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
The Technology behind pixiv Infrastructure
Search
Harukasan
PRO
November 30, 2013
Technology
10
4k
The Technology behind pixiv Infrastructure
pixivのインフラを支える技術2013
at Python Developers Festa 2013.11
Harukasan
PRO
November 30, 2013
Tweet
Share
More Decks by Harukasan
See All by Harukasan
PicoRabbit: a Tiny Presentation Device Powered by Ruby
harukasan
PRO
2
380
pixivを支える技術 / 技育CAMPアカデミア
harukasan
PRO
3
500
20240401 新卒研修 - ピクシブにおける技術領域
harukasan
PRO
1
800
ピクシブのコンテンツ配信基盤技術 / pixiv TECH SALON
harukasan
PRO
5
5.6k
Goにおける画像ファイル処理 / golang.tokyo #19
harukasan
PRO
7
6.7k
WebRTC動画をトランスコードする / Transcoding video streams from WebRTC
harukasan
PRO
5
1.6k
ImageFluxを支えるリモート開発 / 20171202
harukasan
PRO
2
1.8k
YAPC::Fukuoka 前夜祭LT / Yet Another Pawoo Commit logs
harukasan
PRO
0
3k
YAPC::Fukuoka lunch session
harukasan
PRO
1
3.1k
Other Decks in Technology
See All in Technology
Amazon ECS & AWS Fargate 運用アーキテクチャ2025 / Amazon ECS and AWS Fargate Ops Architecture 2025
iselegant
16
5.6k
Oracle Cloud Infrastructure:2025年6月度サービス・アップデート
oracle4engineer
PRO
2
250
HiMoR: Monocular Deformable Gaussian Reconstruction with Hierarchical Motion Representation
spatial_ai_network
0
110
BrainPadプログラミングコンテスト記念LT会2025_社内イベント&問題解説
brainpadpr
1
170
AIエージェント最前線! Amazon Bedrock、Amazon Q、そしてMCPを使いこなそう
minorun365
PRO
15
5.2k
生成AIで小説を書くためにプロンプトの制約や原則について学ぶ / prompt-engineering-for-ai-fiction
nwiizo
4
2k
250627 関西Ruby会議08 前夜祭 RejectKaigi「DJ on Ruby Ver.0.1」
msykd
PRO
2
310
Fabric + Databricks 2025.6 の最新情報ピックアップ
ryomaru0825
1
140
AIのAIによるAIのための出力評価と改善
chocoyama
2
560
Amazon S3標準/ S3 Tables/S3 Express One Zoneを使ったログ分析
shigeruoda
4
530
Postman AI エージェントビルダー最新情報
nagix
0
110
Amazon Bedrockで実現する 新たな学習体験
kzkmaeda
2
580
Featured
See All Featured
CoffeeScript is Beautiful & I Never Want to Write Plain JavaScript Again
sstephenson
161
15k
Embracing the Ebb and Flow
colly
86
4.7k
Understanding Cognitive Biases in Performance Measurement
bluesmoon
29
1.8k
Performance Is Good for Brains [We Love Speed 2024]
tammyeverts
10
930
Typedesign – Prime Four
hannesfritz
42
2.7k
Writing Fast Ruby
sferik
628
61k
GitHub's CSS Performance
jonrohan
1031
460k
Fireside Chat
paigeccino
37
3.5k
Imperfection Machines: The Place of Print at Facebook
