Using MySQL for Distributed Database

Architectures

Peter Zaitsev
CEO, Percona
Percona Technical Webinars
July 31, 2018

© 2018 Percona.
1
Presentation

Cover Basics

Why Going Distributed

How to do it
© 2018 Percona.
2
Distributed ?

MySQL Deployment on More

than one System

© 2018 Percona.
3
Modern MySQL Scalability

© 2018 Percona.
4
Even more Modern MySQL Scalability

© 2018 Percona.
5
Single MySQL Instance Can Do
Hundreds of Thousands of Queries/Sec

Tends of Thousands of Updates/Sec

Traverse Tens of Millions of Rows/Sec

Comfortably Handle Several TB Database size

© 2018 Percona.
6
Lets Do Some Math
100.000 QPS

10 Queries per User Interaction

10.000 User Interactions/sec

864.000.000 User Interactions/Day

30 User Interactions/User Avg

28.000.000 Daily Active Users Possible

15M of Daily Active Users counting time of day skew

© 2018 Percona.
7
Distributed Systems Tend To be
More Complicated to Develop Against

More Complicated to Operate

Have Additional Performance Bottlenecks

Have Complicated Failure Modes

© 2018 Percona.
8
With All of this ?

Why “Go Distributed” ?

© 2018 Percona.
9
Reasons to “Go Distributed”

High Data
Scalability
Availability Distribution

© 2018 Percona.
10
High Availability with MySQL

Cold Standby (ie DRBD)

Failover (Classical Replication)

Active-Active Clustering (PXC, MySQL Group Replication)

© 2018 Percona.
11
Q1:What Failure Modes Do you Consider ?
Server Crash/Server Hardware Failure

Software Bugs and Storage Corruption

Network Failure

“Datacenter” Failure

Developer Mistakes/Intruder Actions

© 2018 Percona.
12
Q2:What Data Am I allowed to lose ?

Limited Data Loss Allowed

No Data Loss

No Transactions In Flight Loss

© 2018 Percona.
13
Q3:How Quickly do you need to Recover ?
“Immediate?”

“Seconds?”

“Minutes?”

“Hours?”
© 2018 Percona.
14
Speed of Light Realities

Data •Synchronous – Slow

Propagation •Asynchronous –
Can Be Data Loss Can Occur
© 2018 Percona.
15
Scalability

Scaling Scaling Scaling

Reads Writes Data Size

© 2018 Percona.
16
Data Distribution

Some Data
Must be in •User Latency Reasons
Specific •Legal and Compliance
Geographic Reasons
Location
© 2018 Percona.
17
Distributed Architectures with MySQL

© 2018 Percona.
18
Main Concepts
Replication

Sharding

Failover Management

Traffic Management
© 2018 Percona.
19
Replication

Having Multiple Copies of the

data, updated with changes

© 2018 Percona.
20
Availability

Service Stays up when

component fails

© 2018 Percona.
21
Availability via Redundancy

Have more than one system

Works well for stateless systems

Is not enough for databases

© 2018 Percona.
22
Availability via Replication

Redundant Paired with

Computing Replicated
Resource Data
© 2018 Percona.
23
Where Replication Happens

Storage Level

Database Level

Application Level
© 2018 Percona.
24
Storage Level Replication

Replication in SAN/NAS, DRBD

Typically provides cold standby

Simple choice which works with many systems

Amazon Aurora – Smart Storage

© 2018 Percona.
25
Database level

Most Flexible

Most Common

Hot/Warm Spare

Some can do Active-Active

© 2018 Percona.
26
Application Level
Hard to get right

Rarely used, especially “used right”

Partial Replication/Syncronization

Smart conflict resolution

Cross Vendor Redundancy

© 2018 Percona.
27
Sharding

Split data set by certain criteria

and store such “shards” on
separate “clusters”
© 2018 Percona.
28
Typical Sharding

By User

By Customer Account/Company
© 2018 Percona.
29
Sharding and Replication

Sharding Almost Always

used with some kind of
Replication
© 2018 Percona.
30
Failover Management

Many Replication Modes Require “Failover”

Need some Tool To Decide on Failover

Need some Tool to Perform Failover

© 2018 Percona.
31
Traffic Management
Scaling with Large Number of Connections

Routing Traffic to Right “Shard”

Read-Write Splitting

Load Management

Avoiding “Dead” Nodes

© 2018 Percona.
32
Options in MySQL

© 2018 Percona.
33
Replication Options in MySQL
Classical MySQL Replication

MySQL Group Replication

Percona XtraDB Cluster and Galera

MySQL (NDB) Cluster

© 2018 Percona.
34
Classical MySQL Replication

© 2018 Percona.
35
Classical MySQL Replication Properties
Asynchronous or Semi-Synchronous

Parallel (since MySQL 5.7)

Many Masters to Many Slaves (MySQL 5.7)

No Conflict Resolution or “Protection”

No Built-in Failover
© 2018 Percona.
36
Advanced MySQL Replication Topologies

© 2018 Percona.
37
PMM Dashboard for Replication

© 2018 Percona.
38
MySQL Group Replication

© 2018 Percona.
39
MySQL Group Replication (New in 5.7)
“Group of Peers”

