8. key value databases laboratory

Cloning Twitter With
Redis
Dr. Fabio Fumarola

Motivation
• The programming community considers that key-
value cannot be used as replacement for a relational
database.
• Here we show how a key-value layer is an effective
data model to implement many kinds of applications.
2

A Twitter Clone
• One of the most successful new Internet services of
recent times is Twitter.
• Since its launch it has exploded from niche usage to
usage by the general populace, with celebrities such
as Oprah Winfrey, Britney Spears, and Shaquille
O'Neal, and politicians such as Barack Obama and Al
Gore jumping into it.
3

Why Twitter?
• Simple: it does not care what you share, as a long it is less
than 140 characters
• A means to have public conversation: Twitter allows a user
to tweet and have users respond using '@' reply, comment,
or re-tweet
• Fan versus friend
• Understanding user behavior
• Easy to share through text messaging
• Easy to access through multiple devices and applications
4

Main Features
• Allow users to post status updates (known as
'tweets' in Twitter) to the public.
• Allow users to follow and unfollow other users. Users
can follow any other user but it is not reciprocal.
• Allow users to send public messages directed to
particular users using the @ replies convention (in
Twitter this is known as mentions)
5

Main Features
• Allow users to send direct messages to other users,
messages are private to the sender and the recipient
user only (direct messages are only to a single
recipient).
• Allow users to re-tweet or forward another user's
status in their own status update.
• Provide a public timeline where all statuses are
publicly available for viewing.
• Provide APIs to allow external applications access.
6

Redis CLI
• Save a Key Value
– SET foo bar
• Get a value for a key
– GET foo => bar
• Del a key value
– DEL foo
8

Redis CLI
• Increment a value for a key
– SET foo 10
– INCR foo => 11
– INCR foo => 12
– INCR foo => 13
• INCR is an atomic operation
– x = GET foo
– x = x + 1
– SET foo x
9

Redis CLI
• The problem with this kind of operation is when
multiple client update the same key
– x = GET foo (yields 10)
– y = GET foo (yields 10)
– x = x + 1 (x is now 11)
– y = y + 1 (y is now 11)
– SET foo x (foo is now 11)
– SET foo y (foo is now 11)
10

Redis CLI: LIST
• Beyond key-value stores: lists
– LPUSH mylist a (now mylist holds 'a')
– LPUSH mylist b (now mylist holds 'b','a')
– LPUSH mylist c (now mylist holds 'c','b','a')
• LPUSH means Left Push
• There is also the operation RPUSH
• This is very useful for our Twitter clone. User updates
can be added to a list stored in username:updates,
for instance.
11

Redis CLI: LIST
• LRANGE returns a range from the list
– LRANGE mylist 0 1 => c,b
– LRANGE mylist 0 -1 => c,b,a
• The last-index argument can be negative, with a
special meaning: -1 is the last element of the list, -2
the penultimate, and so on
12

Redis CLI: SET
• SADD is the add to set operation
• SREM is the remove from set operation
• SINTER is the perform intersection operation
• SCARD to get the cardinality of a Set
• SMEMBERS to return all the members of a Set.
13

Redis CLI: SET
– SADD myset a
– SADD myset b
– SADD myset foo
– SADD myset bar
– SCARD myset => 4
– SMEMBERS myset => bar,a,foo,b
14

Redis CLI: Sorted SET
• Sorted Set commands are prefixed with Z. The
following is an example of Sorted Sets usage:
– ZADD zset 10 a
– ZADD zset 5 b
– ZADD zset 12.55 c
– ZRANGE zset 0 -1 => b,a,c
• In the above example we added a few elements with
ZADD, and later retrieved the elements with ZRANGE
15

Redis CLI: Sorted SET
• The elements are returned in order according to
their score.
• In order to check if a given element exists, and also
to retrieve its score if it exists, we use the ZSCORE
command:
– ZSCORE zset a => 10
– ZSCORE zset non_existing_element => NULL
16

