A company is running a critical web application on EC2 instances behind an Application Load Balancer. During a recent deployment, a new version of the application was released, and within minutes, the error rate spiked. The operations team rolled back the deployment, but the error rate remained high. The team suspects that the ALB is still routing traffic to unhealthy instances. Which step should the team take to immediately stop routing traffic to the problematic instances?
Deregistering immediately stops routing traffic to those instances.
Why this answer
Deregistering the problematic EC2 instances from the target group immediately stops the Application Load Balancer from routing any new traffic to them. The ALB performs health checks on registered targets, but if an instance is unhealthy, it still remains in the target group and can receive traffic if the health check logic is misconfigured or if the instance is in a 'draining' state. Deregistration forces the ALB to complete any in-flight requests and then remove the instance from the routing table, providing an immediate stop to traffic.
Exam trap
The trap here is that candidates often confuse 'health check failure' with 'immediate traffic removal,' not realizing that health checks only update the instance's status but do not stop traffic until the instance is explicitly deregistered or the health check threshold is met, which introduces a delay.
How to eliminate wrong answers
Option A is wrong because modifying the ALB listener rules to point to a different target group does not stop traffic to the problematic instances; it only changes the routing target for new requests, but the original target group and its unhealthy instances remain active and could still receive traffic if the listener rules are not fully replaced or if there are overlapping rules. Option B is wrong because modifying the health check interval to 5 seconds and the unhealthy threshold to 2 would only accelerate the detection of unhealthy instances, but it does not immediately stop traffic; the ALB would still route requests to instances until they are marked unhealthy, which could take up to 10 seconds (2 checks × 5 seconds), and during that time errors persist. Option C is wrong because stopping the EC2 instances from the AWS Management Console will terminate the instances, but this action is disruptive and does not leverage the ALB's built-in traffic control; it also does not gracefully drain connections, potentially causing dropped requests, and it is not the recommended method for removing instances from load balancing.