DataStax: Relational Scaling and the Temple of Gloom

Luke TillmanTechnical Evangelist at DataStax (@LukeTillman)

• Evangelist with a focus on Developers

• Long-time Developer on RDBMS (lots of .NET)

Who are you?!

2

The Good ol' Relational Database

© 2015. All Rights Reserved. 3

First proposed in 1970

The Relational Database Makes us Feel Good


SQL is ubiquitous and allows

flexible querying

Data modeling is well understood

(3NF or higher)

ACID guarantees make us feel good


Building and Scaling our Applications





I'm getting too old for this…

Scaling Up


All these JOINs are Killing Us

9

SELECTarray_agg(players),player_teams

FROM (SELECT DISTINCT

t1.t1player AS players,t1.player_teams

FROM (SELECT

p.playerid AS t1id,concat(p.playerid, ':', p.playername, ' ') AS t1player,array_agg (pl.teamid ORDER BY pl.teamid) AS player_teams

FROM player pLEFT JOIN plays pl

ON p.playerid = pl.playeridGROUP BY p.playerid, p.playername

) t1INNER JOIN (

SELECTp.playerid AS t2id,array_agg (pl.teamid ORDER BY pl.teamid) AS player_teams

FROM player pLEFT JOIN plays pl

ON p.playerid = pl.playeridGROUP BY p.playerid, p.playername

) t2ON t1.player_teams = t2.player_teamsAND t1.t1id <> t2.t2id

) innerQueryGROUP BY player_teams

© 2015. All Rights Reserved.


9

SELECT * FROM denormalized_view

Let's Denormalize!



9

But I thought data modeling was 3NF or

higher?! There can be only one!© 2015. All Rights Reserved.

Read Replication

12

Client

Users Data

Replica 2

Replica 1

Primary

Write Requests

Read

Req

uest

s


Replication Lag

Consistent results? Nope, now

eventually consistent

Replication speed is limited by

the speed of light

Sharding


Client

Router

A-F G-M N-T U-Z

Users Data

Queries that aren't on the shard key

require scatter-gather

Resharding can be a painful, manual process

Replication for Availability


Client

Users Data

Failover

Process

Monitor

Failover

Failover takes time. How long are

you offline while it's happening?

And while you're offline...


Putting it All Together


Client

Users Data



Client

Users Data

Sharding

A-F G-M N-T U-Z



Client

Users Data

Sharding

Router

A-F G-M N-T U-Z



Client

Users Data

Router

A-F G-M N-T U-Z

Sharding and Replication (and probably Denormalization)



Client

Users Data

Failover

Process

Router

A-F G-M N-T U-Z




Client

Users Data

Failover

Process

Monitor

Failover

Router

A-F G-M N-T U-Z




Client

Users Data

Failover

Process

Monitor

Failover

Router

A-F G-M N-T U-Z




Client

Users Data

Failover

Process

Monitor

Failover

Router

A-F G-M N-T U-Z

Replication Lag




Client

Users Data

Failover

Process

Monitor

Failover

Router

A-F G-M N-T U-Z

Replication Lag




Client

Users Data

Failover

Process

Monitor

Failover

Router

A-F G-M N-T U-Z

Replication Lag




Client

Users Data

Failover

Process

Monitor

Failover

Router

A-F G-M N-T U-Z

Replication Lag


What is Cassandra?


A linearly scaling and fault tolerant

distributed database

What is Cassandra?




• Data spread over many nodes

• All nodes participate in a cluster

• All nodes are equal

• No SPOF (shared nothing)

• Run on commodity hardware

What is Cassandra?




• Have more data? Add more nodes.

• Need more throughput? Add more nodes.

What is Cassandra?




• Nodes down != Database Down

• Datacenter down != Database Down

• No middle of the night phone calls

Multi Datacenter with Cassandra


America Zamunda

Client

Fault Tolerance in Cassandra


You have the power to control fault

tolerance in Cassandra

Replication Factor


How many copies of the data should exist?

Client

Write Beetlejuice

RF=3

Beetlejuice Beetlejuice

Beetlejuice

Consistency Level


How many replicas do we need to hear

from before we acknowledge?

CL=ONE

Copy #1 Copy #2

Copy #3

Client

Consistency Level


How many replicas do we need to hear

from before we acknowledge?

CL=QUORUM

Copy #1 Copy #2

Copy #3

Client

Consistency Levels and Speed


Use a lower consistency level like ONE to

get faster reads and writes

Consistency Levels and Eventual Consistency


Use a higher consistency level like

QUORUM if you don’t want to be surprised

by data from the past (stale data)

Before you get too excited...


Cassandra is not...

• A Data Ocean, Lake, or Pond

• An In-Memory Database

• A Queue

• A magical database luck dragon

that will solve all your database

use cases and problems


How bad of an idea?


Actually a 90's movie Actually a 70's movie

Why Arnold?! Why?

Cassandra is good when...

• Uptime is a top priority

• You have unpredictable or high scaling requirements

• The workload is transactional (i.e. OLTP not OLAP)

• You are willing to put the time and effort into understanding how it works and how to use it



Movie References(in order of appearance)

Leap of Faith (1992)

Patton (1970)

The Aristocats (1970)

When Harry Met Sally (1989)

Beverly Hills Cop (1984)

Lethal Weapon (1987)

Big (1988)

Trading Places (1983)

Highlander (1986)

Spaceballs (1987)


Highlander (1986)

Spaceballs (1987)

Rain Man (1988)

Ghostbusters (1984)

Gremlins (1984)

Star Trek II: The Wrath of Khan (1982)

Star Wars Episode VI: Return of the Jedi (1983)

Weekend at Bernie's (1989)

Coming to America (1988)

Masters of the Universe (1987)

Beetlejuice (1988)

The Goonies (1985)

Top Gun (1986)


Top Gun (1986)

Back to the Future (1985)

Footloose (1984)

The NeverEnding Story (1984)

Batman and Robin (1997)

Star Wars (1977)

Twins (1988)

The Karate Kid (1984)

Find me on Twitter: @LukeTillman

Thank You

DataStax: Relational Scaling and the Temple of Gloom

Technology