Edge routing without drama
Most routing incidents are not exotic. They are a missing route, a wrong MTU, a stale DNS answer, or a rollback nobody tested.
Start with reachability, not theory
Before changing a path, confirm that the source can reach the destination on the exact protocol and port you care about. ICMP is a hint, not proof. A TCP handshake is better. A sustained transfer is better still.
Check MTU early
Tunnels and overlays quietly reduce the usable MTU. Small requests work, large responses stall, and the symptom looks like an application bug. Test with a payload large enough to expose it.
curl -o /dev/null -w '%{size_download} %{time_total}\n' https://example.net/5mb.bin
Make DNS boring
Use one resolver path, document it, and make sure the fallback does not leak somewhere unexpected. If DNS changes as part of the rollout, treat it as a routing change.
Keep a rollback
A change without a tested rollback is an outage waiting for a convenient moment. Keep the previous configuration, know the exact command to restore it, and verify that command before you need it.
If the first kilobyte works and the fifth megabyte does not, you have a path problem, not a content problem.