Webinar Slides: MySQL HA/DR/Geo-Scale - High Noon #2: Galera Cluster

The MySQL Availability Company
MySQL HA/DR/Geo-Scale -- High Noon #2
Galera vs. Continuent Tungsten Clusters
Building a Geo-Scale, Multi-Region and Highly Available MySQL Cloud Back-End
Matt Lang, Director of Customer Success
Nov 1, 2020
“Play Hard!”

Matt Lang
Early Adopter of Linux, MySQL, and Cloud technologies.
Architected systems for startups, supporting explosive
growth in employees, customers, and sales, in automotive,
steel, medical, and tech industries.
• MySQL DBA
• Linux Systems Admin
• Site Reliability Engineer/Cloud Architect
• Author of replication for proprietary database (still in use after 10+
years)
• Weekend musician
“Play Hard!”

Continuent
“The MySQL Availability Company”

Where Does Continuent Fit In?
Continuent has witnessed the rise of Open
Source and Cloud Computing since 2004.
We have been at the forefront of the market
need for platform-agnostic, highly-available,
globally-scaling, clustered MySQL databases
that are driving businesses to the cloud today.
Continuent is not the biggest. Never.
We are the best at what we do. Always!
Geo-Scale
MySQL
DBaaS
Cloud
Computing
Open
Source
MySQL

Goals for the High Noon Webinar Series
Explore and compare how to deploy Geo-Scale MySQL using various MySQL Solutions:
• Local rapid-failover, automated high availability
• Geographically distributed, low-latency data replication with a single consolidated view
• Fast local response times for read traffic
• Ability to deploy MySQL primaries in multiple regions
• No changes to application code
• Complex schema changes while keeping applications available
• Avoid provider lock-in
MySQL is understood in a broad context, including MySQL, MariaDB and Percona Server.
Also AWS RDS MySQL, AWS RDS Aurora, Azure MySQL, and Google Cloud SQL are applicable in some context.

High Noon Series - Tungsten Clustering vs. Others

Clustering Competitive Scorecard
Top prioritized
use cases
Tungsten
Clustering
AWS Aurora with
Read Replica
Google Cloud
SQL HA
Orchestrator +
Proxy
Galera Cluster
(aka XtraDB Cluster)
Local HA
Yes, automatic
failover for primary
and replicas
Yes, 30 seconds with
DNS change
Just one replica, at
least 60 seconds for
failover
Yes – no automatic
recovery of primary
in switch
Failover not
necessary
DR (local/
remote)
Yes, local, remote
and global DR are
supported
Failover only, takes
“a few minutes,”
replication breaks
Not supported
Yes– no automatic
recovery of primary
in switch
No native DR. DR
with async but
without
management
Load balancing
Yes, automatic and
transparent
Supported but
requires application
awareness
Supported but
requires application
awareness
Yes, automatic and
transparent
Requires additional
3rd party tool
Zero downtime
maintenance
Enables both
database, operating
system and
hardware changes
Requires
maintenance
window, downtime
Requires
maintenance
window, downtime
Yes but needs
manual
reconfiguration
Changes are
supported, but can
be costly and time
consuming

Top prioritized
use cases
Tungsten
Clustering
AWS Aurora with
Read Replica
Google Cloud
SQL HA
Orchestrator +
Proxy -- DIY
Galera Cluster
Performance
High, built-in load
balancing and
read/write splitting
High performance
but read/write
splitting requires app
awareness
Requires read
replicas, app
awareness,
replication lag
High performance
Synchronous nature
implies additional
overhead during
writes
Transparency
Yes, automatic
and replicas
Automatic failover
but applications
disconnect (unless
using RDS proxy)
Automatic failover
but applications
disconnect
Yes
Requires cluster and
app made aware of
changes
Deployment Easy to deploy Easy to deploy Easy to deploy Globally easy
Complex, easier with
Galera Manager
Recovery
Fast and simple, local
and cross region
Local recovery
simple, cross region
takes time
Local recovery
simple
Simple as long as
automation works
Time consuming,
potentially blocking
donor node

