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
Cheating Your Way to Webscale
Search
Matt Robenolt
May 02, 2014
Programming
13
1.4k
Cheating Your Way to Webscale
Python Nordeste, May 2nd 2014
Matt Robenolt
May 02, 2014
Tweet
Share
More Decks by Matt Robenolt
See All by Matt Robenolt
Everything is broken and I don't know why.
mattrobenolt
0
44
I am bad at my job.
mattrobenolt
0
220
Everything is broken, and I don't know why. Python edition.
mattrobenolt
1
210
Everything is broken, and I don't know why. Python edition.
mattrobenolt
2
600
Varnish: How We Do It
mattrobenolt
1
220
Everything is broken, and I don't know why.
mattrobenolt
7
1.5k
HTTP for Great Good
mattrobenolt
85
200k
Caching is Hard: Varnish @ Disqus
mattrobenolt
52
2.1M
Developing & Deploying "Large" Scale Web Applications
mattrobenolt
25
1.3k
Other Decks in Programming
See All in Programming
CSC307 Lecture 01
javiergs
PRO
0
670
LLMで複雑な検索条件アセットから脱却する!! 生成的検索インタフェースの設計論
po3rin
4
1.1k
Patterns of Patterns
denyspoltorak
0
450
CSC307 Lecture 02
javiergs
PRO
1
760
ELYZA_Findy AI Engineering Summit登壇資料_AIコーディング時代に「ちゃんと」やること_toB LLMプロダクト開発舞台裏_20251216
elyza
2
1k
LLM Observabilityによる 対話型音声AIアプリケーションの安定運用
gekko0114
2
260
SQL Server 2025 LT
odashinsuke
0
150
re:Invent 2025 のイケてるサービスを紹介する
maroon1st
0
170
TestingOsaka6_Ozono
o3
0
270
The Art of Re-Architecture - Droidcon India 2025
siddroid
0
160
QAフローを最適化し、品質水準を満たしながらリリースまでの期間を最短化する #RSGT2026
shibayu36
0
1.9k
從冷知識到漏洞,你不懂的 Web,駭客懂 - Huli @ WebConf Taiwan 2025
aszx87410
2
3.4k
Featured
See All Featured
Odyssey Design
rkendrick25
PRO
0
460
jQuery: Nuts, Bolts and Bling
dougneiner
65
8.4k
Designing Powerful Visuals for Engaging Learning
tmiket
0
200
Building Better People: How to give real-time feedback that sticks.
wjessup
370
20k
VelocityConf: Rendering Performance Case Studies
addyosmani
333
24k
Conquering PDFs: document understanding beyond plain text
inesmontani
PRO
4
2.2k
Dealing with People You Can't Stand - Big Design 2015
cassininazir
367
27k
A Modern Web Designer's Workflow
chriscoyier
698
190k
YesSQL, Process and Tooling at Scale
rocio
174
15k
The SEO identity crisis: Don't let AI make you average
varn
0
48
Why Our Code Smells
bkeepers
PRO
340
58k
Breaking role norms: Why Content Design is so much more than writing copy - Taylor Woolridge
uxyall
0
130
Transcript
Python Nordeste May 2nd 2014 Matt Robenolt Cheating Your Way
to #webscale
Hello < me irl
Lead Operations Engineer
Core Contributor
So what is #webscale?
10 million requests per second 4ms mean response time asynchronous
io mongodb
10 million requests per second 4ms mean response time asynchronous
io mongodb NOPE
Disqus only does 150 req/s per web server. * we
also write some bad code
150 12,960,000 388,800,000 per second per day per month real
world #webscale
Scale is about hiding the fact that your application is
actually really slow.
If your application feels fast, then it’s probably good enough.
Users hate waiting for shit.
So how do we do it?
Cheating 101
When a user asks for new data, let’s give them
old data instead.
When a user asks for new data, let’s give them
old data instead. Caching
When telling us to do something, let’s say we did
and maybe do it later.
When telling us to do something, let’s say we did
and maybe do it later. Queueing
Rule #1 Don’t get caught.
Rule #2 Don’t get caught.
