PCD Designing highly scalable, available, and reliable cloud-native applications Practice Test 12 — 15 Questions

Question 1

A company is designing a globally distributed application on Google Cloud. They need to ensure that user data is stored close to users for low latency and that the data is highly available across regions. The application requires strong consistency for writes. Which storage solution should they use?

Choose an answer to begin — your selection is scored in the full session.

Accepted Answer

Cloud Spanner with a multi-region configuration.. Cloud Spanner is the only Google Cloud database that provides strong consistency and global distribution with multi-region configurations. It ensures low-latency access for users worldwide while maintaining ACID transactions, making it the correct choice for this scenario.

Answer

Firestore in Datastore mode with multi-region replication.

Answer

Cloud SQL with cross-region read replicas.

Answer

Cloud Bigtable with a multi-cluster configuration.

Question 2

A developer is building a microservices application on Google Kubernetes Engine (GKE). They want to ensure that each service can scale independently based on CPU utilization and that the cluster can automatically add nodes when resource demand increases. Which GKE feature should they use?

Accepted Answer

Horizontal Pod Autoscaler (HPA) combined with Cluster Autoscaler.. Horizontal Pod Autoscaler scales the number of pod replicas based on metrics like CPU utilization, ensuring each service scales independently. Cluster Autoscaler adds or removes nodes in the node pool when pods cannot be scheduled or when nodes are underutilized. This combination provides both pod-level and node-level automatic scaling, which is exactly what the scenario requires.

Answer

Istio service mesh with automatic sidecar injection.

Answer

Pod Disruption Budget (PDB) combined with node pool auto-scaling.

Answer

Vertical Pod Autoscaler (VPA) combined with node auto-provisioning.

Question 3

A media company runs a video transcoding pipeline where each incoming upload must be processed exactly once, and the order of processing for a given video's segments matters. They are using Pub/Sub and want to guarantee that messages for the same video are delivered in order and processed sequentially. Which Pub/Sub feature should they enable to meet this requirement?

Accepted Answer

Message ordering using ordering keys. The need is to guarantee that messages for the same video are processed in order. Pub/Sub's message ordering feature, enabled by setting an ordering key, ensures that all messages with the same key are delivered in publish order. This directly meets the scenario's requirement without requiring custom sequencing logic in the application.

Answer

Push subscriptions with flow control

Answer

Exactly-once delivery with a dead-letter topic

Answer

Message retention duration set to 7 days

Question 4

A media company ingests millions of small event records per minute into Cloud Bigtable from a Dataflow pipeline. The pipeline uses row keys formatted as the event timestamp in reverse order followed by the device ID, and writes are evenly distributed. After several weeks, scans for a specific device ID become very slow, and single-row reads by timestamp are fast. They need to query by device ID efficiently while keeping write throughput high. What should they change?

Accepted Answer

Change the row key to device ID followed by the reverse timestamp, and enable a column family with a single version.. Bigtable stores rows sorted lexicographically by row key, and range scans are efficient only when the queried attribute is the leading part of the key. Placing the device ID first makes each device's events contiguous, so device queries become narrow range scans, and appending the reverse timestamp keeps recent events first. High-cardinality device IDs still distribute writes evenly across tablets, preserving ingest throughput.

Answer

Keep the existing row key and add a secondary index in a separate Bigtable table keyed by device ID.

Answer

Change the row key to a salted hash of the device ID followed by the reverse timestamp.

Answer

Keep the existing row key and switch reads to a Cloud Bigtable SQL query that filters on the device ID column.

Question 5

A company runs a stateful application on Compute Engine that uses a single persistent disk for data storage. They need to ensure that the application can survive a zonal failure without data loss and with minimal downtime. The application cannot be modified to use a distributed database. Which approach should they take?

Accepted Answer

Create a regional persistent disk and attach it to a VM in the primary zone, with a standby VM in another zone ready to attach the disk on failover.. A regional persistent disk replicates data synchronously across two zones, so if the primary zone fails, the disk can be attached to a VM in the other zone. This provides automatic data redundancy and allows the application to continue with minimal downtime, without requiring changes to the application itself. Other options either involve data loss or do not protect against zone failure.

Answer

Use a standard persistent disk and create snapshot schedules to back up data every hour.

Answer

Replicate the persistent disk to a disk in another zone using a custom script that runs periodically.

Answer

Attach multiple persistent disks to the VM using RAID 0 for increased performance and redundancy.

Question 6

A startup is deploying a stateless web front end on Compute Engine. They expect traffic to vary significantly throughout the day and want the deployment to automatically add and remove VM instances based on demand, while ensuring that instances are replaced automatically if they fail a health check. Which Compute Engine feature should they use?

