GCP Professional Cloud Architect Practice Question

Your fintech startup exposes a latency-sensitive gRPC microservice to mobile clients worldwide. The service currently runs in a single zonal GKE Standard cluster in us-central1. European users experience 600 ms p95 latency, exceeding the 250 ms SLO. Traffic surges unpredictably, and the small platform team wants to minimize operational toil while containing cost. Which redesign best aligns with the Performance Optimization pillar of the Google Cloud Well-Architected Framework?

Create regional GKE Standard clusters in us-central1 and europe-west1, enable multi-cluster Ingress with container-native load balancing, and direct users using Cloud DNS latency-based policies.
Repackage the service for Cloud Run, deploy it in us-central1 and europe-west1 with a minimum instance of 1 per region, and front both regions with a global external HTTP(S) load balancer backed by serverless NEGs.
Add additional zonal GKE Standard clusters in europe-west1 and use Traffic Director with DNS round-robin records to split traffic across clusters.
Move the service to a regional GKE Autopilot cluster in us-central1 and expose it with a global external TCP proxy load balancer protected by Cloud Armor.

Report Issue

Answer Description

Deploying the service as a Cloud Run workload in multiple regions and placing the instances behind a global external HTTP(S) load balancer backed by serverless NEGs gives several performance benefits. Cloud Run automatically scales horizontally to meet bursty demand with no cluster management, lowering operational overhead. Setting at least one minimum instance per region keeps a warm container ready, reducing cold-start latency. Because the global load balancer uses anycast, it routes each client to the closest healthy backend, cutting round-trip time for European users while providing built-in fail-over across regions. The other options either remain single-region (thus not improving trans-Atlantic latency) or require the team to operate and tune multiple GKE clusters, increasing complexity and delaying reaction to spikes, which conflicts with the Performance Optimization pillar's guidance to use managed, auto-scaling services whenever possible.

Ask Bash

Bash is our AI bot, trained to help you pass your exam. AI Generated Content may display inaccurate information, always double-check anything important.