Courseiva

CCNA Designing highly scalable, available, and reliable cloud-native applications Questions

75 of 114 questions · Page 1/2 · Designing highly scalable, available, and reliable cloud-native applications · Answers revealed

Practice these questionsDomain overviewAll questions
1
Multi-Selecthard

A team is designing a globally distributed application on Google Cloud that requires strong consistency for writes but can tolerate eventual consistency for reads. The application expects millions of concurrent users. Which two strategies should they implement? (Choose two.)

Select 2 answers
A.Use Cloud Spanner for write operations requiring strong consistency.
B.Use Firestore in multi-region mode for all operations.
C.Use global HTTP(S) Load Balancer with Cloud CDN for read-heavy traffic.
D.Deploy Cloud SQL with cross-region replication for read scalability.
E.Use Cloud Bigtable for reading data with strong consistency.
AnswersA, C

Cloud Spanner provides externally consistent reads and writes across global regions using TrueTime and atomic clocks, satisfying the strong consistency requirement for write operations in the stem. Its synchronous replication ensures that all writes are immediately visible to subsequent reads, directly addressing the need for write consistency despite eventual read tolerance.

Why this answer

Cloud Spanner provides strongly consistent writes globally through synchronous replication using the TrueTime API and Paxos-based consensus. This ensures that all write operations are immediately consistent across regions, meeting the requirement for strong consistency on writes.

Exam trap

The PCD exam often tests the misconception that a single database can handle both strong consistency and high scalability for reads and writes, leading candidates to choose Firestore or Bigtable without considering the specific consistency requirements for writes versus reads.

Practice this question →
2
Multi-Selectmedium

A company is designing a cloud-native application on Google Kubernetes Engine. They want to ensure high availability and scalability for their microservices. Which two best practices should they follow?

Select 2 answers
A.Use a single cluster per region.
B.Use a single replica for each service to reduce cost.
C.Use horizontal pod autoscaling based on custom metrics.
D.Use stateful sets for all services.
E.Deploy services across multiple zones.
AnswersC, E

HPA allows scaling based on application-specific metrics.

Why this answer

Horizontal Pod Autoscaling (HPA) based on custom metrics allows the application to automatically scale the number of pod replicas in response to application-specific signals (e.g., requests per second, queue depth) rather than just CPU/memory. This ensures that each microservice can handle varying load efficiently, maintaining high availability and scalability without over-provisioning.

Exam trap

The PCD exam often tests the misconception that high availability requires a single cluster per region, but the trap is that true resilience demands multi-zone or multi-region deployment to survive zonal failures, not just cluster redundancy.

Practice this question →
3
MCQhard

A company runs a critical application on Compute Engine with a stateful database. They need to achieve 99.99% availability for the database tier. Which architecture should they implement?

A.A Compute Engine instance group with managed instance groups and a regional persistent disk configured for synchronous replication.
B.Use Cloud SQL with automatic failover and read replicas.
C.Two Compute Engine instances in different zones with a shared Zonal persistent disk.
D.Single Compute Engine instance with a persistent disk snapshot scheduled every hour.
AnswerA

Regional persistent disks replicate data synchronously across zones, and the managed instance group can automatically fail over to a new instance in another zone on failure, achieving high availability.

Why this answer

A managed instance group with a regional persistent disk configured for synchronous replication provides the necessary 99.99% availability by ensuring the database runs across two zones with synchronous writes to both replicas. This architecture allows automatic failover within seconds if one zone fails, meeting the high-availability requirement without data loss.

Exam trap

The PCD exam often tests the misconception that a shared zonal persistent disk across two instances provides high availability, but the trap is that a zonal disk is still tied to a single zone and fails if that zone goes down, whereas a regional persistent disk is required for true multi-zone resilience.

