Google Professional Machine Learning Engineer Question 384
Select 4Google Cloud PlatformYou are a Machine Learning Engineer tasked with deploying a trained machine learning model as a REST API for predictions. The model requires minimal preprocessing and should handle variable traffic loads efficiently. You decide to use Cloud Run for hosting this service. Which of the following are best practices to optimize the deployment on Cloud Run?
- A
Set the minimum number of instances to 0 to save costs when there is no traffic.
- B
Use a lightweight container image to reduce startup time and resource usage.
- C
Configure concurrency to 1 to ensure each request is handled in isolation.
- D
Enable Cloud Run’s autoscaling to handle spikes in traffic automatically.
- E
Deploy the model in the same region as the data it accesses to reduce latency.
Show answer and explanation
Correct answers: A, B, D, E
Explanation
When deploying ML models on Cloud Run, optimizing for cost, performance, and scalability is crucial. Setting the minimum number of instances to 0 saves costs during low or no traffic periods. Using lightweight container images reduces startup time and resource usage, improving scalability. Autoscaling ensures the service can handle traffic spikes efficiently. Additionally, deploying the service close to the data it accesses minimizes latency. While setting concurrency to 1 ensures isolation, it is not generally recommended as it can increase costs and reduce scalability unnecessarily.
- A. Correct.
Setting the minimum number of instances to 0 ensures that no resources are being used when there is no traffic, reducing costs. This is a recommended practice for cost efficiency in Cloud Run.
- B. Correct.
Using a lightweight container image reduces the time it takes to start a new instance and minimizes resource requirements. This helps improve scalability and responsiveness.
- C. Incorrect.
While setting concurrency to 1 can ensure isolation for each request, it is not always optimal for performance or cost. Cloud Run is designed to handle multiple concurrent requests efficiently, so this may unnecessarily increase resource usage.
- D. Correct.
Enabling Cloud Run’s autoscaling ensures that the service can automatically scale up or down based on traffic. This is essential for handling variable traffic loads efficiently.
- E. Correct.
Deploying the model in the same region as the data it accesses reduces network latency and improves performance. This is a best practice for any distributed system, including Cloud Run deployments.