Write-Anywhere or Dedicated Writer

Asynchronous Replication with Flow Control

Conflicts Prevented through Certification

Built in Failover

No Automated Provisioning
© 2018 Percona.
40
Percona XtraDB Cluster/Galera

© 2018 Percona.
41
Percona XtraDB Cluster Topology

© 2018 Percona.
42
PXC/Galera Properties
Write to Any Node

Certification Based Replication

Virtually Synchronous; Can ensure no stale reads

Built in Fail-Over

Built-in Node Provisioning

© 2018 Percona.
43
Transaction Commit Flow

© 2018 Percona.
44
PXC 5.7 Performance Improvements
http://bit.ly/2qGCr0T and http://bit.ly/2pzvAIW

© 2018 Percona.
45
PMM Dashboard for PXC

© 2018 Percona.
46
MySQL Cluster

Mostly In Memory Storage

Synchronous Replication

Pessimistic Locking

Conflict Detection with Async Replication

Niche Use Only

© 2018 Percona.
47
Sharding in MySQL

Typically Painful or Very Painful

No Built-In or Standard Solution

Many Custom App-Layer Implementation Exists

© 2018 Percona.
48
Thinking about Sharding

Query Routing to the right Shard

Data Aggregation for Cross-Shard Queries

Shard Balancing
© 2018 Percona.
49
Vitess
Stated by team in YouTube

Now Open Source in CNCF

Deployed at Slack

Supports MySQL Protocol

Supports Cross Shard Queries

Supports Re-Sharding
© 2018 Percona.
50
Vitess Architecture

© 2018 Percona.
51
Failover in MySQL

MHA

MySQL Failover

Orchestrator
© 2018 Percona.
52
Orchestrator Screenshot

© 2018 Percona.
53
Traffic Management Solutions
DNS • Not Immediate, Does not handle existing connections

HAProxy • TCP/IP Port level. Does not understand MySQL Protocol

Elastic Load Balancer (etc) • Like HA Proxy in the Cloud

MySQL Router • Currently very basic

MaxScale, ScaleArc • Proprietary Solutions

ProxySQL • MySQL Protocol Aware 100% Open Source Proxy Solution for MySQL

© 2018 Percona.
54
ProxySQL
• 100% OpenSource
• Well Supported
• Read-Write Splitting
• Caching
• Query Rewrites
• Load Balancing
• 10k+ Connection Support

© 2018 Percona.
55
 Dedicated shared ProxySQL 5
6

application server 1 application server 2 application server 3

ProxySQL

PXC node 1 PXC node 2 PXC node 3

© 2018 Percona.
56
 ProxySQL on application side 5
7

application server 1 application server 2 application server 3

ProxySQL ProxySQL ProxySQL

PXC node 1 PXC node 2 PXC node 3

© 2018 Percona.
57
ProxySQL at PMM

© 2018 Percona.
58
Cloud Only Options

DBaaS in
•Amazon RDS
the cloud
•Amazon Aurora
comes with
•Google Cloud SQL
built-in HA •Azure Database for MySQL
Options
© 2018 Percona.
59
Beyond MySQL

© 2018 Percona.
60
MySQL is a Part of your Data Infrastructure

MySQL is Good for Many

Things but MySQL is not
Good for Everything
© 2018 Percona.
61
MySQL can be used together with

Full Text Queue/Stream Analytics at

Caching
Search Processing Scale
• Redis, • Elastic, Solr • Kafka • Hadoop,
Memcache Clickhouse

© 2018 Percona.
62
Cross Technology Replication
Trigger Based • SymmetricDS, Self-Made Solution
Replication Tools

Kafka Gateway • Debezium, Yelp’s MySQLStreamer

Tungsten Replicator • Replication from MySQL to Variety Data Stores

Custom Binary Log • MySQL to Tarantool Replication

Consumers • MySQL to ClickHouse Replication

© 2018 Percona.
63
Summary

Lots of Options Exist for Building

Distributed Architectures with MySQL

Lots of Option for Integrating MySQL

with other Open Source Data Stores
© 2018 Percona.
64
 November 5-7th, 2018
Call for Papers Open!
Connect. Accelerate. Innovate.
● MySQL, MongoDB, PostgreSQL & other Open Source
Databases
● Security, Open Source Databases, Serverless, Cloud
or On Premise
● High Availability, Scalability
● Business Goals, Case Studies, What the Future Holds
● Radisson Blu Hotel, Frankfurt, Germany
Submit Your Proposal by August 10th!
Get Your Tickets for Percona
Live Europe!
● We’re delighted to be in Frankfurt this year! A
vibrant city, in a central location with many direct
flights makes it easy for you to get here!
● Percona Live Europe 2018 includes a new
business track that covers the best ideas for how
open source databases and database
technologies can address and solve business
issues such as application time-to-market,
resource costs, OPEX and CAPEX expenses, etc.
Super Saver Tickets Available Until
August 19th!
Prices Increase on the 20th!

© 2018 Percona.
65
www.percona.com/live/e18/
Thank You!

© 2018 Percona.
66