How failover works for unhealthy endpoints

If there are no healthy endpoints in an endpoint group that have a weight greater than zero, Global Accelerator tries to fail over to a healthy endpoint with a weight greater than zero in another endpoint group. Note that for this failover, Global Accelerator ignores the traffic dial setting. So if, for example, an endpoint group has a traffic dial set to zero, Global Accelerator still includes that endpoint group in the failover attempt.

If Global Accelerator doesn't find a healthy endpoint with a weight greater than zero after trying the three closest endpoint groups (that is, AWS Regions), it routes traffic to a random endpoint in the endpoint group that is closest to the client. That is, it fails open.

Note the following:

For example, let's say your configuration has two endpoints, one healthy and one unhealthy, and you've set the weight for each of them to be greater than zero. In this case, Global Accelerator routes traffic to the healthy endpoint. However, now say you set the weight of the only healthy endpoint to zero. Global Accelerator then tries three additional endpoint groups to find a healthy endpoint with a weight greater than zero. If it doesn't find one, Global Accelerator routes traffic to a random endpoint in the endpoint group that is closest to the client.

When recovery occurs, that is, Regions are healthy again, Global Accelerator returns to regular routing behavior. This means that, typically, routing will start back to healthy endpoints with traffic dials that aren't set to zero in about 30 seconds or so. However, note that established active connections are not moved. They continue to route to the zero weight Region until the connection is reset by the client or the server, or until the client makes a new connection.