An EC2 Auto Scaling group sits behind an Application Load Balancer. Some instances keep running and pass their EC2 status checks, but the application process inside hangs and stops answering, so the ALB target health check marks the target unhealthy while the group leaves the instance in service. Separately, the application needs about three minutes to start after launch, and newly added instances are sometimes terminated as unhealthy before they finish booting. The team needs the group to replace genuinely hung instances automatically while giving fresh launches enough time to start. Which two configuration changes together meet this requirement? (Select TWO.)
- ASet the Auto Scaling group health check type to ELB so the group uses the load balancer target health status, not only EC2 status checks, to decide when to replace an instance. Correct
- BConfigure a health check grace period on the Auto Scaling group long enough to cover the application start time so new instances are not evaluated until after they have booted. Correct
- CLeave the Auto Scaling group health check type as EC2 so the group acts only when the underlying instance fails its system or instance status checks.
- DAttach a target tracking scaling policy on average CPU utilisation so the group adds capacity whenever the hung instances raise the fleet CPU average.
- EEnable an Application Load Balancer connection draining timeout on the target group so hung instances finish in-flight requests before being removed.
Why A is correct: With the ELB health check type the group treats an unhealthy ALB target as an unhealthy instance and replaces the hung application automatically.
Why B is correct: The grace period suspends health evaluation while a new instance starts, stopping the group from terminating instances that are still warming up.
Why C is wrong: EC2 status checks pass while the application hangs, so this is tempting but it never detects the failed process and the instance stays in service.
Why D is wrong: Target tracking changes desired capacity for load, but it neither detects an unhealthy target nor protects a booting instance from premature termination.
Why E is wrong: Deregistration delay only delays closing connections on instances already being removed; it does not trigger replacement or grant boot time to new launches.