Rule #3 Don’t get caught.
HTTP Caching
Introducing
tl;dr Varnish sits between your application and your users Internet
Let’s talk about HTTP. Hypertext Transport Protocol
$ curl -v disqus.com
> GET / HTTP/1.1 > User-Agent: curl/7.24.0 > Host: disqus.com
> Accept: */* > < HTTP/1.1 200 OK < Server: nginx < Date: Fri, 02 May 2014 06:38:37 GMT < Content-Type: text/html; charset=utf-8 < Content-Length: 10453 < Last-Modified: Fri, 30 Aug 2013 00:32:14 GMT < Vary: Accept-Encoding < Expires: Fri, 02 May 2014 06:43:36 GMT < Cache-Control: public, max-age=300
> GET / HTTP/1.1 > User-Agent: curl/7.24.0 > Host: disqus.com
> Accept: */* > < HTTP/1.1 200 OK < Server: nginx < Date: Fri, 02 May 2014 06:38:37 GMT < Content-Type: text/html; charset=utf-8 < Content-Length: 10453 < Last-Modified: Fri, 30 Aug 2013 00:32:14 GMT < Vary: Accept-Encoding < Expires: Fri, 02 May 2014 06:43:36 GMT < Cache-Control: public, max-age=300 Request
> GET / HTTP/1.1 > User-Agent: curl/7.24.0 > Host: disqus.com
> Accept: */* > < HTTP/1.1 200 OK < Server: nginx < Date: Fri, 02 May 2014 06:38:37 GMT < Content-Type: text/html; charset=utf-8 < Content-Length: 10453 < Last-Modified: Fri, 30 Aug 2013 00:32:14 GMT < Vary: Accept-Encoding < Expires: Fri, 02 May 2014 06:43:36 GMT < Cache-Control: public, max-age=300 Method
> GET / HTTP/1.1 > User-Agent: curl/7.24.0 > Host: disqus.com
> Accept: */* > < HTTP/1.1 200 OK < Server: nginx < Date: Fri, 02 May 2014 06:38:37 GMT < Content-Type: text/html; charset=utf-8 < Content-Length: 10453 < Last-Modified: Fri, 30 Aug 2013 00:32:14 GMT < Vary: Accept-Encoding < Expires: Fri, 02 May 2014 06:43:36 GMT < Cache-Control: public, max-age=300 Path
> GET / HTTP/1.1 > User-Agent: curl/7.24.0 > Host: disqus.com
> Accept: */* > < HTTP/1.1 200 OK < Server: nginx < Date: Fri, 02 May 2014 06:38:37 GMT < Content-Type: text/html; charset=utf-8 < Content-Length: 10453 < Last-Modified: Fri, 30 Aug 2013 00:32:14 GMT < Vary: Accept-Encoding < Expires: Fri, 02 May 2014 06:43:36 GMT < Cache-Control: public, max-age=300 Version
> GET / HTTP/1.1 > User-Agent: curl/7.24.0 > Host: disqus.com
> Accept: */* > < HTTP/1.1 200 OK < Server: nginx < Date: Fri, 02 May 2014 06:38:37 GMT < Content-Type: text/html; charset=utf-8 < Content-Length: 10453 < Last-Modified: Fri, 30 Aug 2013 00:32:14 GMT < Vary: Accept-Encoding < Expires: Fri, 02 May 2014 06:43:36 GMT < Cache-Control: public, max-age=300 Headers
> GET / HTTP/1.1 > User-Agent: curl/7.24.0 > Host: disqus.com
> Accept: */* > < HTTP/1.1 200 OK < Server: nginx < Date: Fri, 02 May 2014 06:38:37 GMT < Content-Type: text/html; charset=utf-8 < Content-Length: 10453 < Last-Modified: Fri, 30 Aug 2013 00:32:14 GMT < Vary: Accept-Encoding < Expires: Fri, 02 May 2014 06:43:36 GMT < Cache-Control: public, max-age=300 Response
> GET / HTTP/1.1 > User-Agent: curl/7.24.0 > Host: disqus.com
> Accept: */* > < HTTP/1.1 200 OK < Server: nginx < Date: Fri, 02 May 2014 06:38:37 GMT < Content-Type: text/html; charset=utf-8 < Content-Length: 10453 < Last-Modified: Fri, 30 Aug 2013 00:32:14 GMT < Vary: Accept-Encoding < Expires: Fri, 02 May 2014 06:43:36 GMT < Cache-Control: public, max-age=300 Status
