A media company runs a stateless web tier on an Amazon EC2 Auto Scaling group behind an Application Load Balancer in a single Region. During a major launch the group tried to scale out but stalled at a fraction of its desired capacity, and customers saw errors while healthy demand went unserved. Investigation found the account had hit its Regional On-Demand vCPU service quota for the instance family, and separately the group was pinned to one Availability Zone so it could not place instances elsewhere. The team must ensure scaling events are no longer capped by these limits and that the fleet survives the loss of one Availability Zone, with the least operational overhead. Which TWO actions together MOST effectively resolve the scaling and resilience limits? Select TWO.
- ARequest a Service Quotas increase for the Regional On-Demand vCPU limit for that instance family and set a CloudWatch alarm on the quota utilisation so the team is warned before the fleet approaches the ceiling again. Correct
- BReconfigure the Auto Scaling group to span at least three Availability Zones across multiple subnets so it can place and balance instances in more than one zone and continue serving if a single Availability Zone is lost. Correct
- CLower the scaling cooldown period and raise the group's desired capacity to a fixed high number so enough instances are always running to absorb launch traffic without waiting for a scaling policy to react.
- DSwitch the group to use only Spot Instances for the launch so the additional capacity is drawn from the Spot pool, which is billed separately and therefore not counted against the Regional On-Demand vCPU quota during the surge.
- EMove the web tier behind Amazon CloudFront so cached responses absorb the launch traffic, reducing the number of instances the Auto Scaling group must launch and therefore the vCPU the fleet consumes.
Why A is correct: Raising the On-Demand vCPU quota removes the hard ceiling that capped the scale-out, and alarming on quota utilisation gives early warning so the fleet is not throttled by the same limit during the next launch.
Why B is correct: Spreading the group across multiple zones lets Auto Scaling place capacity beyond the single zone and keeps the fleet serving when one zone fails, removing the zonal single point of failure.
Why C is wrong: Over-provisioning a fixed high desired capacity is tempting but it still cannot exceed the vCPU quota, so the scale-out is capped at the same ceiling while idle cost rises and the zonal risk is untouched.
Why D is wrong: Spot capacity can be interrupted at any time and is unsuitable as the sole capacity for a customer-facing launch, and relying on it to dodge the quota leaves the resilience gap and adds interruption risk.
Why E is wrong: Edge caching can reduce some load but does nothing for the underlying vCPU quota ceiling or the single-zone placement, so a dynamic surge still hits the same scaling and resilience limits.