Question

An e-commerce company runs a stateless web tier on a Compute Engine managed instance group (MIG). Traffic follows a highly predictable weekday pattern: near-idle overnight, a steady ramp starting at 07:00, and a large sustained peak between 09:00 and 18:00. The default CPU-based autoscaler reacts too slowly, causing latency spikes each morning as it scrambles to add instances, and it also keeps a few extra instances running overnight 'just in case.' The team wants to eliminate the morning latency and reduce wasted overnight capacity without over-provisioning. What should they do?

Accepted Answer

Add scheduled autoscaling to the MIG to pre-scale minimum capacity before 07:00 and lower the minimum overnight, keeping the CPU-based scaling policy for unexpected variation.

Answer

Replace the autoscaler with a fixed instance count sized for the daily peak to guarantee capacity is always available.

Answer

Lower the CPU utilization target on the autoscaler so it adds instances sooner and reduce the cool-down period to near zero.

Answer

Switch the MIG instances to Spot VMs to reduce overnight cost and rely on the autoscaler to replace preempted instances during peak hours.