PCD Designing highly scalable, available, and reliable cloud-native applications Practice Test 14 — 15 Questions

Question 1

A startup runs a stateless containerized API on Cloud Run. Traffic is unpredictable and can spike from zero to thousands of requests per second within seconds. The team wants the service to scale out rapidly while keeping cold starts low. Which Cloud Run configuration should they use?

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

Accepted Answer

Set the minimum number of instances to a value greater than zero and configure CPU to be always allocated.. Cloud Run scales to zero by default, which introduces cold starts when traffic resumes. Setting a minimum instance count above zero keeps a pool of warm instances ready, and allocating CPU always ensures those instances remain responsive. Together they provide fast response during sudden traffic spikes while the platform scales out additional instances as needed.

Answer

Enable CPU throttling outside of request processing and set the minimum instances to zero.

Answer

Set the maximum number of instances to a high value and rely on the default CPU allocation.

Answer

Deploy the service with a concurrency setting of one request per instance.

Question 2

A media company streams video metadata through a Cloud Run service in us-central1. The service writes to a Firestore database in the same region. A new requirement says the service must survive a full regional outage in us-central1 with a recovery time objective (RTO) of 15 minutes. The team wants the least operational overhead while keeping a consistent global view of the data. Which design should they choose?

Accepted Answer

Deploy the Cloud Run service to multiple regions behind a global external Application Load Balancer and use Firestore in Native mode with a multi-region location.. The goal is automatic regional failover with a single consistent data view and minimal operations. A global external Application Load Balancer sends traffic to a surviving Cloud Run region, and a multi-region Firestore Native location keeps one logical database replicated across regions. This removes manual failover and meets the 15-minute RTO with little operational effort.

Answer

Deploy the Cloud Run service to multiple regions and use Firestore in Datastore mode with a single-region location in us-central1, relying on scheduled exports for recovery.

Answer

Deploy the Cloud Run service to multiple regions and use Cloud SQL for PostgreSQL with a cross-region replica, promoting the replica during an outage.

Answer

Keep the Cloud Run service in us-central1 and add a Cloud Storage bucket in a dual-region as a write-ahead log that the service replays after an outage.

Question 3

A startup deploys a stateless containerized web application on Google Kubernetes Engine (GKE) Autopilot. The application must remain available even if a single zone fails. The cluster is regional with nodes spread across three zones. What should the team do to ensure the application survives a zone failure?

Accepted Answer

Set the Deployment's replica count to at least three and configure topology spread constraints to spread pods evenly across zones.. For a stateless application on a regional GKE cluster, zone resilience comes from running enough replicas and using topology spread constraints so that no single zone hosts all pods. If a zone fails, the remaining replicas continue serving requests, and the Deployment controller reschedules pods onto healthy nodes in the surviving zones.

Answer

Configure a PodDisruptionBudget that requires at least one pod to be available at all times.

Answer

Create a second GKE cluster in another region and use Multi Cluster Ingress to route traffic to both clusters.

Answer

Enable node auto-repair and node auto-upgrade on the cluster's node pools.

Question 4

A media company serves video metadata from a Cloud Spanner database. The application runs in us-central1 and europe-west1 and uses a regional instance configuration in us-central1. Users in Europe report high read latency. The company wants to reduce latency for European users while keeping strong consistency and a single logical database. What should they do?

Accepted Answer

Change the Spanner instance to a multi-region configuration that includes europe-west1, and keep the same database.. Cloud Spanner multi-region configurations replicate data across regions while preserving strong consistency and a single logical database. Adding europe-west1 to the instance configuration allows European reads to be served locally, reducing latency. This is the only option that maintains the required consistency model and avoids splitting the database.

Answer

Add a Cloud CDN in front of the Spanner database to cache metadata reads for European users.

Answer

Create a second Spanner instance in europe-west1 and use a two-phase commit protocol to keep both instances in sync.

Answer

Enable read-only replicas in europe-west1 by creating a separate Spanner database and configuring the application to read from it.

Question 5

A startup deploys a stateless REST API on Cloud Run. The API must handle unpredictable spikes and remain available if one zone fails. The team wants the simplest configuration that provides automatic scaling and zone redundancy without managing servers. What should they do?

Accepted Answer

Deploy the API as a Cloud Run service and set the maximum number of instances high enough to absorb the spikes.. Cloud Run is a fully managed serverless platform that scales instances automatically with request load and spreads them across zones in the region. Raising the maximum instance count allows the service to absorb spikes, and the team avoids all server management. This gives autoscaling and zone redundancy with the least configuration.

Answer

Deploy the API on a single Compute Engine VM with a startup script that installs the runtime.

Answer

Deploy the API on a Compute Engine managed instance group with an autoscaler and a health check.

Answer

Deploy the API on Google Kubernetes Engine with a single node pool and a Horizontal Pod Autoscaler.

