Production Compute Engine service drops requests under heavy load because a single application process is saturating CPU and autoscaling has reached its max; no other bottlenecks are evident. To restore service quickly, what do you recommend?

The correct answer is to increase the maximum number of instances in the autoscaling group. The problem states that the service is dropping requests under heavy load, and autoscaling has reached its maximum. This means the existing instances cannot handle the current demand, and the autoscaler is prevented from adding more capacity. Increasing the maximum allows the autoscaler to provision more instances, distributing the load and restoring service. Changing the autoscaling metric to memory usage (agent.googleapis.com/memory/percentused) is incorrect because the problem explicitly states CPU saturation is the issue, not memory. Restarting instances or application processes might temporarily alleviate the problem for individual instances but won't address the overall capacity shortage under sustained heavy load, and could even worsen the situation by causing more downtime.