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
Dubious Database Design
Search
Andrew Godwin
September 07, 2015
Programming
0
260
Dubious Database Design
My talk from DjangoCon US 2015.
Andrew Godwin
September 07, 2015
Tweet
Share
More Decks by Andrew Godwin
See All by Andrew Godwin
Reconciling Everything
andrewgodwin
1
270
Django Through The Years
andrewgodwin
0
170
Writing Maintainable Software At Scale
andrewgodwin
0
400
A Newcomer's Guide To Airflow's Architecture
andrewgodwin
0
330
Async, Python, and the Future
andrewgodwin
2
620
How To Break Django: With Async
andrewgodwin
1
690
Taking Django's ORM Async
andrewgodwin
0
690
The Long Road To Asynchrony
andrewgodwin
0
620
The Scientist & The Engineer
andrewgodwin
1
720
Other Decks in Programming
See All in Programming
もう僕は OpenAPI を書きたくない
sgash708
6
1.9k
15分で学ぶDuckDBの可愛い使い方 DuckDBの最近の更新
notrogue
3
670
Better Code Design in PHP
afilina
0
180
ファインディLT_ポケモン対戦の定量的分析
fufufukakaka
0
940
.NET Frameworkでも汎用ホストが使いたい!
tomokusaba
0
200
CloudNativePGを布教したい
nnaka2992
0
120
Jasprが凄い話
hyshu
0
180
Rubyで始める関数型ドメインモデリング
shogo_tksk
0
140
GoとPHPのインターフェイスの違い
shimabox
2
210
読まないコードリーディング術
hisaju
0
110
良いコードレビューとは
danimal141
1
220
Swift Testingのモチベを上げたい
stoticdev
2
140
Featured
See All Featured
Site-Speed That Sticks
csswizardry
4
410
Why You Should Never Use an ORM
jnunemaker
PRO
55
9.2k
Imperfection Machines: The Place of Print at Facebook
scottboms
267
13k
How to train your dragon (web standard)
notwaldorf
91
5.9k
Statistics for Hackers
jakevdp
797
220k
Keith and Marios Guide to Fast Websites
keithpitt
411
22k
The Straight Up "How To Draw Better" Workshop
denniskardys
232
140k
[Rails World 2023 - Day 1 Closing Keynote] - The Magic of Rails
eileencodes
33
2.1k
The Invisible Side of Design
smashingmag
299
50k
The Pragmatic Product Professional
lauravandoore
32
6.4k
VelocityConf: Rendering Performance Case Studies
addyosmani
328
24k
Building Your Own Lightsaber
phodgson
104
6.2k
Transcript
DUBIOUS Database DESIGN
Andrew Godwin Hi, I'm Author of 1.7 Django & South
migrations Senior Software Engineer at Only hates MySQL a little
“Do this. Don't ask why.”
Learning from failure.
Spacelog 1
Spacelog 1
“Redis is fast!”
None
Spacelog 1
Read-only, forever.
“Redis is fast!”
GET chapter-1 GET chapter-2
GET entry-123 GET entry-124 GET entry-125
ZRANGEBYSCORE .... GET entry-123 GET entry-124 GET entry-125
Request page Look up key range Multi-get key range Get
speaker details
SELECT ... JOIN ... WHERE ...
Ignoring JOIN 2
“Joins are slow!”
{"id": 11, "post": "abc", "author": 1} {"id": 12, "post": "def",
"author": 2} {"id": 13, "post", "ghi", "author": 3} {"id": 1, "name": "Andrew"} {"id": 2, "name": "Brenda"} {"id": 3, "name": "Carol"}
n number of authors m × number of posts
scan all posts build dict of author -> posts scan
all authors and emit with posts
HASH JOIN
{ "id": 11, "post":"abc", "author": {"name": "Andrew"} }
{ "id": 11, "post":"abc", "author": { "name": "Andrew", "last_seen": 120993013,
} }
The server's running, it's fine! 3
Write new save file Write new save file Delete old
save file
Write new save file Write new save file ?
Tell payment processor to send Mark as processing Find unpaid
clients Mark as paid
Tell payment processor to send Mark as processing Find unpaid
clients Mark as paid
The Fastidious Modeller 4
TwitterUser FacebookUser LinkedInUser EmailUser
SELECT ... FROM TwitterUser SELECT ... FROM EmailUser SELECT ...
FROM LinkedInUser
None
None
The database isn't magic.
The Table Lover 5
“How do I make tables at runtime?”
None
Tables/columns per language Tables/columns per customer Configurable CMS columns
Columns per language 300 - 400 language variants x A
couple of translated cols per table x
DDL is very expensive.
Use JSON, hstore, or EAV-style table!
6 The Cold Boot
Decent cache hit rate Application servers mostly utilised
Great engineering!
What would happen if I deleted the entire cache?
None
The Optimist 7
Sharded PostgreSQL ElasticSearch Riak Redis Flat files + + +
+
Don't forget redundancy. And backups.
What happens if just one dies?
n services = n points of failure
The Primary Optimist 8
“The highest value PK is the most recent”
“Autoincrement will work and scale forever”
“IDs are numbers we can do maths on”
The Function Lover 9
"Why waste time fetching columns and rendering them separately?"
CREATE FUNCTION
CREATE FUNCTION ... import jinja2
bit.ly/whynotpg
None
SELECT render(template, id) FROM pages WHERE %s ~ url;
None
There's a reason behind every rule.
Ask why, or try yourself. Don't write it off without
context.
Thanks. Andrew Godwin @andrewgodwin