.79 seconds
Approach-1 ! ฒߦʹ
Goroutine
• GoݴޠͰฒߦॲཧΛ࣮ݱ͢Δ • εϨουɺίϧʔνϯͱҧ͏ • Concurrency(ฒߦ)ͱParallelism(ฒྻ) • ܰྔ • go
f()
find grep print go go go
Channel
• Goroutineؒͷϝοηʔδϯά • ͷૹड৴ • όοϑΝʹΑΔϒϩοΫ
find grep print go go go
$IBOFM find grep print $IBOFM go go go
$IBOFM find grep print $IBOFM go go go
// channel files := make(chan *string, self.Option.Cap) matches := make(chan
*grep.Match, self.Option.Cap) done := make(chan bool) ! // find find := find.Find{Files: files, Option: self.Option} go find.Do(self.Root) ! // grep grep := grep.Grep{ Files: files, Matches: matches, Pattern: self.Pattern, Option: self.Option} go grep.Do() ! // print print := print.Print{ Done: done, Matches: matches, Pattern: self.Pattern, Option: self.Option} go print.Do() ! <-done // block
walkFunc := func(path string, info os.FileInfo, err error) error {
if info.IsDir() { return nil } self.Files <- &path // send return nil } ! filepath.Walk(root, walkFunc) close(self.Files) // close
for file := range self.Files { // receive ( <-self.Files
) fh, err := os.Open(*file) if err != nil { panic(err) } ! f := bufio.NewReader(fh) ! var buf []byte var lineNum = 1 for { buf, _, err = f.ReadLine() if err != nil { break } line := string(buf) if strings.Contains(line, self.Pattern) { self.Matches <- &Match{*file, lineNum, line} // send } lineNum++ } fh.Close() } close(self.Matches) // close
for match := range self.Matches { // receive fmt.Printf("%s:%d:%s\n", match.Path,
match.Num, match.Match) } self.Done <- true // send
> the_simple_searcher go $GOROOT > /dev/null
0.79 -> 0.87 seconds
?
buffer
• Channelͷड༰ྔ • ch := make(chan ܕ, ༰ྔ) • ༰ྔ·Ͱड
• ༰ྔ͑Δͱૹ৴ଆडͪ • ड৴͢Δͱ༰ྔ͕ͻͱۭͭ͘ • ༰ྔ͕0ͷ߹ɺৗʹͭ
// channel with buffer files := make(chan *string, self.Option.Cap) matches
:= make(chan *grep.Match, self.Option.Cap) done := make(chan bool) // always wait
> the_simple_searcher go $GOROOT > /dev/null
0.79 -> 0.8 seconds
Approach-2 ! ͬͱฒߦʹ
$IBOFM find grep print $IBOFM go go go
$IBOFM find grep print $IBOFM go go go grep grep
grep
var wg sync.WaitGroup for file := range self.Files { wg.Add(1)
// goroutineͷىಈΛΠϯΫϦϝϯτ (தུ) go func(self *Grep, file *string) { defer wg.Done() // goroutine͕ྃͨ͠ΒىಈΛσΫϦϝϯτ for { ɹɹɹɹɹɹɹɹɹɹɹɹɹɹɹɹɹɹɹ(தུ) } fh.Close() ! }(self, file) // ΫϩʔδϟΛgoroutineʹ͢Δͱ͖มͷڞ༗ʹҙ ! } wg.Wait() // ෆಛఆͷgoroutine͕શͯऴྃ͢ΔͷΛͭ close(self.Matches)
> the_simple_searcher go $GOROOT > /dev/null
panic ! too many open files
var wg sync.WaitGroup sem := make(chan bool, self.Option.Cap) // ىಈ͢ΔgoroutineͷΛ੍ޚ͢Δchannel
for file := range self.Files { sem <- true // goroutineͷىಈ(channelͷbuffer)͕͍ͬͺ͍ͳΒͭ wg.Add(1) (தུ) go func(self *Grep, file *string) { defer wg.Done() for { ɹɹɹɹɹɹɹɹɹɹɹɹɹɹɹɹɹɹɹ(தུ) } fh.Close() <-sem // ಉ࣌ىಈchannelͷbufferʹۭ͖Λͭ͘Δ ! }(self, file) ! } wg.Wait() close(self.Matches)
> the_simple_searcher go $GOROOT > /dev/null
0.79 -> 0.8 seconds
Approach-3 ! ฒྻʹ
GOMAXPROCS
• Goroutineͷฒྻ • σϑΥϧτ1 • runtime.NumCPU()ͰίΞΛऔಘ • runtime.GOMAXPROCS()ͰฒྻΛઃఆ
> the_simple_searcher go $GOROOT > /dev/null
0.79 -> 0.55 ! seconds
benchmark ! • Mac OSX(10.9.3) • CPU: 2.5GHz Core i5(2Core)
• Memory: 8GB • Go: 1.2.2
#V⒎FS (0."9130$4 "QQSPBDI
ฒߦԽͯ͠ͳ͍ͷͰ ฒྻԽͯ͠มΘΒͣ ίΞҎ্ͷࢦఆ ޮՌͳ͠ ଌఆͯ͠ௐ͠ͳ͍ͱ ߹ʹΑͬͯ͘ͳΔ #V⒎FS (0."9130$4 "QQSPBDI
–Rob Pike • Concurrency is powerful. • Concurrency is not
parallelism. • Concurrency enables parallelism. • Concurrency makes parallelism (and scaling and everything else) easy.
એ ϖύϘͰΤϯδχΞΛืू͍ͯ͠·͢ɻ ڞʹαʔϏεΛੜΈग़͠ҭͯͯ͘ΕΔ৽͍ؒ͠ Λ͍ͬͯ·͢ɻ ! http://pepabo.com/recruit/career/engineer/
͓ΘΓ