by peter » Wed Jul 02, 2014 8:51 am
Automatic switching from the client to different data centers (clusters) is very hard to do, and can easily create severe problems. It is very tricky to do.
At internet providers like Yahoo!, Google, etc., a common practice is to roll out a mostly automatic way of doing this but the exact switch is usually triggered manually. This is the safest way.
by bbulkow » Thu Jul 31, 2014 2:30 pm
We generally find that automatic switching is rarely the correct policy.
Our experience with Aerospike is if a cluster in a data center has failed, the datacenter is very messed up, and applications running in that datacenter will have a wide variety of problems. You’ll be having multiple switch, router, and power failures. In that case, you’re on your way to a business continuity fail-over anyway.
Saturating your long-haul links with small database requests is not a good idea when you’ve got these kinds of failures happening.
As Peter mentions, if that is desired - perhaps some small requests - it can easily be done at the application level.