SAA-C03 Practice Question 96: ALB Deregistration Delay Scale in
AWS Certified Solutions Architect – Associate (SAA-C03) · Test 10 · question 6 of 10
A web application runs in an Auto Scaling group behind an Application Load Balancer. Users uploading large files, which take up to 90 seconds, report failed uploads whenever the group scales in. An engineer finds that the target group's deregistration delay was changed from the default to 0 seconds to make scale-in faster.
What should the engineer do to stop the failed uploads?
Show answer and explanation
Answer:
When Auto Scaling removes an instance, Elastic Load Balancing deregisters the target and stops sending it new requests, then waits for the deregistration delay so that in-flight requests can finish before the connection is closed. With the delay set to 0 seconds, the ALB closes the connections at once, so an upload in progress is cut off. Setting the delay to more than the longest request (for example 120 seconds for uploads that take up to 90) lets those requests complete while new requests go to the remaining instances. The default is 300 seconds.
Option A speeds up detection of failed instances, but these instances are healthy and are being removed on purpose, so the health check plays no part.
Option B changes how traffic is spread across Availability Zones. It does nothing about what happens to requests already in flight on a target that is being removed.
Option D keeps a user on the same instance, which makes no difference when that instance is the one being terminated. It could even increase the number of users affected by one scale-in.
Reference: Target groups for your Application Load Balancers