Cloud Developer Advanced Practice Exam: Hard Questions 2025

You are modernizing a global checkout service running on Cloud Run. It writes an order record to Cloud Spanner and then publishes an event for downstream processing. During regional network partitions, you observe occasional duplicate downstream charges and missing downstream events even though the order exists in Spanner. You must guarantee: (1) each successful order results in at-least-once event delivery, (2) downstream processing is effectively exactly-once per order, and (3) the API remains highly available. What architecture best meets these requirements?

A latency-sensitive API on Cloud Run (fully managed) reads from Memorystore (Redis) and falls back to a backend service when the cache misses. Under load tests, you see frequent Redis connection errors and sudden spikes in backend traffic, but no CPU saturation. You also notice many Cloud Run instance starts and stops. You need to stabilize cache connectivity and reduce backend thundering herd while keeping the service highly scalable. What is the best approach?

You operate a multi-tenant SaaS API on GKE. Tenants are isolated by namespace. A new compliance requirement mandates that no tenant can exfiltrate data to the public internet, but tenants must still reach Google APIs (Artifact Registry, Cloud Logging, Pub/Sub) and an internal shared service exposed via HTTPS on a private IP. You also need auditable enforcement at the platform level (not just app code). What design best satisfies the requirement?

Your team maintains a microservice that emits OpenTelemetry traces. In production, trace sampling must be adaptive: keep 100% of traces for requests that exceed 2 seconds or return 5xx, but sample only 1% of the rest to control cost. The service runs on Cloud Run and must not block request handling on telemetry export. Which implementation best meets these constraints?

A Cloud Run service processes Pub/Sub push requests. You notice occasional message redeliveries even though the handler returns 200 OK. Investigation shows the service sometimes exceeds the Pub/Sub push acknowledgment deadline during CPU spikes, and retries cause duplicate side effects in an external payment gateway that is not idempotent. You cannot change the gateway. What is the best design to prevent duplicate payments?

You are writing integration tests for a service that uses Firestore, Pub/Sub, and Cloud Storage. The CI pipeline runs in Cloud Build and must be deterministic and safe: tests must not interact with production data, and the pipeline must validate IAM and runtime configuration (service account permissions, VPC connector existence) before deployment. What is the best approach?

You deploy a new revision of a Cloud Run service using gradual rollout (traffic splitting). After shifting 10% traffic, error rates increase only for one specific customer tenant. Logs show requests for that tenant are routed inconsistently between revisions, causing authentication failures due to a per-revision in-memory session cache. You must keep gradual rollout but ensure tenant affinity to a single revision during the rollout window without introducing stateful sessions at the load balancer. What should you do?

A GKE deployment uses Workload Identity. A new release fails only in production: pods cannot pull a private image from Artifact Registry, but the same manifest works in staging. Nodes are healthy. Pod events show ImagePullBackOff with 403 errors. You must fix the issue with minimal blast radius and without granting broad permissions to the node service account. What is the best fix?

A Cloud Run service is fronted by an external HTTPS load balancer. You need blue/green deployments with fast rollback and the ability to route a subset of requests based on a cookie for internal testers, while keeping a single stable public URL. The service must remain serverless and you want to avoid managing custom proxy infrastructure. What is the best solution?

A data ingestion pipeline receives events from thousands of devices. Devices send JSON over HTTPS to an API. Requirements: handle bursts, maintain per-device ordering for downstream processing, and support replay of events for up to 7 days if a bug is found. You also need to monitor end-to-end lag and quickly identify which component is the bottleneck. What design best meets these requirements on Google Cloud?