Global Traffic Management

An application can run in several regions and still fail its users. A healthy region may be far away. A failed region may remain in cached DNS answers. A health check may test the wrong path. A sudden failover may send more traffic than the surviving region can handle.

Global traffic management addresses these problems. It selects an application location for each user or resolver. The selection can reflect availability, latency, geography, capacity, rollout policy, or a fixed priority.

The key idea is separation. A traffic manager makes a placement decision. The selected location still needs a local load balancer, healthy application instances, sufficient capacity, and consistent data.

The control loop

A useful mental model has five stages:

1. Observe endpoint health and performance from relevant locations.
2. Evaluate a policy against the eligible locations.
3. Publish or enforce a traffic decision.
4. Observe the result from the user’s path.
5. Adjust the policy or endpoint set when evidence changes.

The loop is only as good as its signals. A TCP connection proves that a port accepted a connection. An HTTP check can test a path and expected response. Neither test automatically proves that every dependency or user journey works.

Three steering planes

DNS steering