Accepted Answer

A managed instance group with an autoscaler and an HTTP(S) health check.. A managed instance group with an autoscaler and an HTTP(S) health check provides both horizontal scaling based on demand and automatic replacement of unhealthy instances. It is the standard Compute Engine pattern for stateless, variable-traffic workloads. Fixed-size groups, single instances, and preemptible VMs either lack scaling, lack reliability, or both.

Answer

A managed instance group with a fixed size of ten instances and no autoscaler.

Answer

A zonal managed instance group with preemptible VMs and a network load balancer.

Answer

A single Compute Engine instance with a startup script that monitors CPU and restarts the instance when utilization is high.

Question 7

A financial services company is deploying a stateful application on Google Kubernetes Engine (GKE) that requires each pod to have a stable network identity and persistent storage that survives pod rescheduling. The application also needs to scale horizontally. Which Kubernetes resource should they use to meet these requirements?

Accepted Answer

StatefulSet with a headless Service. StatefulSets are designed for stateful applications, providing stable pod identities, ordered deployment and scaling, and persistent storage through PersistentVolumeClaims. A headless Service gives each pod a unique DNS name, enabling stable network identity. Together, they satisfy the requirements for stable identity, persistent storage, and horizontal scaling.

Answer

DaemonSet with hostPath volumes

Answer

Deployment with a PersistentVolumeClaim

Answer

ReplicaSet with a ClusterIP Service

Question 8

A financial services company runs a containerized trading dashboard on Google Kubernetes Engine (GKE) Standard. The dashboard reads from a Cloud SQL for PostgreSQL instance and must survive a zonal outage without manual intervention. The team wants to minimize cost while ensuring the database remains reachable if the primary zone fails. Which configuration should they implement?

Accepted Answer

Configure the Cloud SQL instance with high availability (HA) enabled, which provisions a standby replica in a different zone in the same region.. High availability on Cloud SQL automatically maintains a standby in a different zone and fails over without manual action. This is the standard, cost-effective way to survive a zonal failure for a single-region application. Read replicas and cross-region replicas serve different purposes, and PITR is for recovery, not availability.

Answer

Deploy the Cloud SQL instance with a cross-region replica and use a global load balancer to route database traffic.

Answer

Create a read replica in a different zone and configure the application to switch to it when the primary becomes unavailable.

Answer

Enable point-in-time recovery (PITR) on the Cloud SQL instance and restore the database to a new instance during an outage.

Question 9

A financial services team runs a Spring Boot service on Cloud Run that must call an external payment API. The external API enforces a strict rate limit and occasionally returns HTTP 503 during maintenance windows. The team wants the Cloud Run service to remain responsive and avoid cascading failures when the dependency degrades. What should they do?

Accepted Answer

Implement retries with exponential backoff and jitter, a circuit breaker around the external call, and a bounded thread or connection pool.. Resilience against a flaky dependency comes from combining bounded resources with failure isolation. Exponential backoff with jitter absorbs transient 503 responses, a circuit breaker stops traffic once failures cross a threshold so the service fails fast, and a bounded connection or thread pool ensures a slow dependency cannot consume all instance capacity. Together these keep the Cloud Run service responsive and prevent failures from propagating to upstream callers.

Answer

Configure Cloud Run request timeout to 1 second and let failed requests return 500 to the caller immediately.

Answer

Move the external API calls into a Cloud Tasks queue and return 202 Accepted to the caller immediately for every request.

Answer

Increase the Cloud Run concurrency setting to the maximum so more requests are handled in parallel while the dependency is slow.

Question 10

A media company is building a video transcoding pipeline on Google Cloud. Jobs are submitted to a Pub/Sub topic, and a pool of worker applications running on GKE consumes the messages. The team needs the pipeline to process each job exactly once, tolerate worker crashes without losing messages, and scale workers based on backlog. Which TWO design choices should they implement? (Choose two.)

Accepted Answer

Configure the GKE deployment to use a Kubernetes HorizontalPodAutoscaler driven by a custom metric that reflects the Pub/Sub backlog.. Reliable Pub/Sub consumption requires acknowledging messages only after successful processing and extending the acknowledgement deadline for long jobs, so crashes cause redelivery instead of loss. Scaling on backlog requires an autoscaler driven by a metric that reflects unacknowledged messages. Together these choices make the transcoding pipeline resilient and elastic, while short deadlines, alternative services, and local state do not meet the requirements.

Answer

Store job state in a local file on each worker pod and use a Kubernetes liveness probe to restart failed pods.

Answer

Set the Pub/Sub subscription to use a short acknowledgement deadline and rely on automatic retries to handle failures.

Answer

Switch the pipeline to use Cloud Tasks with HTTP targets instead of Pub/Sub to guarantee exactly-once delivery.

Question 11

A company is building a microservices application on Cloud Run. They want to ensure that if a downstream service becomes unavailable, the upstream service does not overwhelm it with retries. Which pattern should they implement to handle transient failures gracefully and prevent cascading failures?