scottboms
267
13k
The Psychology of Web Performance [Beyond Tellerrand 2023]
tammyeverts
48
2.8k
Designing for humans not robots
tammielis
253
25k
Art, The Web, and Tiny UX
lynnandtonic
299
21k
Transcript
harukasan / Δ͔͞Μ at #pyfes 2013.11 ͷΠϯϑϥΛࢧ͑Δٕज़ 2013
Harukasan@pixiv a.k.a MICHII Shunsuke ಓҪढ़հ 2003 Kurume National College of
Tech. - NHK ROBOCON - ACM-ICPC 2008 Kyushu Inst. of Technology 2010 Tsukuba Univ. - Computational Vision Science 2012 pixiv Inc. - Infrastructure team
None
͜͜ʹpixivͷઆ໌͕ೖΔ
Server 400+ Traffic 10Gbps+ Team member 6 Monthly PV 3.7
Billion
Bases of pixiv Infrastructure Office IDCF DC Developments Testing Log
Analytics Small Services Main Applications DB Image Cluster New DC Image Cluster
ISP Backbone Office IDCF DC New DC 100M 1G 10G
1G 1G line 1G pixiv Network
ͷΠϯϑϥΛࢧ͑Δٕज़ 2013
http://www.slideshare.net/kamipo/pixiv @kamipo pixivͷΠϯϑϥΛࢧ͑Δٕज़ / ςΫηϛ2009
@cubicdaiya inside pixiv’s infrastructure / PHP Conference 2013 http://www.slideshare.net/cubicdaiya/inside-pixiv-infrastructure
Agenda pixiv Image Cluster Log Analysis Basis Management Tools
pixiv Image Cluster QJYJWͷը૾৴Ϋϥελʹ͍ͭͯ
pixiv Image Cluster • 2010͔Βӡ༻։࢝ • pixivͷϝΠϯίϯςϯπͰ͋ΔΠϥετΛ ߴʹॲཧ͢ΔͨΊʹ࠷దԽ • શτϥϑΟοΫͷ90%Ҏ্Λࡹ͍͍ͯΔ
Image Cluster nginx Front Cache DNS Round-Robin ATS Cache nginx
Dispatch nginx Front Cache nginx Front Cache nginx Front Cache ATS Cache ATS Cache ATS Cache Consistent Hashing nginx Dispatch nginx Dispatch nginx Dispatch Apache Origin nginx Thumbnail DNS Round-Robin i1.pixiv.net i2.pixiv.net SmallLight
Cache strategy • ϝϞϦͱσΟεΫͷ2ஈΩϟογϡߏ • τϥϑΟοΫ͕૿͑ΔʹͭΕͯ εΠονؒτϥϑΟοΫ͕ແࢹͰ͖ͳ͘ͳͬͨ • ωοτϫʔΫτϥϑΟοΫΛ͑ͭͭ Ωϟογϡ༰ྔͷ֬อ͢Δඞཁ
Cache strategy • ϝϞϦՁ֨ͷԼ • SSDͷՁ͕֨མ • ߴIOPSͷSSD͕ొ 2011 ¥40,000
2013 ¥20,000 256G READ WRITE PRICE ioDrive2 785GB MLC*1 215,000/230,000 IOPS I don’t know Intel 910 800GB 100,000/ 75,000 IOPS ¥400,000 SSD 256GBx3 RAID0 80,000/ 50,000 IOPS ¥60,000 *1: ioDrive2ެশ Intel 910ɺ SSD RAID0ʹ͍ͭͯfioʹΑΔଌఆ 16G ECC RDIMM ¥20,000
Cache strategy ound-Robin nginx Front Cache nginx Front Cache nginx
Front Cache ATS Cache ATS Cache ATS Cache Consistent Hashing nginx Dispatch nginx Dispatch nginx Dispatch Apache Origin nginx Thumbnail DNS Round-Robin i1.pixiv.net i2.pixiv.net 64GB Memory - nginx cache on tmpfs - cache hit rate: 50% - reduce network traffic 256GB SSD x3 RAID0 - Apache Traffic Server (standalone) - cache hit rate: 80-90% SSD SSD HDD Original & BIG Thumb. Small Thumbnails
