Will the current beta version of HA, and/or future versions of controller HA support having controllers spread across multiple cloud regions?
Questions around this in considerations.
HA Latency requirements?
How does HA react if three regions are used and 2 of them go down, how will split brain make decisions?
Yes, this is inherently how it is built. We use Raft as a consensus algorithm under the hood to achieve it.
I would suggest reading these:
If a region or more go down, the remaining controllers will elect a leader (itself if needed).
Thank you for the quick reply, I have been reading through those already, so the only limiting factor on the regional disbursement would be quorum?
Yes, quorum is the limiting factory. There are three classes of operations:
- Updates to model (adding/changing services, identities, routers) etc
This needs a quorum to work
- Reads from model
Reads will use what the controller you are accessing has in local storage, so you only need one controller up, but it may also be behind.
- Using services
This should work as long as any controller is up, though may be slower if an SDK or router has to work with a controller that's further away than usual.
Cheers,
Paul
2 Likes