How to eliminate wrong answers

Option B is wrong because Cloud SQL with automatic failover and read replicas is a managed database service that offers up to 99.95% availability, not 99.99%, and read replicas are asynchronous, which can lead to data loss during failover. Option C is wrong because two Compute Engine instances in different zones with a shared zonal persistent disk cannot achieve 99.99% availability, as a zonal disk is tied to a single zone and becomes inaccessible if that zone fails, causing a single point of failure. Option D is wrong because a single Compute Engine instance with hourly persistent disk snapshots provides no high availability; recovery from a snapshot can take minutes to hours, far exceeding the downtime allowed for 99.99% availability (approximately 52.56 minutes per year).

Practice this question →
4
Multi-Selecthard

A company uses Cloud Spanner for a globally distributed application. They need to design their table schema for maximum scalability and performance. Which two design considerations are critical? (Choose two.)

Select 2 answers
A.Use interleaved tables to colocate related data.
B.Store large binary blobs directly in Spanner.
C.Define secondary indexes on every column.
D.Use monotonically increasing primary keys.
E.Choose primary keys that distribute write load evenly across nodes.
AnswersA, E

Interleaved tables store parent and child rows in the same split, reducing the number of participants in transactions and improving performance.

Why this answer

Interleaved tables in Cloud Spanner physically colocate parent and child rows on the same split, reducing cross-node reads and improving join performance. This design is critical for globally distributed applications because it minimizes latency and ensures that related data is stored together for efficient access.

Exam trap

The PCD exam often tests the misconception that secondary indexes on every column improve query performance, but in Spanner they increase write latency and storage costs without benefiting all queries.

Practice this question →
5
MCQeasy

A developer is using Cloud Spanner for a global application. They need to design a schema to avoid hotspots. Which practice should they follow?

A.Use a UUID primary key
B.Use a monotonically increasing primary key
C.Use a composite primary key with a timestamp
D.Use a hash prefix on the primary key
AnswerD

A hash prefix distributes write load evenly across nodes, avoiding hotspots.

Why this answer

Cloud Spanner uses a distributed architecture that splits data across splits based on the primary key. A monotonically increasing or timestamp-based key causes all new writes to hit a single split, creating a hotspot. By using a hash prefix on the primary key, writes are distributed uniformly across all splits, avoiding hotspots and maximizing write throughput.

Exam trap

The PCD exam often tests the misconception that any random key (like UUID) automatically avoids hotspots, but in Cloud Spanner, the key distribution must be explicitly designed to avoid sequential patterns, and a hash prefix is the recommended pattern.

How to eliminate wrong answers

Option A is wrong because a UUID primary key, while random, is 128 bits and can still lead to uneven distribution and hotspotting if the UUID generation is not truly random or if the application uses sequential UUIDs (e.g., UUID v1). Option B is wrong because a monotonically increasing primary key (e.g., auto-increment integer) causes all new rows to be written to the last split, creating a severe write hotspot. Option C is wrong because a composite primary key with a timestamp as the leading column causes all writes at the same time to target the same split, again creating a hotspot.

Practice this question →
6
Multi-Selecthard

A company runs a microservices architecture on GKE with gRPC services. They want to implement traffic splitting for canary deployments. Which THREE components should they use?

Select 3 answers
A.ClusterIP service
B.Istio or Anthos Service Mesh
C.Ingress resource
D.Google Cloud Load Balancer
E.Headless service
AnswersB, C, D

Provides advanced traffic management, including weight-based canary deployments.

Why this answer

Istio or Anthos Service Mesh provides fine-grained traffic splitting capabilities for canary deployments in a GKE environment. It uses Envoy sidecar proxies to route a percentage of traffic to different service versions based on HTTP headers or weight, enabling controlled rollouts without modifying application code.

Exam trap

The trap here is that candidates often confuse ClusterIP or Headless services with traffic splitting capabilities, but these are only for basic service discovery and do not provide the advanced routing needed for canary deployments.

