Google Professional Machine Learning Engineer Question 376
Select 2Google Cloud PlatformYour team is tasked with hosting an MLFlow-based third-party machine learning pipeline on Google Cloud. The pipeline must integrate with custom storage backends for artifact tracking, leverage managed services to minimize operational overhead, and provide a scalable deployment solution. Which combination of Google Cloud services would best meet these requirements?
- A
Use Google Cloud Storage as the artifact store and deploy MLFlow on Google Kubernetes Engine (GKE).
- B
Deploy MLFlow on a Compute Engine instance and use Cloud SQL as the backend for metadata tracking.
- C
Leverage Vertex AI Pipelines for managing the MLFlow pipeline and use BigQuery for artifact storage.
- D
Host MLFlow on Google Kubernetes Engine (GKE) and configure Cloud SQL as the backend for metadata storage.
- E
Use App Engine to host MLFlow and integrate Firestore as the metadata storage backend.
Show answer and explanation
Correct answers: A, D
Explanation
Hosting MLFlow on Google Kubernetes Engine (GKE) ensures scalability and flexibility in managing the MLFlow server, and using managed services like Google Cloud Storage for artifact storage and Cloud SQL for metadata storage reduces operational overhead. These choices align with best practices for integrating third-party pipelines on Google Cloud.
- A. Correct.
Using Google Cloud Storage as the artifact store ensures scalability and durability for artifacts, while deploying MLFlow on GKE provides flexibility in managing the MLFlow server and its dependencies.
- B. Incorrect.
Deploying MLFlow on a Compute Engine instance is an option, but it does not leverage managed services effectively, leading to higher operational overhead.
- C. Incorrect.
Vertex AI Pipelines is not directly compatible with MLFlow as it is a separate orchestration tool, and BigQuery is not designed for artifact storage, making this option unsuitable.
- D. Correct.
Hosting MLFlow on GKE ensures scalability and flexibility, while using Cloud SQL as the backend provides a managed solution for metadata storage, reducing operational burden.
- E. Incorrect.
App Engine is not well-suited for hosting MLFlow due to its limitations in running long-lived processes and custom dependencies, and Firestore is not an ideal choice for metadata storage in this context.