I just took over this matter, before seen Galera Cluster method, and before and NDB scheme (though many people say that bad maintenance), but previous outside the DBA would only make decision from, and then manually switch (fault, last week is not a high availability), he said have been switch with script switching may occur misjudgment, before also look at said ali mycat scheme in the QQ Group, I think this is mainly in order to improve the performance, I think the first is the solution stability at present, this year I have seen the MySQl 5.7 official MGR (MySQl Group Replication), don't know what is this hole, would like to ask you is how to deal with the big company,

Demand in general should come from standard, I this small company has been a demand uncertainty, users data volume are small, big cake, because self-built rooms (tube) to other department itself operations staff checking not in time, so, want to use software strategy to compensate for priority, so the stability on the premise of ensuring stability, performance, the higher the better,

Currently use hardware configuration is E5-2650, 8 g memory, disk with Shared storage, before each project database independent, now going to database are put together in order to save the cost, so more is considered stable,

CodePudding user response:

Generally is a master-slave, main hang up after manual switch, complicated the cost is high, and, to the maintenance and the higher technical requirements,,,

CodePudding user response:

Can use double GTID + keeplive Lord made automatic switching.

CodePudding user response:

Before the Lord, and then make switch KeepAlived

CodePudding user response:

reference oyljerry reply: 3/f

master-slave first, then make switch KeepAlived

MHA on?

CodePudding user response:

Mha is very by spectrum,

CodePudding user response:

Strictly speaking, the master-slave mysql is unreliable, it does not provide a complete master-slave synchronization, currently only support asynchronous and synchronous
So in a strict sense, automatic switch there is a certain risk

CodePudding user response:

Master-slave + keepalive, MHA, PXC and so on

CodePudding user response:

The master-slave + keepalive is very by spectrum

CodePudding user response:

Master slave is the lowest cost, high availability at least need only two nodes,
Galera Cluster the high cost, but can realize more, need at least three nodes, if there are more than the main requirement, use Galera Cluster,
NBD calculate, the cost is very high,

CodePudding user response:

With q, if a web site need to repeatedly query mysql, obtain a variable's value, so how to realize the mysql link at a time, multiple use instantiation objects?

CodePudding user response:

MGR can not automatically switch? I don't understand the MySQL ha

CodePudding user response:

The heartbeat detection the most practical, to hang up before you start,,,

CodePudding user response:

Don't MGR and mysql cluster, will pit to life you doubt,
My present high availability is to use mysqlfailover and proxysql do,
Two from a main structure,
Premise is in GTID mode,

CodePudding user response:

The DBA treatment high