Practice this question →
7
MCQhard

A financial services company runs a transaction processing microservice on Google Kubernetes Engine (GKE). The service uses Cloud Spanner as its database. After migrating from Cloud SQL to Spanner to improve scalability, the team notices that a small percentage of transactions fail with an 'ABORTED' error due to deadlock detection. The application currently performs no retries, and the failures cause customer-facing errors. The team also observes that under peak load, transaction latencies are around 500ms, which is acceptable but they want to ensure the system remains reliable. They need to implement a solution that minimizes failures while maintaining acceptable performance. Which course of action should they take?

A.Increase the number of Spanner nodes to reduce the probability of deadlocks.
B.Reduce the size of each transaction by splitting them into smaller ones.
C.Change the transaction isolation level to READ UNCOMMITTED to avoid deadlocks.
D.Implement retry logic with exponential backoff and random jitter for aborted transactions.
AnswerD

Retrying with backoff and jitter is the standard pattern for handling Spanner aborts, ensuring transient conflicts are resolved without significant latency impact.

Why this answer

In Cloud Spanner, 'ABORTED' errors due to deadlock detection are a normal part of the optimistic concurrency control mechanism. The correct solution is to implement retry logic with exponential backoff and random jitter, as recommended by Google's own documentation. This approach transparently handles transient deadlocks without requiring infrastructure changes or sacrificing consistency, and it maintains acceptable latency by spacing out retries.

Exam trap

The PCD exam often tests the misconception that scaling infrastructure (more nodes) or reducing transaction size alone can eliminate deadlocks, when in fact retry logic is the required pattern for handling transient aborts in distributed databases like Cloud Spanner.

How to eliminate wrong answers

Option A is wrong because increasing the number of Spanner nodes improves throughput and storage capacity but does not directly reduce the probability of deadlocks; deadlocks are a function of transaction contention, not node count. Option B is wrong because splitting transactions into smaller ones can reduce the chance of conflicts but does not eliminate the need for retry logic; it also may break application-level atomicity requirements. Option C is wrong because Cloud Spanner does not support READ UNCOMMITTED isolation; it provides serializable isolation (and optional stale reads), and lowering isolation is not possible and would violate consistency guarantees.

Practice this question →
8
MCQmedium

Refer to the exhibit. A company configured an HPA for their deployment. They notice that the HPA is not scaling based on the 'packets-per-second' metric. What is the most likely reason?

A.The metric is not available in the cluster.
B.The metric name 'packets-per-second' is incorrect.
C.The target type should be 'Value' instead of 'AverageValue'.
D.The HPA is using the wrong scaleTargetRef.
AnswerA

Custom metrics must be exposed via an adapter; if not, HPA cannot access it.

Why this answer

The 'packets-per-second' metric is a custom metric. If it is not registered in the cluster's metrics server (e.g., via custom metrics adapter), the HPA will not be able to collect it. The metric name and target type are correct.

The scaleTargetRef matches the deployment. Therefore, the metric being unavailable is the most likely issue.

Practice this question →
9
MCQmedium

A team is migrating a monolithic application to a microservices architecture on Google Kubernetes Engine (GKE). They want to ensure that failures in one microservice do not cascade to others. Which design pattern should they implement?

A.Implement retry logic with exponential backoff for all inter-service calls.
B.Implement a circuit breaker pattern that opens when failure thresholds are exceeded.
C.Use synchronous HTTP calls with timeouts to detect failures quickly.
D.Use bulkheads to separate thread pools for each service.
AnswerB

Circuit breaker fails fast and prevents unnecessary load on failing services.

Why this answer

The circuit breaker pattern is the correct choice because it prevents cascading failures by monitoring inter-service calls and opening the circuit when failures exceed a threshold, allowing the system to fail fast and recover gracefully. In a GKE-based microservices architecture, this pattern is typically implemented using libraries like Resilience4j or Istio's circuit breaker, which can be configured to trip after a certain number of consecutive failures, thus protecting downstream services from being overwhelmed.

Exam trap

The PCD exam often tests the distinction between patterns that isolate failures within a component (bulkheads) versus patterns that prevent failures from propagating across components (circuit breaker), leading candidates to confuse the scope of each pattern.