Top prioritized
use cases
Tungsten
Clustering
AWS Aurora with
Read Replica
Google Cloud
SQL HA
Orchestrator +
Proxy
Galera Cluster
Monitoring
Full monitoring via
web-based GUI
Full monitoring,
though enhance
monitor is extra
charge
Easy to use basic
monitoring
Full monitoring on
web interface
3rd party tool like
ClusterControl or
Galera Manager
(newly released)
Geo-Scale
Across regions,
availability zones
Only basic read
replicas in AWS
regions
Not supported Yes
No Geo-scale with
Sync. Async option
without
management layer
Replication
To most popular
high-performance
NoSQL and data
analytics engines
Basic MySQL
replication only,
manual setup (5.5 or
later)
Basic MySQL
replication only,
manual setup (5.5 or
later)
Basic
Synchronous locally,
async globally,
manual setup
Support
24/7 with <5 min.
initial response to
urgent support
tickets
Basic support > 12
hours, critical
support extremely
expensive
Not included, critical
support extremely
expensive
Only through mailing
list/forums
Support through
MariaDB,Percona,
Codership < 4HR
response

Galera Characteristics
• True multi-master for cluster on LAN
• Synchronous replication. Almost no slave lag/very little slave divergence
• No VIP needed (all nodes are readable and writable)
• Highly available, failover not needed since all nodes are read/write
• Various MySQL versions are available
• Must use specific Galera build (cannot use just off the shelf MySQL)
• NEW – Galera Manager, GUI for Galera
• Replication is handled by a “Certification” process to eliminate conflicts and enforce ordering

Certification Based Replication
• Transaction begins as normal on source host
• When COMMIT is issued, all changed rows are collected into a ”write set.” The transaction is not
committed yet.
• Transaction is given a sequence number
• The node checks the last applied sequence number. If there is a gap:
• Node collects sequence numbers from other nodes within the gap
• Node determines if there will be primary key conflicts
• If not, all transactions are committed.
• If so, transaction is rolled back
• This happens on each node and guarantees ordering of transactions

Galera Multi Site Requirements
• Use Synchronous replication
• Writes will be slow, and proportional to the round trip time (RTT) of the WAN
• Read speed is unaffected
• Using Asynchronous replication
• Enable binary logging
• Use MySQL native replication
• Management becomes more complex when mixing 2 technologies
• Not easy to deploy multi site topologies due to above and thus not often used

Limitations Using Galera
• Write intensive workloads may perform poorly due to synchronous replication
• Poor WAN performance due to above
• Thus multi-site topologies are rarely used
• No Proxy included, so a third party tool is required
• All tables in database must have a primary key
• One slow node will slow down the entire cluster
• Replication out of cluster requires another technology (i.e. different replication method, binary
logging, etc)
• Must use official database distribution with Galera compiled in.

How To Do Better MySQL HA/DR/Geo-Scale?

Tungsten Clustering
Manager
Replicator
Manager
Replicator
Manager
Replicator
monitoring
&
control
Dataservice
Replica Replica
Primary
• Continuous MySQL Operations
• Zero Downtime Maintenance
• Geo-distributed MySQL Clusters
• Multi-Cloud / Hybrid-Cloud MySQL
• Read/Write Splitting with Intelligent Proxy
• Full MySQL Support, No App Changes

Tungsten Active-Passive Clustering
• Scale to multiple Cloud Regions or
datacenters
• Platform-agnostic means you can span
vendors and create hybrid topologies using
any combination of cloud, VM and/or bare-
metal servers
• Active/Passive
• Control all clusters from any node
• Move the write primary from site to site
with a single command when in
Active/Passive mode

Tungsten Active-Active Clustering
• Scale to multiple Cloud Regions or
datacenters
• Platform-agnostic means you can span
vendors and create hybrid topologies using
any combination of cloud, VM and/or bare-
metal servers
• Active/Active
• Control all clusters from any node
• Move the write primary from site to site with
a single command when in Active/Passive
mode

Zero Downtime Maintenance
Replica
Upgrade
Additiona
l Replicas
Switch
Primary
Upgrade
Shun replica
Upgrade MySQL
Return node to
cluster
Discard and re-
provision on failure
Repeat for
remaining replica(s)
Switch primary to
promote an
upgraded replica
Upgrade old
primary
Maintenance is
now done!