Question 6

A financial services company runs a critical application on Google Kubernetes Engine (GKE) that must be resilient to zonal failures. The application uses a PostgreSQL database deployed as a StatefulSet with a PersistentVolumeClaim. They want to ensure that if a zone fails, the database pod is rescheduled to another zone and can reattach to its persistent disk. What should they do?

Accepted Answer

Use a regional GKE cluster with a StorageClass that provisions regional persistent disks (e.g., pd-balanced with replication-type: regional-pd).. Regional persistent disks replicate data across two zones, allowing a pod to reattach to the disk in a surviving zone. A regional GKE cluster spreads nodes across multiple zones, so the pod can be rescheduled. This combination ensures that a zonal failure does not cause data loss or prolonged downtime. The StorageClass must be configured to provision regional disks, and the StatefulSet should not have zone-pinning constraints.

Answer

Create a regional GKE cluster and use a StorageClass with volumeBindingMode: WaitForFirstConsumer and a zonal persistent disk.

Answer

Deploy the StatefulSet with a node affinity rule that pins the pod to a specific zone, and use a local SSD for storage.

Answer

Use a zonal GKE cluster and configure the StatefulSet to use a PersistentVolumeClaim with a ReadWriteMany access mode on Filestore.

Question 7

A team is building a highly available application on Google Cloud that must survive a zonal failure. The application runs on Compute Engine and uses a managed instance group. They want to ensure that the application remains available if a single zone becomes unavailable. (Choose two.)

Accepted Answer

Distribute the managed instance group across multiple zones in the region.. To survive a zonal failure, compute capacity must exist in multiple zones, and traffic must be routed only to healthy instances. A regional managed instance group distributes instances across zones, and an HTTP(S) load balancer with health checks automatically avoids failed zones. Together they provide continuous availability during a zonal outage.

Answer

Attach a persistent disk to each instance and configure the disks to replicate across zones.

Answer

Set the instance group's autoscaling policy to scale based on CPU utilization.

Answer

Use a single-zone managed instance group and enable autoscaling.

Question 8

You are deploying a stateless API on Google Kubernetes Engine (GKE) that must survive a zone outage in a single region without manual intervention. The cluster is regional, and the Deployment runs three replicas. What should you configure to ensure pods are spread across zones and that the Deployment maintains availability?

Accepted Answer

Configure topology spread constraints with topologyKey set to the zone label and a balanced skew policy.. Topology spread constraints are the native GKE mechanism for spreading pods across zones in a regional cluster. By targeting the zone topology key with a balanced skew policy, the scheduler distributes replicas evenly, so a zone outage degrades capacity rather than taking the whole Deployment offline. Disruption budgets and autoscalers address different availability concerns.

Answer

Set a PodDisruptionBudget with minAvailable: 1 on the Deployment.

Answer

Create a horizontal Pod autoscaler that scales the Deployment based on CPU utilization.

Answer

Add pod anti-affinity rules that require pods to run on nodes in different zones.

Question 9

A logistics company runs an event-driven order-tracking service on Cloud Run that consumes messages from a Pub/Sub push subscription. During traffic spikes, the service returns HTTP 500 errors for some messages, and the subscription's dead-letter topic is filling up. The team wants failed messages to be retried automatically before being sent to the dead-letter topic, and they want to avoid duplicate processing. What should they do?

Accepted Answer

Configure a retry policy with exponential backoff on the subscription and make the consumer idempotent using a message ID deduplication store.. A retry policy with exponential backoff lets Pub/Sub redeliver failed messages with increasing delays before the dead-letter threshold is reached, giving the service time to recover. Idempotent consumers keyed on message ID prevent duplicates from redelivery. Together they ensure reliable retries without duplicate order processing.

Answer

Set the dead-letter topic's max delivery attempts to one so messages move to the dead-letter topic before any retry occurs.

Answer

Increase the subscription's ack deadline to the maximum and disable the dead-letter topic so messages are retried indefinitely.

Answer

Switch from a push subscription to a pull subscription and have each Cloud Run instance acknowledge messages immediately upon receipt.

Question 10

A retail company runs a web application on Compute Engine behind an external Application Load Balancer. They want the application to remain available if an entire zone becomes unhealthy, and they want unhealthy instances removed from serving traffic automatically. The application instances are managed by a regional managed instance group (MIG) with an autoscaler. What should they do?

Accepted Answer

Configure an HTTP health check on the backend service and set the MIG to use autohealing.. A regional managed instance group spreads instances across zones in a region, and combining a load balancer health check with MIG autohealing ensures unhealthy instances are both removed from traffic and automatically recreated. This gives zone-level resilience and self-healing without the overhead of a multi-region design. Capacity changes and connection draining do not provide these behaviors.

Answer

Increase the MIG's maximum number of instances so more capacity is available during a zone failure.

Answer

Add a second regional MIG in another region and use global load balancing across both.

Answer

Enable connection draining on the backend service to allow in-flight requests to complete.

Question 11

A company runs a microservices application on Google Kubernetes Engine (GKE). They want to ensure that if a single node fails, the application's Pods are rescheduled and the service remains available. The application uses a Deployment with multiple replicas. Which GKE feature should they configure to ensure Pods are spread across nodes and zones?