How to eliminate wrong answers

Option A is wrong because retry logic with exponential backoff alone does not prevent cascading failures; it can actually exacerbate them by continuing to send requests to an already failing service, potentially causing resource exhaustion. Option C is wrong because synchronous HTTP calls with timeouts, while useful for detecting failures, do not provide a mechanism to stop repeated calls to a failing service, leading to thread pool starvation and cascading failures. Option D is wrong because bulkheads separate thread pools to isolate failures within a single service instance, but they do not prevent failures from propagating across different microservices in a distributed system.

Practice this question →
10
MCQeasy

A web application uses Cloud SQL for MySQL. The team expects a sudden spike in read-only traffic from a reporting tool. What should they use to offload read queries?

A.Automatic storage increase
B.Cross-region replication
C.Read replicas
D.Failover replica
AnswerC

Read replicas allow you to offload read queries from the primary instance, improving performance.

Why this answer

Read replicas in Cloud SQL for MySQL allow you to offload read traffic from the primary instance by creating one or more read-only copies. This is the correct approach for handling a sudden spike in read-only queries from a reporting tool, as it distributes the load without affecting write performance or requiring application changes beyond updating the connection string.

Exam trap

The PCD exam often tests the distinction between read replicas (for scaling reads) and failover replicas (for high availability), tempting candidates to choose failover replica because it sounds like it can handle traffic, but it cannot serve reads independently in Cloud SQL for MySQL.

How to eliminate wrong answers

Option A is wrong because automatic storage increase only adds disk space when the instance runs low, which does nothing to offload read queries or reduce CPU/memory load from read traffic. Option B is wrong because cross-region replication is designed for disaster recovery and geographic redundancy, not for scaling read capacity; it introduces latency and does not provide a local read endpoint for the reporting tool. Option D is wrong because a failover replica (also called a standby or HA replica) is a synchronous copy used for high availability and automatic failover, not for offloading read queries; it cannot serve read traffic independently in Cloud SQL for MySQL.

Practice this question →
11
MCQmedium

A company runs a batch job daily that processes large files from Cloud Storage and stores results in BigQuery. The job requires significant compute for about 10 minutes and is fault-tolerant. Which compute option is most cost-effective?

A.Cloud Run Jobs
B.Always-on Compute Engine VM
C.GKE cluster with a single node
D.Preemptible VM
E.Cloud Functions (9-minute timeout)
AnswerD

Low cost, suitable for fault-tolerant and short-lived workloads.

Why this answer

Preemptible VMs offer the same compute capacity as regular VMs at a significantly lower cost (up to 80% discount), and since the batch job runs for only 10 minutes daily and is fault-tolerant, it can handle the occasional preemption without data loss. The job's short duration and fault tolerance make preemptible instances ideal, as they can be restarted if terminated.

Exam trap

The PCD exam often tests the misconception that Cloud Functions or Cloud Run Jobs are always the cheapest serverless options, but the trap here is that the 9-minute timeout of Cloud Functions disqualifies it, and candidates overlook the cost savings of preemptible VMs for fault-tolerant, short-duration batch jobs.

How to eliminate wrong answers

Option A is wrong because Cloud Run Jobs have a maximum timeout of 60 minutes, which is sufficient, but they are designed for stateless containers and may incur higher costs per vCPU-hour compared to preemptible VMs for sustained batch processing. Option B is wrong because an always-on Compute Engine VM incurs costs 24/7, even when the job is not running, making it far more expensive than a preemptible VM that only runs for 10 minutes daily. Option C is wrong because a GKE cluster with a single node introduces unnecessary orchestration overhead and cost (including cluster management fees) for a simple batch job that does not require container orchestration.

Option E is wrong because Cloud Functions has a 9-minute timeout, which is insufficient for a job requiring 10 minutes of compute, and it is not designed for long-running batch processing.

Practice this question →
12
MCQmedium

A team is migrating a monolithic application to microservices on Google Kubernetes Engine (GKE). They want to ensure that if one microservice fails, it does not cascade to other services. Which design pattern should they implement?