Accepted Answer

Circuit breaker pattern. The circuit breaker pattern monitors for failures and, when a threshold is reached, opens the circuit to stop further requests to the failing service. This gives the downstream service time to recover and prevents the upstream service from wasting resources on doomed requests. It is the standard pattern for preventing cascading failures in microservices.

Answer

Retry with exponential backoff

Answer

Timeout and bulkhead pattern

Answer

Load balancing with round-robin

Question 12

A financial services firm runs a Java service on GKE Standard. The service holds in-memory session state and must survive a zonal failure without losing sessions. The team wants the application to remain available if a single zone becomes unavailable. Which design should they implement?

Accepted Answer

Deploy the service as a Deployment spread across zones, and externalize session state to a highly available store such as Memorystore for Redis with a replica in another zone.. Surviving a zonal failure requires both compute redundancy across zones and session data that does not live only in a pod's memory. A multi-zone Deployment keeps serving replicas available, while a replicated Memorystore for Redis instance stores sessions outside the pods, so a client can be routed to a surviving replica and still find its session. Together these provide continuity without relying on any single zone.

Answer

Deploy the service as a DaemonSet so that one pod runs on every node in the cluster, ensuring coverage across all zones.

Answer

Deploy the service as a StatefulSet with a PersistentVolumeClaim per replica, and pin each replica to a specific zone using node affinities.

Answer

Deploy the service as a Deployment with a single replica and configure a PodDisruptionBudget to prevent eviction during maintenance.

Question 13

A retail company runs a stateless order-processing API on Cloud Run. The service has a 30-second request timeout configured, and each request writes a record to Firestore before returning a confirmation to the client. During a flash sale, Firestore returns intermittent DEADLINE_EXCEEDED errors, and some orders are silently lost because the client never receives a confirmation and does not retry. The team wants to guarantee that every accepted order is eventually persisted without changing the client. What should they do?

Accepted Answer

Publish each order to a Pub/Sub topic and have a separate Cloud Run consumer write to Firestore with idempotent retry handling.. The reliable pattern is to accept the order into a durable buffer and process the write asynchronously. Pub/Sub persists the message immediately and can redeliver it, so a consumer that retries idempotently guarantees eventual persistence. Extending timeouts, relying on platform retries, or swapping datastores all leave the synchronous write path vulnerable to transient failures.

Answer

Add an HTTP retry policy to the Cloud Run service so the platform automatically resends failed requests to the same instance.

Answer

Increase the Cloud Run request timeout to 60 seconds and enable CPU always allocated so the container keeps running while Firestore retries.

Answer

Move the order data to Cloud SQL and use a connection pool with automatic retries on transient serialization failures.

Question 14

A team is building a new API on Cloud Run that must serve users across North America, Europe, and Asia with minimal latency and remain available if an entire region fails. The API is stateless and uses Firestore in Native mode for persistence. Which deployment configuration should the team use?

Accepted Answer

Deploy the Cloud Run service to multiple regions behind a global external Application Load Balancer with a serverless network endpoint group (NEG) and a global anycast IP address.. A global external Application Load Balancer with a serverless NEG is the Google Cloud-recommended pattern for globally distributed, stateless Cloud Run services. It provides a single anycast IP, routes users to the nearest healthy region, and can fail over automatically if a region becomes unhealthy. Firestore in Native mode is multi-region, so no additional data replication is required for this stateless API.

Answer

Deploy the Cloud Run service to a single region with a minimum instance count set high enough to handle peak traffic and enable CPU always allocated for all instances.

Answer

Deploy the Cloud Run service to multiple regions and configure Cloud DNS with a geolocation routing policy that resolves to the nearest regional Cloud Run URL.

Answer

Deploy the Cloud Run service to a single region and front it with Cloud CDN using a custom origin to cache API responses at edge locations.

Question 15

A media company ingests telemetry from millions of devices into Pub/Sub. A Dataflow streaming pipeline enriches the messages and writes results to BigQuery. During peak hours the pipeline falls behind and the Pub/Sub subscription backlog grows continuously. The pipeline uses a fixed number of 20 workers. What should the team do to keep the backlog low while controlling cost?

Accepted Answer

Switch the pipeline to use Streaming Engine and enable autoscaling with a maximum number of workers.. A streaming pipeline that falls behind is capacity-constrained, and the most effective remedy is to let it scale with load. Streaming Engine separates execution from worker VMs so autoscaling responds quickly to backlog, and a maximum worker count places a ceiling on spend. This keeps the subscription backlog low during peaks while avoiding the cost of a permanently large static worker pool.

Answer

Increase the BigQuery write throughput by switching the sink to streaming inserts with a higher quota.

Answer

Increase the number of workers to a fixed value of 200 and leave the pipeline running at that size permanently.

Answer

Reduce the Pub/Sub message retention duration so old messages expire before they are processed.