Galera vs Tungsten Clustering Scorecard
Top prioritized
use cases
Tungsten
Clustering
Galera Cluster
Local HA
Yes, automatic
and replicas
Failover not
necessary
DR (local/
remote)
Yes, local, remote
and global DR are
supported
No native DR. DR
with async but
without
management
Load balancing
Yes, automatic and
transparent
Requires additional
3rd party tool
Zero downtime
maintenance
Enables both
database, operating
system and
hardware changes
Changes are
supported, but can
be costly and time
consuming

Top prioritized
use cases
Continuent
Clustering
Galera Cluster
Performance
High, built-in load
balancing and
read/write splitting
Synchronous nature
implies additional
overhead during
writes
Transparency
Yes, automatic
and replicas
Requires cluster and
app made aware of
changes
Deployment Easy to deploy
Complex, easier with
Galera Manager
Recovery
Fast and simple, local
and cross region
Time consuming,
potentially blocking
donor node

Top prioritized
use cases
Continuent
Clustering
Galera Cluster
Monitoring
Full monitoring via
web-based GUI
3rd party tool like
ClusterControl or
Galera Manager
(newly released)
Geo-Scale
Across regions,
availability zones,
clouds, on-prem
No Geo-scale with
Sync. Async option
without
management layer
Replication
To most popular
high-performance
NoSQL and data
analytics engines
Synchronous locally,
async globally,
manual setup
Support
24/7 with <5 min.
initial response to
urgent support
tickets
Support through
MariaDB,Percona,
Codership < 4HR
response

Continuent Products
• Tungsten Clustering
• Tungsten Replicator
• Tungsten Dashboard

Tungsten Clustering
• Continuous MySQL Operations
• Zero Downtime Maintenance
• Geo-distributed MySQL Clusters
• Multi-Cloud / Hybrid-Cloud MySQL
• Read/Write Splitting with Intelligent Proxy
• Full MySQL Support, No App Changes

Tungsten Key Benefits
Industry Best 24/7 MySQL Customer Service
• Highly Qualified 24/7 support. Our support team member have 20 or more years of MySQL DBA and
Site Reliability Experience
• 24/7 support comes with 1-hour SLA, with response times for urgent requests averaging less than 3
minutes
• MySQL uptime measured in months or years
Cost Savings
• Use the free open-source MySQL for your business-critical needs
• Optimize costs by selecting the most cost-effective cloud environment(s) at any given time
• Eliminate downtime risks and their associated costs, including during maintenance operations (zero-
downtime maintenance operations)
• Reduce DBA time spent on admin and recovery operations, lowering your costs while increasing
reliability.

Tungsten Clustering Key Benefits
Continuous MySQL Operations
• MySQL High Availability and Disaster Recovery solution, which provides redundancy within and
across data centers
• Immediate failover for maximum availability and data protection of business-critical MySQL
applications
• Reduce MySQL recovery time from hours or days to mere seconds
• Dashboard provides graphical view and management of all globally distributed MySQL clusters
Zero Downtime MySQL
• Site-level and cross-site failover ensures application availability
• Upgrade hardware, software and data without taking applications offline
• MySQL compatibility means seamless migration of your data and applications

Geo-Scale MySQL
• Load-balance MySQL read operations across multiple replicas, locally and globally
• Geo-distributed MySQL clusters bring data close to your application users for faster response times
• Easily add more MySQL clusters as needed for unlimited scaling, both locally or across the globe
Hybrid-Cloud and Multi-Cloud MySQL
• Deploy in the cloud, VM and bare metal environments
• Mix-and-match on-premises, private and public clouds (incl. Amazon AWS, Google Cloud and
Microsoft Azure)
• Easy, seamless migration from cloud to cloud to avoid vendor lock-in in any specific cloud provider
• Withstand node, data center, zone or region failures or outages