A.Circuit Breaker pattern
B.Event-driven architecture
C.Retry with exponential backoff
D.Bulkhead pattern
AnswerA

Circuit Breaker pattern prevents cascading failures by opening the circuit when failures exceed a threshold.

Why this answer

The Circuit Breaker pattern is correct because it prevents cascading failures by monitoring for failures in a downstream microservice and, once a threshold is exceeded, immediately failing requests to that service without attempting the call. In GKE, this can be implemented using tools like Istio or Envoy sidecar proxies, which can be configured with circuit breaker settings to stop traffic to unhealthy pods, allowing the system to recover gracefully.

Exam trap

The PCD exam often tests the distinction between patterns that prevent cascading failures (Circuit Breaker) versus patterns that handle transient failures (Retry) or isolate resources (Bulkhead), so candidates mistakenly choose Retry or Bulkhead because they sound like they prevent failure spread, but they do not provide the fail-fast mechanism that stops the cascade.

How to eliminate wrong answers

Option B (Event-driven architecture) is wrong because it describes a communication style where services produce and consume events asynchronously, but it does not inherently provide failure isolation or prevent cascading failures; it can actually increase complexity in failure handling. Option C (Retry with exponential backoff) is wrong because it is a technique for handling transient failures by retrying with increasing delays, but it does not stop cascading failures; in fact, retrying a failing service can exacerbate the problem by adding load. Option D (Bulkhead pattern) is wrong because it isolates resources (e.g., thread pools or connections) per service or component to prevent a failure in one from exhausting shared resources, but it does not directly stop a failing service from being called; it limits blast radius but does not provide the fail-fast behavior of a circuit breaker.

Practice this question →
13
MCQhard

An organization runs a critical application on Compute Engine with a regional managed instance group. They want to achieve 99.99% availability. Which architecture should they use?

A.Regional MIG with instances in two zones
B.Single zone MIG with multiple instances
C.Regional MIG with instances in three zones
D.Multi-region deployment with global load balancer
AnswerC

Three zones provide higher availability within a region.

Why this answer

To achieve 99.99% availability, the architecture must tolerate both a zonal failure and a single instance failure. A regional managed instance group (MIG) with instances in three zones ensures that even if one zone becomes unavailable, the remaining two zones can still serve traffic, meeting the 99.99% uptime target. Three zones provide the necessary redundancy because a two-zone regional MIG can only survive a single zone failure but not a simultaneous instance failure in the remaining zone, whereas three zones allow for a rolling update or failure of one zone while still maintaining quorum.

Exam trap

The PCD exam often tests the misconception that two zones are sufficient for 99.99% availability, but the trap here is that two zones only provide 99.9% availability because they cannot tolerate a simultaneous instance failure in the remaining zone during a zonal outage or maintenance event.

How to eliminate wrong answers

Option A is wrong because a regional MIG with instances in only two zones can survive a single zone failure, but if an instance in the remaining zone fails or a rolling update is performed, the application may drop below the required capacity, failing to achieve 99.99% availability. Option B is wrong because a single zone MIG with multiple instances cannot survive a zonal outage; if the entire zone fails, all instances are lost, making 99.99% availability impossible. Option D is wrong because while a multi-region deployment with a global load balancer can provide even higher availability, the question specifically asks for an architecture using Compute Engine with a regional managed instance group, and a multi-region deployment is not a regional MIG architecture; it introduces cross-region latency and complexity not required for the stated 99.99% target.

Practice this question →
14
Multi-Selectmedium

A team is designing a cloud-native application that must be highly available and resilient to zone failures. Which three practices should they follow? (Choose three.)

Select 3 answers
A.Use a single Load Balancer with multiple backends.
B.Deploy resources across multiple zones.
C.Use zonal managed instance groups with 100% target utilization.
D.Store data in regional persistent disks.
E.Implement health checks and autohealing.
AnswersB, D, E

Distributing instances across zones protects against zone-level failures.

