Load Balancing and Auto Scaling Questions and Answers

Answer: C

Both are true and the reason gives both objections, and it also means the loss of one backend loses the sessions it was holding, so a restart during a rolling deploy logs those users out. Every deployment then becomes visible to users who were midway through a task.

Answer: D

Scheduled scaling is a separate mechanism for known patterns. Choosing the right metric matters, since processor use is a poor proxy for load in a memory or input output bound service. The metric should be whichever resource actually saturates first under load.

Answer: A

The first two are correct. The third is wrong, and scaling a memory bound service on processor use produces a system that never scales when it actually needs to. Queue depth is often the honest signal where processor use is a misleading one.

Answer: B

Without it the group reacts to a metric that has not yet responded and overshoots. Flapping between scaling out and in is the usual symptom of too short a cooldown. Each cycle costs launch time and money without ever settling at the right size.

Answer: C

Both are true and the reason gives the mechanism, and the oscillation costs money and destabilises the service at the same time, and it is entirely avoidable by widening the thresholds. A gap of twenty points between scale out and scale in is a common starting value.

Answer: D

Scale out and scale in thresholds that are too close together cause it, as does too short a cooldown. Widening the gap between them is the usual first remedy. Lengthening the evaluation window before a decision has much the same effect.

Answer: A

Scaling begun when load is already at the limit arrives too late if warm-up takes minutes. Scaling on a leading indicator rather than a lagging one is the usual answer. Otherwise every scaling decision arrives one warm-up period behind the traffic.

Answer: B

Both are true and the reason gives the timing problem, and queue depth or request rate often lead processor saturation by exactly the interval needed. Choosing the metric well matters more than tuning the thresholds around it.

Answer: C

Reacting after the rise begins cannot beat a warm-up longer than the rise itself. Pre-warming, predictive scaling or a larger baseline are the available answers. None of them is free, so the choice depends on how costly a brief overload is.

Answer: D

A longer cooldown makes the group slower to respond, which is the opposite of what a spike needs. The other three all bring capacity forward in time. A spike is won or lost on how early capacity starts, not on how fast it scales.