Location via proxy:   [ UP ]  
[Report a bug]   [Manage cookies]                
Scribd
Upload
343 views

MySQL Architecture Design Patterns For Performance, Scalability, and Availability Presentation

This document discusses various MySQL architecture design patterns for performance, scalability, and availability. It begins with an overview of high availability and scale-out approaches. It then covers specific architectural tools like MySQL replication, DRBD clustering, and MySQL Cluster. The remainder of the document discusses and compares various design patterns for MySQL architectures, including master-slave replication, master-master replication, DRBD pairs, sharding, and combining tools like replication, DRBD, and sharding.

Uploaded by

umi
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
343 views

MySQL Architecture Design Patterns For Performance, Scalability, and Availability Presentation

This document discusses various MySQL architecture design patterns for performance, scalability, and availability. It begins with an overview of high availability and scale-out approaches. It then covers specific architectural tools like MySQL replication, DRBD clustering, and MySQL Cluster. The remainder of the document discusses and compares various design patterns for MySQL architectures, including master-slave replication, master-master replication, DRBD pairs, sharding, and combining tools like replication, DRBD, and sharding.

Uploaded by

umi
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 26

<Insert Picture Here>

MySQL Architecture Design Patterns for


Performance, Scalability, and Availability
Brian Miezejewski
Principal Manager Consulting

Alexander Rubin
Principal Consultant

Agenda
HA and Scale-Out
Architectural Tools
MySQL Replication
DRDB Clustering
MySQL Cluster

Design Patterns

Simple Master-Slave
Master and Many Slaves
Master-Master
DRBD Pair
DRBD Master with Many Slaves
MySQL Cluster

Scale-out/Sharding

MySQL High Availability


What is High Availability?
Availability of resources in a computer system

Continuous Availability
Non-stop service
No disruption of service even during a fail over

The Five 9s of Availability


Clustering &
Geographical
Redundancy

disasters
power
failures
network failures

Clustering
Technologies

hardware failures
software failures

Replication
Technologies

maintenance operations
hardware upgrades

Well-Managed

software upgrades
Unmanaged
35 days
Small
Business

4 days
ISPs &
Mainstream
Business

9.

8 hours

50 mins

Data
Centers

Banking
Medical

5 mins
Telco
Military
Defense

Scale-Up vs. Scale-Out

Scale-Up

Vertical
Expensive SMP hardware
Proprietary software
Platform lock-in
Fork Lift to increase capacity & performance

High Cost

Scale-Out

Horizontal
Commodity Intel/AMD hardware
Open source software
Low Cost
Platform independence
Add servers to increase capacity & performance

MySQL Replication
Writes & Reads

Web/App
Server

Writes
relay
binlog

mysqld

I/O
Thread

SQL
Thread

index &
binlogs
data
Replication

binlog
mysqld

MySQL Master

MySQL Slave

data

MySQL Replication (Scale-Out)


Web/App
Server

Writes & Reads

Reads

Master Server

Slave Server
Backups

MySQL Replication
Writes

Writes
Index &
Bin Log
Rotation

Possible Roles
Fail over server
Used for performing backups
Read/Write load balancing
Additional slaves allow Scale-Out

MySQL Replication - Fail Over


Web/App
Server

ual
Man ver
lO
Fai

Reads

ver Server
O
Slave
l
Fai

MasterXServer

Backups

MySQL Replication
X

Writes

Writes
Index &
Bin Log
Rotation

Possible Roles
Fail over server
Used for performing backups
Read/Write load balancing
Additional slaves allow Scale-Out