Why this answer

Deploying resources across multiple zones ensures that the application remains available even if an entire zone fails. In Google Cloud, zones are independent failure domains, and distributing workloads across them is a fundamental pattern for achieving high availability and resilience to zone-level outages.

Exam trap

The trap here is that candidates may think a single load balancer is sufficient for high availability, but in cloud-native design, the load balancer itself is a managed service that is inherently resilient, while the real risk is having backends in only one zone or no spare capacity to absorb failures.

Practice this question →
15
Multi-Selectmedium

A company is designing a highly available application on Google Cloud using multiple regions. Which TWO strategies should they implement to achieve this?

Select 2 answers
A.Use zonal persistent disks for stateful data.
B.Use a global load balancer to distribute traffic across regions.
C.Deploy a single instance group in one region for simplicity.
D.Configure managed instance groups in multiple regions.
E.Store all data in a single Cloud Storage bucket.
AnswersB, D

Global load balancers route traffic to the closest healthy backend, enabling multi-region high availability.

Why this answer

A global load balancer (e.g., Google Cloud External HTTPS Load Balancer) can distribute traffic across multiple regions, providing cross-region failover and low-latency routing. This is a fundamental pattern for multi-region high availability, as it allows traffic to be directed to healthy backends in any region, even if an entire region fails.

Exam trap

The trap here is that candidates often confuse zonal resources (like persistent disks) with regional or multi-regional resources, or they assume that a single-region deployment with a load balancer is sufficient for high availability, ignoring the need for geographic redundancy.

Practice this question →
16
MCQeasy

An application running on Cloud Run experiences cold starts causing latency spikes. What is the most cost-effective solution to reduce cold starts?

A.Set a minimum number of instances
B.Increase the container's CPU allocation
C.Enable HTTP keep-alive connections
D.Use a larger container memory size
AnswerA

Minimum instances keep the specified number of instances always warm, eliminating cold starts for those instances.

Why this answer

Setting a minimum number of instances ensures that Cloud Run always keeps at least one instance warm (idle) to serve incoming requests instantly, eliminating cold start latency. This is the most cost-effective solution because you only pay for the minimum instances when they are idle (at a reduced rate), whereas other options increase per-request cost or do not address the root cause of cold starts.

Exam trap

The PCD exam often tests the misconception that scaling resources (CPU or memory) or optimizing network connections can eliminate cold starts, but the only way to prevent cold starts is to keep instances warm, which is achieved by setting a minimum number of instances.

How to eliminate wrong answers

Option B is wrong because increasing CPU allocation does not prevent cold starts; it only speeds up request processing after the instance is already running, and it increases cost per instance without keeping instances warm. Option C is wrong because HTTP keep-alive connections reduce latency for subsequent requests over the same connection but do not eliminate the initial cold start when a new instance is created. Option D is wrong because larger memory size does not prevent cold starts; it may even increase cold start time due to longer container initialization, and it raises the cost per instance without guaranteeing a warm instance.

Practice this question →
17
MCQmedium

A team is deploying a microservices application on Google Kubernetes Engine (GKE). They want to ensure that if a pod fails, Kubernetes automatically replaces it and maintains the desired number of replicas. Which Kubernetes resource should they use?

A.StatefulSet
B.Deployment
C.Job
D.DaemonSet
AnswerB

A Deployment provides declarative updates for pods and ReplicaSets. It ensures that the desired number of pods are running and replaces failed pods automatically.

Why this answer

A Deployment is the correct Kubernetes resource for managing stateless microservices that require automatic pod replacement to maintain a desired replica count. It uses a ReplicaSet to ensure the specified number of pod replicas are running, and if a pod fails, the ReplicaSet controller immediately creates a new pod to restore the desired state.

Exam trap

The PCD exam often tests the distinction between stateless and stateful workloads, where candidates mistakenly choose StatefulSet for any application that needs high availability, overlooking that Deployments are the standard for stateless microservices with automatic replacement.

