Redis 101
A whirlwind tour of the next big
thing in NoSQL data storage
PETER COOPER
h t t p : / / tw i t te r.c o m / p e te rc
h t t p : / / c o d e r. i o /
Whirlwind tour?
No overbearing detail.
A quick whizz-through.
Enough to get you excited
(if Redis is for you.)
Official docs, etc, are an awesome way to
continue.
Redis is... NoSQL
an “advanced key-value
store”
by
S A L V AT O R E S A N F I L I P P O
(@antirez)
NoSQL?
An informal, loosely-defined
term for non-relational,
structured data storage systems
Like MongoDB, memcached, CouchDB, and Redis
See http://en.wikipedia.org/wiki/Structured_storage for comparisons
memcached
The canonically simple example
a networked “hash in the sky” behind a simple protocol
Keys Values
page:index.html <html><head>[...]
user:123:session xDrSdEwd4dSlZkEkj+
login_count “7464”
user:100:last_login_time “102736485756”
Everything’s a string (or a “blob”)
Commands just set or get data (mostly)
Take memcached’s simplicity,
Add more data types,
Add persistence,
Add more commands,
.. and more™
Redis
Redis Data Types
Strings
Lists
Sets
Sorted/Scored Sets
Hashes
all accessed by a string “key”
Redis Data Examples
Keys Values
page:index.html <html><head>[...] String
login_count 7464
users_logged_in_today { 1, 2, 3, 4, 5 } Set
latest_post_ids [201, 204, 209,..] List
user:123:session time => 10927353 Hash
username => joe
users_and_scores joe ~ 1.3483 Sorted
bert ~ 93.4 (scored)
fred ~ 283.22 Set
chris ~ 23774.17
Strings
Redis command Key Value
line client app
./redis-cli SET mystring “hello world”
./redis-cli GET mystring returns
“hello world”
Strings
GETSET INCR Works on
strings that
MGET INCRBY
appear to be
integers. Magic!
SETNX DECR
SETEX DECRBY
MSET APPEND
MSETNX SUBSTR
http://code.google.com/p/redis/wiki/CommandReference
Expiration
When caching, you don’t want things to
live forever.
Any item in Redis can be made to expire
after or at a certain time.
seconds
EXPIRE your_key 1234
TTL your_key == 1234
Deleting Keys
You can also delete data at will.
DEL your_key
EXISTS your_key == 0 (false)
Lists
LPUSH RPUSH
a b c d e f
LPOP RPOP
e.g. RPUSH my_q f
Lists
LLEN == 6
a
Xb
} c d e f
LRANGE 2 3
LREM 1 b LINDEX 5
Queues
NOT A NATIVE TYPE RPUSH
Still just a list!
a b c d e f
RPUSH my_q abc
LPOP RPUSH my_q def
LPOP my_q == “abc”
Or BLPOP to block (wait) LPOP my_q == “def”
until something can be
popped LPOP my_q == (nil)
Sets
SREM contains:aba hello
abacus cabal baba hello teabag
contains:aba
base cabaret database
SMOVE contains:aba contains:ase base
vase vaseline baseline uncase
contains:ase
unbased phase database tease
SADD contains:ase suitcase
Sets
abacus cabal baba teabag
contains:aba
cabaret database
SCARD contains:aba == 6
SISMEMBER contains:aba chips == 0 (meaning false)
SRANDMEMBER contains:aba == “teabag”
vase vaseline baseline unbased
contains:ase
phase database suitcase
SMEMBERS contains:ase == vase, vaseline,
baseline, unbased,
phase, database,
suitcase
Sets
contains:aba
abacus cabal baba teabag
cabaret
database
vase vaseline baseline
unbased phase suitcase
contains:ase
SINTER contains:aba contains:ase == database
This is only a simple example. SINTER can take any number of arguments!
SUNION is another command that will join sets together.
contains:aba
Sets
abacus cabal baba teabag
cabaret
database
vase vaseline baseline
unbased phase suitcase
resultset
contains:ase
database
SINTERSTORE resultset contains:aba contains:ase
SUNIONSTORE does the same for set unions.
Sorted Sets?
Sorry - no time!
Basically, like normal sets but each
element can have a “rank” or “score” and
be returned or sorted by it.
Hashes
product:1
HSET product:1 created_at 102374657
created_at 102374657
product_id 1 HSET product:1 product_id 1
name Twinkies HSET product:1 name “Twinkies”
available 10
HSET product:1 available 10
HGET product:1 name == Twinkies
HLEN product:1 == 4
HKEYS product:1 == created_at, product_id,
name, available
HGETALL product:1 == created_at => 102374657
product_id => 1
[.. etc ..]
Also...
HVALS HEXISTS HINCRBY HMGET HMSET
Session Storage
Session 8d3e4
created_at: 102374657
user_id: 1
It’s basically a hash
HSET session:8d3e4 created_at 102374657
HSET session:8d3e4 user_id 1
OR
HMSET session:8d3e4 created_at 102374657 user_id 1
Then let Redis automatically expire it in 24 hours!