> GET / HTTP/1.1 > User-Agent: curl/7.24.0 > Host: disqus.com
> Accept: */* > < HTTP/1.1 200 OK < Server: nginx < Date: Fri, 02 May 2014 06:38:37 GMT < Content-Type: text/html; charset=utf-8 < Content-Length: 10453 < Last-Modified: Fri, 30 Aug 2013 00:32:14 GMT < Vary: Accept-Encoding < Expires: Fri, 02 May 2014 06:43:36 GMT < Cache-Control: public, max-age=300 Headers
> GET / HTTP/1.1 > User-Agent: curl/7.24.0 > Host: disqus.com
> Accept: */* > < HTTP/1.1 200 OK < Server: nginx < Date: Fri, 02 May 2014 06:38:37 GMT < Content-Type: text/html; charset=utf-8 < Content-Length: 10453 < Last-Modified: Fri, 30 Aug 2013 00:32:14 GMT < Vary: Accept-Encoding < Expires: Fri, 02 May 2014 06:43:36 GMT < Cache-Control: public, max-age=300
> GET / HTTP/1.1 > User-Agent: curl/7.24.0 > Host: disqus.com
> Accept: */* > < HTTP/1.1 200 OK < Server: nginx < Date: Fri, 02 May 2014 06:38:37 GMT < Content-Type: text/html; charset=utf-8 < Content-Length: 10453 < Last-Modified: Fri, 30 Aug 2013 00:32:14 GMT < Vary: Accept-Encoding < Expires: Fri, 02 May 2014 06:43:36 GMT < Cache-Control: public, max-age=300 For 300 seconds, all users will get the same response without talking to our application.
> GET / HTTP/1.1 > User-Agent: curl/7.24.0 > Host: disqus.com
> Accept: */* > < HTTP/1.1 200 OK < Server: nginx < Date: Fri, 02 May 2014 06:38:37 GMT < Content-Type: text/html; charset=utf-8 < Content-Length: 10453 < Last-Modified: Fri, 30 Aug 2013 00:32:14 GMT < Vary: Accept-Encoding < Expires: Fri, 02 May 2014 06:43:36 GMT < Cache-Control: public, max-age=300 With power comes great responsibility.
GET / INTERNET Varnish Web servers
GET / INTERNET Varnish Web servers CACHED! “Cache-Control: max-age=300”
GET / INTERNET Varnish Web servers
GET / INTERNET Varnish Web servers CACHED!
BUT WAIT… THERE’S MORE
COLLAPSING REQUEST
GET / INTERNET Varnish Web servers
INTERNET Varnish Web servers GET /
INTERNET Varnish Web servers GET / If multiple users request
the same object, Varnish makes one fetch and returns to all users.
Queueing
Do as little work as possible, and return a promise
that this work will be done.
INTERNET Web servers Task workers Slow/Fast Data store Queue POST
/foo
INTERNET Web servers Task workers Slow/Fast Data store Queue POST
/foo
INTERNET Web servers Task workers Slow/Fast Data store Queue POST
/foo Workers can rate limit, debounce, increment counters, generate a fast materialized view, etc.
INTERNET Web servers Task workers Slow/Fast Data store Queue POST
/foo Make sure your tasks finish before a user tries to read the data back.
Final Thoughts
Understand your application. Where can you cheat without disrupting user
experience? Is seeing a few seconds old data going to damage a product?
Cheating should only enhance user experience.
Cheat any way you can, just don’t get caught.
Django & Varnish & RabbitMQ & Celery & PostgreSQL &
Redis & Cassandra & Riak Thanks
Questions? I have answers. ^ github.com/mattrobenolt @mattrobenolt some