How to eliminate wrong answers

Option A is wrong because StatefulSet is designed for stateful applications that require stable network identities and persistent storage; it does not automatically replace pods in the same way as a Deployment for stateless workloads, and its pod replacement behavior is ordered and graceful, not immediate. Option C is wrong because a Job is used for batch or one-time tasks that run to completion, not for maintaining a desired number of continuously running replicas. Option D is wrong because a DaemonSet ensures that a copy of a pod runs on every node (or a subset of nodes) in the cluster, which is used for node-level services like logging or monitoring, not for maintaining a specific replica count across the cluster.

Practice this question →
18
MCQhard

A company uses Cloud Run for a serverless application that processes user uploads. Users report that sometimes the first request after a period of inactivity takes very long (cold start). The application is stateless. They want to minimize cold start latency while keeping costs low. The application is deployed with default settings: min instances = 0, max instances = 100, CPU always off, and a container image of 1GB. What should they do to reduce cold start latency?

A.Set min instances to 1 to keep a warm instance.
B.Increase container memory from the default to reduce startup time.
C.Use a larger container image to include more dependencies.
D.Enable CPU always on allocation.
AnswerA

Keeping a minimum number of instances eliminates cold starts.

Why this answer

Setting min instances to 1 ensures that at least one instance is always warm and ready to serve requests, eliminating the cold start for the first request after a period of inactivity. Since the application is stateless and the default min instances is 0, Cloud Run scales down to zero, causing a cold start on the next request. By keeping one instance warm, you minimize latency without significantly increasing cost, as you only pay for the single idle instance.

Exam trap

The PCD exam often tests the misconception that increasing resources (memory or CPU) or enabling CPU always on reduces cold start latency, when in fact the root cause is the instance being scaled to zero and the solution is to keep at least one instance warm via min instances.

How to eliminate wrong answers

Option B is wrong because increasing container memory does not reduce startup time; it only affects the CPU and memory resources available during execution, not the time to initialize the container. Option C is wrong because using a larger container image increases the download and extraction time during cold start, worsening the latency problem. Option D is wrong because enabling CPU always on allocation keeps the CPU active even when the instance is idle, which increases cost without addressing the cold start issue—the instance still scales to zero if min instances is 0.

Practice this question →
19
MCQmedium

A company is migrating a monolithic application to a microservices architecture on Google Cloud. They want to decouple services and ensure that a failure in one service does not impact others. Which pattern should they implement?

A.Implement caching with Memorystore
B.Increase the number of instances of each service
C.Use synchronous HTTP calls with retries
D.Implement circuit breaker pattern using a service mesh like Istio
AnswerD

Circuit breaker trips on failures, isolating the fault.

Why this answer

The circuit breaker pattern, implemented via a service mesh like Istio, is the correct approach because it prevents cascading failures by monitoring service health and stopping requests to a failing service until it recovers. Istio's Envoy sidecar proxies enforce circuit breaking at the network layer, allowing the system to degrade gracefully without impacting other services.

Exam trap

The PCD exam often tests the misconception that scaling instances (Option B) or adding caching (Option A) is sufficient for fault isolation, but these patterns address performance and availability, not decoupling or failure containment.

How to eliminate wrong answers

Option A is wrong because caching with Memorystore improves read performance and reduces latency but does not decouple services or prevent failure propagation; a failing service still receives requests. Option B is wrong because increasing instance count improves scalability and fault tolerance through redundancy but does not isolate failures—a failing service can still overwhelm downstream services or cause cascading issues. Option C is wrong because synchronous HTTP calls with retries increase coupling and can exacerbate failures by causing retry storms, overwhelming already failing services and violating the goal of decoupling.

Practice this question →
20
MCQhard

Refer to the exhibit. The user developer@example.com tries to create a firewall rule and receives a permission denied error. What is the most likely reason?

A.The user lacks compute.networkAdmin role
B.The user lacks compute.securityAdmin role
C.The user is missing compute.firewalls.create permission
D.All of the above
AnswerD

All three statements are correct; the user needs one of the roles or the specific permission.

