A Compute Engine autoscaling policy should trigger when total memory usage exceeds 80%. You installed the Cloud Monitoring agent but scaling does not occur under high memory load. Which change fixes the policy?

The Cloud Monitoring agent reports memory usage with a 'state' label, which can be 'used', 'buffered', 'cached', or 'free'. To trigger autoscaling based on total memory usage exceeding 80%, you need to specifically target the 'used' state. The default autoscaling policy might not correctly aggregate these different states. Changing the metric filter to metric.label.state = 'used' ensures the autoscaler monitors only the actively used memory. Changing the Target type to DELTAPERMINUTE is incorrect because memory usage is a gauge metric, not a delta. Changing the metric identifier to agent.googleapis.com/memory/bytesused is unnecessary as the base metric is likely correct, but the specific state needs filtering. Filtering for 'free' memory and setting utilization to 20% would scale down when free memory is low, which is the opposite of the desired behavior for high memory usage.