A developer needs to deploy a containerized machine learning inference service that experiences unpredictable traffic patterns, with long periods of inactivity followed by sudden bursts of requests. The service must scale to zero to minimize costs during idle times. Which Google Cloud compute option is the most cost-effective and appropriate for this scenario?

Why the other options are wrong

• A. Compute Engine instances, even with custom images, incur costs when idle and require manual configuration for scaling, which is not cost-effective for unpredictable, bursty workloads that need to scale to zero.
• C. Cloud Functions are serverless, but they are designed for event-driven functions, not typically for containerized services that might require more complex runtime environments or custom binaries beyond simple functions.
• D. GKE Autopilot manages the underlying infrastructure, but it's designed for continuous workloads or applications that need more control over infrastructure, and it doesn't scale to zero in the same cost-effective manner as Cloud Run for individual services.