Aggregate image domains • ը૾αʔόϢʔβʔIDϕʔεͰࢄ img01.pixiv.net - img1XX.pixiv.net • 1ϖʔδͰ40-60ճDNSϦΫΤετ͕ൃੜ
Ոఉ༻ϧʔλDNSղܾ͕Ͱ͖ͳ͘ͳΔ • શը૾ͷURIΛมߋͯ͠ରԠ OLD: http://img01.pixiv.net/img/****/*****.jpg NEW: http://i1.pixiv.net/img01/img/****/*****.jpg
New Image Store ৽͍͠ը૾ετϨʔδํࣜʹ͍ͭͯ
New image store • ࡞+IDϕʔεͷγʔέϯγϟϧͳURI • 1ॻ͖ࠐΜͩϑΝΠϧRead Only • ࠶ߘॲཧ࡞Λߋ৽
• ngx_lua/OpenRestyΛ༻͍ͨཧআ
Logical Delete ! Kyototycoon ngx_lua / nginx null 404 /img02.png
403 /img03.png 404 /img05.png 404 /img08.png 403 " " GET /img01.png GET /img03.png 404
Logical Delete local memcached = require "resty.memcached" local uri =
ngx.var.request_uri local memc = memcached:new() . . . local val, flags, err = memc:get(request_uri) if val and val ~= "200" then exit(tonumber(val)) end logical_delete.lua location / { access_by_lua_file logical_delete.lua; }
Log Analysis Basis QJYJWͷϩάղੳج൫ʹ͍ͭͯ
Log Analysis Basis PHP Application MySQL/neoagent Front server - Error
Log - Login Log - Activity Log - Slow Query - Access Log MongoDB Elasticsearch Fluentd File System
Error Viewer
Slow Query Viewer
Kibana 3
Output Log • JSONΛॻ͖ग़ͯ͠Fluentd͕tail͢ΔΈ • ϓϩηε͕େྔʹىಈͯ͠εϧʔϓοτ͕མͪͳ͍ PHP Exception Handler Logger::write($type,
$data) # JSON Fluentd in_tail
Fluentd config <source> type tail path /var/tmp/log/activity.log pos_file /var/tmp/fluentd/activity.pos tag
activity format json # JSONܗࣜΛಡΈࠐΉ </source> <match activity> type forward_with_hostname # HostΛೖΕͯforward flush_interval 1s # ඞͣ1ඵ ! buffer_type file # buffer typefile/࠶ىಈͯ͠ফ͑ͳ͍ buffer_path /var/tmp/fluentd/buffer/activity.*.buffer buffer_chunk_limit 2m # chunkαΠζখ͞Ί buffer_queue_limit 128 # Ͳͷ͘Β͍ফ͑ͪΌ͍͚ͳ͍͔ … </match>
Management tools ཧܥɺࢹܥπʔϧʹ͍ͭͯ
Monitoring servers/services • ͘͘͝͝ҰൠతͳࢹϓϩμΫτΛ͍ͬͯΔ Ϧιʔεάϥϑ ϗετ/αʔϏεࢹ ϗετϓϩηεࢹ εΫϦϓτ Munin Nagios
Monit Cron
Cluster Admin ϋʔυΣΞใ ϗετͷ༻్ ࢹঢ়ଶ
Capistrano/Subversion • /etc/ҎԼͷઃఆϑΝΠϧ͕ͦͷ··subversionͷ ཧԼͷσΟϨΫτϦʹ • ઃఆөcapistranoΛ༻ͯ͠શʹσϓϩΠ • ϗετҰཡAPIܦ༝ͰऔಘͰ͖ΔΑ͏ʹ $cap dns:update
$cap dns:check $cap dns:reload ex: update DNS Record
Management Tools • LVSཧը໘ • MySQLͷԆࢹ • αʔϏεͷϦιʔεϞχλϦϯά
Conclusion • pixivΛࢧ͑Δج൫γεςϜʹ͍ͭͯհ • ͍͍͢πʔϧΛ͕͍͍ࣗͨͪ͢Α͏ʹ • ઑͬͨ͜ͱͤͣɺແཧͤͣӡ༻Ͱ͖Δঢ়ଶʹ ͍ͬͯ͘