Fast Master Failover without Data Loss

High Availability
14 April 11:30AM - 12:20PM @ Ballroom B

Experience level: 
50 minutes conference

Rate This Session

In this session, we will present about our enhancements and deployments for reducing MySQL server downtime dramatically at Facebook -- by fast master failover without data loss. There are a couple of well-known fast master failover solutions like mysqlfailover/MHA. In addition to that, we started using Semi-Synchronous replication to achieve failover without dataloss. To make failover work at scale (across datacenters, in timely fashion), we extended MySQL internals and developed some tools. We'll cover below topics. * What is Semi-Synchronous Replication * Loss-Less Semisync * mysqlbinlog speaking Semisync protocol * Master failure detection * Promoting a slave to the new master, with coordinating other slaves and binlogs * How GTID helps faster failover and recovery * Recovering a master without re-creating a whole instance


Database Engineer, Facebook
Yoshinori is a database engineer at Facebook. Before joining Facebook,
Yoshinori was a database and infrastructure architect at DeNA, living in Tokyo. Yoshinori's primary responsibility at DeNA is to make our database infrastructure more reliable, faster and more scalable. Before joining DeNA, Yoshinori worked at MySQL/Sun/Oracle as a lead consultant in APAC for four years. Yoshinori has written eight MySQL related technical books so far and has published technical articles about MySQL, Linux, and Java for a monthly database magazine since 2004.