Which solution will meet these requirements?

Explanations:

The round robin routing algorithm does not consider the current load of instances and will continue to forward requests evenly, which may overload some instances. Relying solely on RequestCountPerTarget and ActiveConnectionCount does not directly address the issue of preventing overloaded instances from receiving new requests.

The least outstanding requests algorithm is designed to send new requests to instances with the fewest outstanding requests. By using the RequestCountPerTarget and ActiveConnectionCount metrics, this approach ensures that traffic is distributed based on the current load, effectively preventing overloaded instances from receiving additional requests.

Similar to option A, the round robin algorithm does not account for the state of each instance under load. Using TargetResponseTime as a metric does not help manage the load effectively, as it may still send requests to overloaded instances without regard for their actual performance at that moment.

While the least outstanding requests algorithm is suitable for load balancing, using RequestCount and TargetResponseTime does not provide the right metrics for decision-making in this context. TargetResponseTime may not accurately reflect the instance’s ability to handle more requests, thus not effectively addressing the overload issue.