Redis CLI: HASH
• Redis Hashes are basically like Ruby or Python
hashes, a collection of fields associated with values:
– HMSET myuser name Salvatore surname Sanfilippo
country Italy
– HGET myuser surname => Sanfilippo
17

Data Layout
• When working with a relational database, a database
schema must be designed so that we'd know the
tables, indexes, and so on that the database will
contain.
• We don't have tables in Redis, so what do we need
to design?
• We need to identify what keys are needed to
represent our objects and what kind of values this
keys need to hold.
19

Users
• We need to represent users, of course, with their
– username, userid, password, the set of users following a
given user, the set of users a given user follows, and so on.
• The first question is, how should we identify a user?
• Asolution is to associate a unique ID with every user.
• Every other reference to this user will be done by id.
– INCR next_user_id => 1000
– HMSET user:1000 username antirez password p1pp0
20

Users
• Besides the fields already defined, we need some
more stuff in order to fully define a User
• For example, sometimes it can be useful to be able
to get the user ID from the username, so every time
we add an user, we also populate the users key,
which is an Hash, with the username as field, and its
ID as value.
– HSET users antirez 1000
21

Users
– HSET users antirez 1000
• We are only able to access data in a direct way,
without secondary indexes.
• It's not possible to tell Redis to return the key that
holds a specific value.
• This new paradigm is forcing us to organize data so
that everything is accessible by primary key, speaking
in relational DB terms.
22

Followers, following and updates
• A user might have users who follow them, which
we'll call their followers.
• A user might follow other users, which we'll call a
following.
• We have a perfect data structure for this. That is...
Sorted Set.
23

• So let's define our keys:
– followers:1000 => Sorted Set of uids of all the followers
users
– following:1000 => Sorted Set of uids of all the following
users
• We can add new followers with:
– ZADD followers:1000 1401267618 1234 => Add user 1234
with time 1401267618
24

• Another important thing we need is a place were we
can add the updates to display in the user's home
page.
• We'll need to access this data in chronological order
later
• Basically every new update will be LPUSHed in the
user updates key, and thanks to LRANGE, we can
implement pagination and so on.
25

• Note, we use the words updates and posts
interchangeably, since updates are actually "little
posts" in some way.
– posts:1000 => a List of post ids - every new post is
LPUSHed here.
• This list is basically the User timeline.
• We'll push the IDs of her/his own posts, and, the IDs
of all the posts of created by the following users.
Basically we implement a write fanout.
26

Following Users
• We need to create following / follower relationships.
– If user ID 1000 (antirez) wants to follow user ID 5000
(pippo), we need to create both a following and a follower
relationship.
• We just need to ZADD calls:
– ZADD following:1000 5000
– ZADD followers:5000 1000
27

Following Users
• Note the same pattern again and again.
• In theory with a relational database the list of
following and followers would be contained in a
single table with fields like following_id and
follower_id.
• With a key-value DB things are a bit different since
we need to set both the 1000 is following 5000 and
5000 is followed by 1000 relations.
28

Making it horizontally scalable
• Our clone is extremely fast, without any kind of
cache.
• On a very slow and loaded server, an apache
benchmark with 100 parallel clients issuing 100000
requests measured the average pageview to take 5
milliseconds.
• This means we can serve millions of users every day
with just a single Linux box.
29

Making it horizontally scalable
• However you can't go with a single server forever,
how do you scale a key-value store?
• It does not perform any multi-keys operation, so
making it scalable is simple:
1. you may use client-side sharding,
2. or something like a sharding proxy like Twemproxy,
3. or the upcoming Redis Cluster.
30

8. key value databases laboratory

More Related Content

What's hot (20)

Similar to 8. key value databases laboratory (20)

More from Fabio Fumarola (19)

Recently uploaded (20)

8. key value databases laboratory

Editor's Notes