Intelligent MySQL Proxy
• Provides intelligent traffic routing to a valid MySQL Primary, locally and globally
• Scale read queries via query inspection and other methods
• Application and active users do not disconnect during MySQL failover events
Most Advanced MySQL Replication
• Filter and transform your data in-flight
• No more ETL, get real-time data feeds into your analytics
• Replicate directly into popular analytic repositories: AWS RedShift, Hadoop (HDFS), Kafka and Vertica
• Unlimited real-time transactional data transfer to eliminate escalating replication cost of ETL-based
alternatives

Full MySQL Support, No Application Changes
• Deploy and Configure MySQL clusters in minutes
• Not just a ‘MySQL-compatible” solution, but fully MySQL-compliant. Use any of your off-the-shelf
MySQL, MariaDB and Percona Server versions
• Support for all modern MySQL (5.x through 8.x) and MariaDB (5.x and 10.x) versions and features
• SSL support for all in-flight traffic
• Native MySQL support means easy and complete migration of your data and applications

Tungsten Replicator
• High performant Real-Time replication
• Multiple Heterogeneous Targets
• AWS Aurora / RDS
• AWS Redshift
• HPE Vertica
• Hadoop
• Kafka
• and many more
• Fan-in / Fan-Out / Primary-Primary
• Flexible, Advanced Filtering
• Available as an AMI through Amazon
Marketplace (with free 14-day trial)

Tungsten Dashboard

Proven Continuent Team
• Our core team has been building data availability solutions since 2004
• We have an expert team of data replication, database connectivity and database cluster management
engineers
• We have a world-class support team
• Each member with 20 or more years of operational database experience
• Our team is available to respond effectively and immediately to our customers’ needs
• Average response time for an urgent support request less than 3 minutes

Proven Solutions
Our tightly-integrated, complete and fully tested solutions reliably handle billions of transactions per
day and support businesses with billions of dollars in combined revenue:

Why to Choose Continuent
Customer Satisfaction
during the most recent customer survey
Customer Renewal Rate
many multi-year subscription renewals
Average Response Time
in minutes for urgent customer support
cases
Average Customer
Life-Span
years using Continuent solutions

Continuent Timeline
• The MySQL Availability Company
• Established in 2004
• Customers Industries include
• SaaS
• eCommerce
• Finance
• Telco
• Online Gaming
• Business-critical MySQL apps
• From Startup to Enterprise companies
2004 Continuent Established
2009 3rd Generation Continuent Tungsten, the first
fully MySQL-compliant clustering solution
2014 VMware Acquired Continuent to build
DBaaS for vCloud Air
2016 “New” Continuent breaks away from VMware
(after they abandoned vCloud Air)
2017 Major Tungsten release (v5) with hardened
security features
2018 Tungsten v6 ships with major multi-site
clustering changes
2020 Marketo is the first 10th anniversary Tungsten
customer (grown from 4 clusters to over 100
clusters)
202x The Future….is in the Cloud – Tungsten Cloud

THANK YOU FOR LISTENING
continuent.com
The MySQL Availability Company
Matthew Lang
matthew.lang@continuent.com

Webinar Slides: MySQL HA/DR/Geo-Scale - High Noon #2: Galera Cluster

Recommandé

Recommandé

Contenu connexe

Tendances

Tendances (20)

Similaire à Webinar Slides: MySQL HA/DR/Geo-Scale - High Noon #2: Galera Cluster

Similaire à Webinar Slides: MySQL HA/DR/Geo-Scale - High Noon #2: Galera Cluster (20)

Plus de Continuent

Plus de Continuent (20)

Dernier

Dernier (20)

Webinar Slides: MySQL HA/DR/Geo-Scale - High Noon #2: Galera Cluster