Accepted Answer

Configure topology spread constraints for the Deployment.. Topology spread constraints are designed to distribute Pods across failure domains such as nodes and zones. By defining constraints with appropriate topology keys and whenUnsatisfiable policies, the scheduler ensures that replicas are spread out, so a single node or zone failure affects only a subset of Pods. This is the GKE-native way to achieve even distribution.

Answer

Use node affinity rules to schedule Pods on specific nodes.

Answer

Configure a PodDisruptionBudget for the Deployment.

Answer

Enable the Horizontal Pod Autoscaler for the Deployment.

Question 12

A retail company runs an e-commerce application on Compute Engine instances in a managed instance group (MIG). The application stores session data in a local SSD on each instance. During a rolling update, the company wants to ensure that active user sessions are not lost and that the application remains available. They also want to minimize downtime during the update. What should they do?

Accepted Answer

Move session data to a centralized, highly available data store such as Cloud Firestore or Memorystore, and then perform a rolling update on the MIG.. The core issue is that session data is stored on local SSDs, which are ephemeral and tied to the instance. To ensure sessions are not lost during updates, the data must be moved to a centralized, highly available store. Cloud Firestore or Memorystore provide low-latency, durable storage for session state. Once sessions are externalized, a rolling update can be performed without affecting users, as new instances can retrieve session data from the central store. This approach ensures availability and minimizes downtime.

Answer

Create a new MIG with the updated instance template and gradually shift traffic using a load balancer.

Answer

Configure the MIG with a rolling update and set the maximum surge to 1 and maximum unavailable to 0.

Answer

Use a stateful MIG and configure the update policy to recreate instances while preserving the persistent disk.

Question 13

A media company is designing a highly available video transcoding pipeline on Google Cloud. Jobs are submitted to a Pub/Sub topic, and workers run on GKE. They need the pipeline to continue processing if a single zone fails and to avoid losing messages if a worker crashes mid-job. Which TWO design choices should they implement? (Choose two.)

Accepted Answer

Deploy the GKE worker Deployment across multiple zones using a regional cluster and topology spread constraints.. High availability for the pipeline requires workers to survive a zone failure and messages to survive worker crashes. Spreading the worker Deployment across zones in a regional cluster keeps processing alive during a zone outage, and acknowledging only after successful completion lets Pub/Sub redeliver unacknowledged messages to healthy workers. Short retention, immediate acknowledgment, and single-zone deployment all undermine these goals.

Answer

Use a single-zone GKE cluster with a high number of worker replicas to maximize in-zone throughput.

Answer

Configure the worker pods to acknowledge messages immediately upon receipt to reduce duplicate processing.

Answer

Set the Pub/Sub subscription's message retention duration to the minimum allowed value.

Question 14

A media company runs a video transcoding pipeline on Google Kubernetes Engine. The pipeline uses a Deployment with pods that pull jobs from a Pub/Sub subscription and write output to Cloud Storage. During a node upgrade, several pods are terminated mid-transcode, and jobs are lost. The team wants terminated pods to finish their current job before shutdown, up to a bounded time. What should they do?

Accepted Answer

Set the pod's terminationGracePeriodSeconds to a value larger than the longest transcode and handle SIGTERM in the application to stop fetching new messages, finish the current job, acknowledge the message, then exit.. Graceful termination on Kubernetes depends on the application reacting to SIGTERM and on the pod having enough time to finish. Handling SIGTERM lets the worker stop fetching new messages and complete the current transcode, while a terminationGracePeriodSeconds longer than the longest job gives it room to acknowledge and exit. Together these prevent job loss during node upgrades.

Answer

Configure the Deployment with a PodDisruptionBudget of maxUnavailable: 0 and rely on it to prevent pod termination during node upgrades.

Answer

Add a preStop lifecycle hook that sleeps for the maximum transcode duration and increase the terminationGracePeriodSeconds on the pod spec to match.

Answer

Change the workload from a Deployment to a DaemonSet so that one pod runs per node and is not evicted during upgrades.

Question 15

A team runs a stateful order-processing service on GKE that uses a Redis instance on Compute Engine for session state. The service must survive a zonal failure without losing session data. The Redis instance currently runs on a single VM in us-central1-a. What should the team do to meet the requirement with the least operational overhead?

Accepted Answer

Migrate the session state to Memorystore for Redis with a Standard Tier instance that has automatic failover to a replica in another zone.. Memorystore for Redis Standard Tier provides a primary and replica in different zones with automatic failover. The managed service handles replication and failover, so the application continues to read and write session data after a zonal failure. This meets the availability requirement with far less operational effort than self-managed replication.

Answer

Move the Redis instance to a larger machine type and enable live migration on the Compute Engine VM.

Answer

Create a managed instance group of Redis VMs across three zones and configure Redis replication manually.

Answer

Configure the application to write session data to both Redis and a Cloud Storage bucket for durability.