Question on fly-proxy routing during partial region degradation: is routing purely based on health-check pass/fail plus concurrency/load, or does it do any latency-aware avoidance?
If a region is slow but still passes health checks, will the proxy continue routing traffic to it? And in that case, is manually stopping/cordoning machines in the degraded region the recommended way to force traffic away from it
Depending on what you mean by partial region degradation. In the case of something like the current sjc outage, since it is happening on our “edge” nodes, all traffic hitting the sjc edges (i.e. users near sjc) will be affected regardless of whether you have machines there or health checks; this is simply due to how the internet works and is outside of our control. However, for users not located near sjc, we do attempt to automatically route traffic away from affected nodes, and sometimes full regions. Removing or cordoning machines will also help in that case.
Noted, thanks - just added lax on top of sjc machines, seems to help with non-US users