EXPIRE session:8d3e4 86400
Redis Social Network
Users
have names, can follow others, and be
followed
Posts
are things like messages, photos, etc.
User Post Post Post
id: 1
name: joe has many..
User Post Post Post
id: 2
name: fred has many..
User Post Post Post
id: 1
name: joe
User Post Post Post
id: 2
name: fred
user:1:name joe
username:joe 1 So we can do a two
way reference
post:1:content hello world
post:1:user 1 Ditto
Building unique key names
with colons like
user:1:name
is just a
convention
Any string will dooooo.....
User Post Post Post
id: 1
name: joe
User Post Post Post
id: 2
name: fred
set user:1:name joe
set username:joe 1
set post:1:content “hello world”
set post:1:user 1
Remember, SET and GET are used for string values
User Post Post Post
id: 1
name: joe
User Post Post Post
id: 2
name: fred
user:1:posts [3, 2, 1] List
User Post Post Post
id: 1
name: joe
User Post Post Post
id: 2
name: fred
user:1:posts [3, 2, 1]
lpush user:1:posts 1
lpush user:1:posts 2
lpush user:1:posts 3
LPUSH and RPUSH add items to the start or end of a list
User User
id: 1 id: 3
name: joe name: bill
User User
id: 2 id: 4
name: fred name: jane
user:1:follows {2, 3, 4} Set
Order not
important
sadd user:1:follows 2
sadd user:1:follows 3
sadd user:1:follows 4
SADD and SREM add or remove elements to/from a set
User User
id: 1 id: 3
name: joe name: bill
User User
id: 2 id: 4
name: fred name: jane
You might want to track the relationship
in the opposite direction too. Just create
another set!
user:1:followed_by {3}
sadd user:1:followed_by 3
A Simple Social Network
Keys Values
user:1:name joe
user:2:name fred
username:joe 1
username:fred 2
user:1:follows {2,3,4} Set
user:2:follows {1}
user:1:followed_by {2}
user:2:followed_by {1}
post:1:content “Hello world”
post:1:user 2
post:2:content “Blah blah”
post:2:user 1
user:1:posts [2,3,4] List
user:2:posts [1,5,6]
Simplified from the earlier graphs due to lack of space :-)
Unique IDs
INCR next_post_id
If next_post_id doesn’t exist or doesn’t
contain a number, it’ll be set at 0,
incremented, and 1 will be returned.
returns 1 post:1:etc
INCR next_post_id
INCR increments the element by 1 and
returns the new value. Great for unique IDs!
returns 2
or next_user_id!
Creating a new user
INCR next_user_id returns [uid]
SET user:[uid]:name [username]
SET username:[username] [id]
Creating a new post
INCR next_post_id returns [pid]
SET post:[pid]:content [content]
SET post:[pid]:user [pid]
LPUSH user:[uid]:posts [pid]
LPUSH posts:global [pid]
SORT SUBSCRIBE
MONITOR
ZCARD
PUBLISH
SLAVEOF
RENAME SAVE
SELECT
Enough commands!
I haven’t covered them all though..
On to softer issues.
Atomicity
Redis is single threaded
No locking necessary
In other words, commands like INCR
won’t tread on each other’s toes coming
from multiple clients simultaneously!
Redis Factoids
BSD licensed (free, open)
Sponsored by VMware
Written in ANSI C
Good community (list, IRC & wiki)
Works on all POSIX-compliant UNIXes
An unofficial Windows/Cygwin build is available
Installation
Download a tarball or clone the git repo
Run make
redis-server and redis-cli are ready to roll
(You can make a config file later, if you want.)
http://code.google.com/p/redis/
Performance
Depends a lot on configuration
and operation complexity.
Common range from 5000 to
120,000 rps for basic ops
GET/SET/LPUSH/LPOP, etc.
(ultra low end to high end hardware)
Performance
redis-benchmark tool on a CentOS virtual
machine on a 2009 iMac
}
GET: 28011 rps
SET: 36101 rps average
INCR: 36496 rps ~36000
LPUSH: 38759 rps
LPOP: 38610 rps
And that’s with 1024 byte payloads!
Persistence
Dump data to disk after certain
conditions are met. Or
manually. SAVE and BGSAVE
commands
AND/OR
An append only log file
(which can be optimized/rebuilt automatically)
but you need to set
this up in a config
file
Language Support
Ruby, Python, PHP, Erlang,
Tcl, Perl, Lua, Java, Scala,
Clojure, C#, C/C++,
JavaScript/Node.js, Haskell,
IO, Go
i.e. anything actually worth using
Missed a lot, so where next!?
Google “Redis”
the official site is great
http://coder.io/tag/redis
for news and articles
P.S. I’m writing a Redis book a little like this presentation.
E-mail peter@peterc.org to be put on an announce list!
