Cloud Run: Transitioning from Min-Instances to Scale-to-Zero for Cost Reduction

A low-traffic workload is currently configured with a minimum instance count greater than zero to prevent cold-start latency. To reduce monthly spend, the goal is to transition to a scale-to-zero configuration where CPU is only allocated during request processing.

The primary constraint is the trade-off between the continuous billing of warm instances and the potential latency spikes experienced by the first user after a period of inactivity. While the default billing model only charges for vCPU and memory during request processing, the impact of container initialization varies by runtime and image size.

Technical Uncertainties

• Does the cost saving of removing minimum instances outweigh the business impact of cold-start latency for this specific traffic pattern?
• How does increasing the concurrency limit affect the frequency of cold starts during bursty, low-volume traffic?