A web service performs requests that can take 90 seconds. During planned instance removal, requests are cut off because the ALB target group deregistration delay is 30 seconds. Instances remain alive for several minutes after deregistration. Which change most directly addresses the interrupted in-flight requests?

1. Increase only the ALB idle timeout while keeping the 30-second deregistration delay.
2. Increase the target group deregistration delay to accommodate the expected in-flight request duration.
3. Raise the unhealthy threshold but keep deregistration delay at 30 seconds for planned removals.
4. Set the deregistration delay to zero to make draining more immediate.