On 09/05/2018 12:04 PM, Thomas Poty wrote:
> I want to get :
> Automatic failover (promoting a slave)
> Automatic Routing traffic to master
> Fencing in case of node failure.
Why would you want to fence a node that's already failed? -- You want to
fence off the master during the failover so it does not accept any more
DB updates.
Anyway, here's how we run:
- we provide services to customers; customer-facing service access
points is where we want to reduce downtime.
- We buy decent quality hardware with redundant disks and power
circuits. 99.9999% of the downtime is maintenance reboots.
- We carefully partition or services so that most of them require
read-only access. Those run off slaves.
- The master is not even visible to the customers. Its downtime does not
affect service availability.
Obviously, this works because we can run read-only. If you need
read-write on a public-facing database, the above won't work for you.
--
Dimitri Maziuk
Programmer/sysadmin
BioMagResBank, UW-Madison -- http://www.bmrb.wisc.edu