I'm trying to map this to a mental model I get, like a Kubernetes ingress controller or a global load balancer. The docs talk about "nearest" and "best" but the logic is fuzzy.
From my packet captures and some support tickets, I've seen it's a mix of:
* Your physical location (IP geolocation)
* The service you're trying to reach (SaaS app vs internal corporate app)
* Current data center load and health (their telemetry)
* Pre-defined policies you set (like steering certain apps to a specific DC)
It's not just simple BGP anycast. The client agent (or network connector) seems to get a sorted list of gateways from a controller. The real question is the weighting of those factors. Is health check data from my endpoint part of it?
Example: If my user in London tries to reach Salesforce, does it:
1. Always go through the London PoP?
2. Go to the PoP with the best latency to Salesforce's backend at that moment?
3. Or just follow a static mapping they have internally?
Anyone reverse-engineered this or got concrete log examples showing the decision?
Ship it, but test it first