26 Jul
2015
26 Jul
'15
3:53 p.m.
For what it's worth, it might be worth assigning a /24 and setting that up as an
Anycast subnet so users can just connect to a single IP which is advertised from multiple
locations. Setup local health-check scripts that will cause a BGP failure if any of the
local services (Wiki / Portal / API) fail.
Use the Anycast IP as the primary inbound server connection with some other public IP (or
VPN) on the backside for database replication and management.
Alternatively there are a few load balancer vendors who do offer global load balancing
using a combination of DNS and translation to allow physically diverse servers to answer a
single URL, though they are probably a bit too heavy weight for AMPR and a bit too
expensive.
--
Will
On 7/26/15 4:12 AM, Chris wrote:
I am working on a HA solution for the portal with
real-time replicated systems at separate geographical locations.
Failover will be via short TTL DNS records but there would be no issue with scripts using
the API polling one server then the other if they got no reply from the first.