using continuent tungsten to survive mysql failures...using continuent tungsten to survive mysql...
TRANSCRIPT
©Continuent 2013
Using Continuent Tungstento Survive MySQL Failures
Je! Mace, Director of Professional Services
©Continuent 2013
Introduction
• Continuent Tungsten manages large and small MySQL deployments
• Automatic local failover
• Managed failover between datacenters
• Connectivity options for all applications
2
©Continuent 2013 3
1
Our Customers
©Continuent 2013
What’s Your Plan?
• Local failover
• Disaster recovery
• Multiple datacenters
• Erroneous SQL statements
4
©Continuent 2013
Sources of Instability
• Network instability
• DNS outages
• Hardware degradation and failure
• Memory management
• System administration
5
©Continuent 2013
Keeping It Local
6
©Continuent 2013
What is a Dataservice?
7
• Encapsulate data availability as a service
• React to changes in system status with automatic failover
• Role based access to database servers
• Provide connectivity to the master
• Optionally connect to a slave for reads
©Continuent 2013 8
MasterSlave Slave
App Logic
Tungsten Connector
Replicator Replicator Replicator
App Logic
Tungsten Connector
Manager Manager Manager
Monitoring and control
Monitoring and control
db2 db1 db3
Data Service: sj
©Continuent 2013
Keeping It Local
9
©Continuent 2013
Keeping It Local
10
©Continuent 2013
Keeping It Local
11
©Continuent 2013
Keeping It Local
12
• Di!erent connectivity options
• Limit writes to a single server at any time
• Splitting writes introduces inconsistency
• Continuent Tungsten automatically promotes a master and sends tra"c there
©Continuent 2013
Data Overload
• Promote sharding to support extra large data sets
• Support parallel replication in Q4 2013
• Single customer is supporting ~300TB across 30 Continuent Tungsten dataservices
13
©Continuent 2013
Going Global
14
©Continuent 2013
Going Global
15
©Continuent 2013
Going Global
16
©Continuent 2013
Going Global
17
©Continuent 2013
Going Global
18
©Continuent 2013
Going Global
19
• Primary/DR replication allows for backup datacenter without application changes
• True multi-master replication allows for large distributed systems
• Must account for eventual consistency in the application
• Local failover can reduce the e!ects of a single server outage on the global system
©Continuent 2013
When Replication Works Too Well
• Replication applies all events, even the ones you don’t want
• Delayed replication
• Filtered replication
• Point In Time Recovery (PITR)
20
©Continuent 2013
Questions?
21