Open JeremyYu2018 opened 5 years ago
before promote one ideal slave, the promoted slave could apply the unfinished relay-logs(do not stop slave)
I'm expecting a PR from a contributor which does exactly that.
and then get diff binlogs from dead master,
There is no plan to make that happen.
in all promoting-one-slave-process, could orchestrator give a proper time to ideal slave to apply apply the unfinished relay logs
Repeats the first question. I'm expecting a PR.
@shlomi-noach if we can design an agent to get diff binlog between Slave IO THREAD and Master Executed GTID, that would be excellent! we could make consistent failover(no data loss) when not using semi-sync replication。just like MHA, still no plan?
There is a scenrio that where
one-master/two-direct-slaves
asynchronous mysql topology, where a master instance found dead but host is fine,stop slave
), and then get diff binlogs from dead master, when all is done, do 'reset slave all' thing。orchestrator
give a proper time to ideal slave to apply apply the unfinished relay logs, if not, this could result in data loss. I found variable 'FailMasterPromotionIfSQLThreadNotUpToDate' doesn't satisfy this.