I ran across this little gotcha recently.

Scenario: A service hosted in two regions. Each is fronted by either an ALB or ELB with an attached Autoscale group, that uses the LB healthcheck to determine instance health. A Route53 configuration balances trafic between the two. Route53 ‘Evaluate Target Health’ is set to yes and no healthcheck is attached.

Under the ELB, if the backend application fails in a region, the ELB will trigger termination of the application nodes. Route53 will consider the region unhealthy if all the backends are sick or unregistered and fail over to the remaining region.

Under the ALB, the same occurs except Route53 does not consider an empty ALB as unhealthy and will continue to send traffic to a region with no registered backends.

This is to a certain extent understandable, as ALBs allow attaching multiple target groups and its not immediately obvious what to do when there is a mix of statuses. I suspect the common case is that most ALBs have exactly one target group attached though and that could be used as the status, or allow Route53 to be bound to a specific target group.

The current workaround is to use a Route53 Healthcheck (at an additional $1+ per month per check) to have Route53 perform an application healthcheck against each origin.