Linux Heartbeat, DRBD & MySQL


Web/App
Server

Linux Heartbeat
= Virtual IP =
10.10.10.10

Passive Server

Active Server

= Private IP =
10.10.10.20

Primary DRBD

= Private IP =
10.10.10.21

DRBD

Secondary DRBD

Linux Heartbeat, DRBD & MySQL


Web/App
Server

Linux Heartbeat
= Virtual IP =
10.10.10.10

ive Server
t
Passive
c
A

Active
X Server

= Private IP =
10.10.10.20

Primary DRBD

= Private IP =
10.10.10.21

DRBD

D
RB
D
Secondary
DRBD
ry
ma
Pri

Scale-Out Cluster Architecture

MySQL
Server

MySQL
Server

MySQL
Server

MySQL
Server

MySQL
Server

NDB API

Data
Node

Data
Node

NDB
Storage Engine

Data
Node

Data
Node

MySQL Cluster

Management
Server

Management
Server

MySQL Cluster Data Node Architecture


ID

Capital

Country

UTC

Copenhagen

Denmark 2

Berlin

Germany 2

New York City USA

-5

Tokyo

Japan

Athens

Greece

Moscow

Russia

Oslo

Norway

Beijing

China

Partition 1
Partition 2

Data
Node

P1-Primary

P1-Secondary

P2-Secondary

P2-Primary

Partition 3

Node Group 1
Partition 4

Data
Node

Four Data Nodes


Two Replicas
Two Node Groups

Data
Node

Data
Node

P3-Primary

P3-Secondary

P4-Secondary

P4-Primary

Node Group 2

Scale-Out Cluster Architecture Fail Over

XMySQL
Server

MySQL
Server

MySQL
Server

MySQL
Server

MySQL
Server

NDB API

Data
Node

Data
X
Node

NDB
Storage Engine

Data
Node

Management
Server

Data
Node

Management
X
Server

MySQL Cluster

Patterns

Master/Slave Replication
Master with Read Only Slaves
Master to Master
Ring Replication
DRDB HA
DRBD Master Pair with Read Only Slaves
Master to Master Cross Datacenter + DRDB
Simple Sharding
Large Sharing/Scale-out solution + DRDB
Sharding + DRDB Fail-Over + Geographical
Redundancy

Pattern: Master/Slave Replication


Master

Asynchronous

Slave

MySQL
Replication

Advantages

Can read from either server


Up to 2x read scaling
Easy to setup, configure, and maintain
Easy fail-over, fail-back
Poor mans HA, ~99.9% (3 nines) HA

Disadvantages

~.5 write scaling or slave will fall behind


All writes must go to the master
Replication is asynchronous; potential delay in writes to slave
Data can be left un-replicated on the Master in a fail-over.
Saving and re-syncing data left in the master on fail-over is a manual task
that is not always straight forward.

Pattern: Master with Read Only Slaves


Master
Slave 1

Slave 2

Slave 4

Slave 3

A master can have many slaves,

- ~1% load on master per slave


- 5+ slaves are not unusual, Most Ive heard of is 80 slaves

Reading data is HA, can be designed to be 4-5 nines


Add more slaves to scale read load
Can load balance reads across slaves
Can do long running reports on slave
Use slave for backup
JDBC drivers support this architecture

Pattern: Master to Master Replication


Master 1

Master 2

Advantages

Can read from or write to either server


Works across WAN - Supports geographical redundancy
Can be setup as active/passive
Can get some performance advantage by splitting load and
keeping it sticky, i.e. east coast always on master 1 and west
always on master 2

Disadvantages
Total insert/update/delete load is ~.5 one server or slaves
begin falling behind
Applications and schema must be designed to eliminate
duplicates that would break replication
Fail-back can be more complicated

Pattern: Ring Replication


Master 1

Master 3
Asynchronous
MySQL

Master 2

Replication

Master 4

Advantages
Can read from or write any server
Works across WAN - Supports multiple geographical redundancy
BP - Use DRDB HA for each

Disadvantages
Total insert/update/delete load for all servers is ~.5 of typical load of one server
or slaves begin falling behind
Applications and schema must be designed to eliminate duplicates that would
break replication
Fail-over and Fail-back is very complicated

Dont try this at home!


Unless you really know what you are doing!

Pattern: DRBD HA
Hot Primary

Synchronous
DRBD
Replication

Warm Secondary

Advantages
No data loss
Can replicate from the pair via the fail-over VIP
Allows for heavier load on primary than replication on master
Fully automated fail-over when using Heartbeat
Very easy to rebuild old primary as new secondary after fail-over
Can be 3-4 9s Available
Disadvantages
Secondary is passive
30-300 Second delay in recovery after fail-over
Only works on Linux

Pattern: DRBD Master Pair with Read Only Slaves


Hot Primary

Synchronous
DRBD
Replication

Warm Secondary

Master Pair

HA - Both for reads and write


Scale-out for reads
Often the optimal solution for most needs
Replication is via fail-over VIP so slaves transparently fail-over to
secondary on primary failure
Can load balance across slaves

Pattern: DRBD Master Pair with Read Only Slaves

Pattern: Master to Master Cross Datacenter + DRDB

Pattern: Simple Sharding

Shard A

Shard B

Shard C

Shard D

Pattern: Large Sharing/Scale-out solution + DRDB

Pattern: Sharding + DRDB Fail-Over + Geographical


Redundancy

The preceding is intended to outline our general


product direction. It is intended for information
purposes only, and may not be incorporated into any
contract. It is not a commitment to deliver any
material, code, or functionality, and should not be
relied upon in making purchasing decisions.
The development, release, and timing of any features
or functionality described for Oracles products
remains at the sole discretion of Oracle.

You might also like