Why this answer

All of the above. In Google Cloud, creating a firewall rule requires the `compute.securityAdmin` role (which includes `compute.firewalls.create` permission) or the `compute.networkAdmin` role (which also includes `compute.firewalls.create` permission). If the user lacks any of these roles or the specific permission, they will receive a permission denied error.

While options A, B, and C are individually true statements, the question asks for the 'most likely reason,' and since all three are valid, the single correct answer is D. This tests the understanding that multiple permission deficiencies can cause the same error.

Exam trap

The trap here is that there are three individually correct statements (A, B, C) and candidates might pick only one, but the question is designed to test whether you recognize that all three are valid reasons for the same error, making 'All of the above' the correct answer.

How to eliminate wrong answers

Option A is correct because the `compute.networkAdmin` role includes the `compute.firewalls.create` permission, and lacking it would cause a permission denied error. Option B is correct because the `compute.securityAdmin` role also includes the `compute.firewalls.create` permission, and lacking it would also cause the error. Option C is correct because the `compute.firewalls.create` permission is the specific IAM permission required to create firewall rules, and missing it directly results in a permission denied error.

Since all three options are individually valid reasons, the question expects the candidate to recognize that multiple factors can cause the same error, making D the only fully correct answer.

Practice this question →
21
MCQhard

A company runs a stateful application on Compute Engine instances with local SSDs. They need to perform maintenance that requires stopping the instances. What is the best approach to ensure data durability and minimal downtime?

A.Create a snapshot of the local SSD before stopping the instance
B.Use instance groups with autohealing to automatically recreate instances
C.Enable live migration on the instance
D.Migrate data to persistent disks and configure the application to use persistent disks
AnswerD

Persistent disks are durable and can be detached and reattached to other instances, ensuring data persistence during maintenance.

Why this answer

Local SSDs provide ephemeral storage that is tied to the lifecycle of the Compute Engine instance. When an instance is stopped or terminated, data on local SSDs is permanently lost. To ensure data durability during maintenance that requires stopping the instance, the application must use persistent disks, which are durable network-attached storage that persists independently of the instance.

Option D is correct because migrating the application to persistent disks ensures data survives the stop and allows the instance to be restarted with the same data, minimizing downtime.

Exam trap

The trap here is that candidates assume local SSDs can be snapshotted or that live migration works with local SSDs, but Google Cloud explicitly disables both features for local SSDs, making persistent disks the only durable option for stateful workloads requiring maintenance.

How to eliminate wrong answers

Option A is wrong because snapshots cannot be created directly from local SSDs; local SSDs are ephemeral and do not support snapshot creation. Option B is wrong because instance groups with autohealing recreate instances based on health checks, but they do not preserve data on local SSDs, which are lost when instances are terminated or recreated. Option C is wrong because live migration is enabled by default for instances with persistent disks, but it is not supported for instances with local SSDs; local SSDs prevent live migration, so the instance must be stopped for maintenance.

Practice this question →
22
MCQhard

A financial services company has a critical application that must survive a regional outage. They deployed on Compute Engine across multiple zones within a single region and now want to redirect traffic to a secondary region if the primary region becomes unavailable. Which load balancing solution should they use?

A.SSL Proxy Load Balancer
B.External HTTP(S) Load Balancer
C.Proxy Network Load Balancer
D.Internal TCP/UDP Load Balancer
E.Network Load Balancer
AnswerB

Global load balancer that can distribute traffic to backends in multiple regions and perform health-check-based failover.

Why this answer

The External HTTP(S) Load Balancer is the correct choice because it supports global load balancing across multiple regions, enabling traffic failover to a secondary region when the primary region becomes unavailable. It uses anycast IP addresses and is designed for HTTP/S traffic, making it suitable for a critical application that must survive a regional outage.

Exam trap

The trap here is that candidates often confuse regional load balancers (like Network Load Balancer or SSL Proxy) with global ones, assuming any load balancer can handle cross-region failover, but only the External HTTP(S) Load Balancer (and the External TCP/UDP Network Load Balancer with global access) supports multi-region failover for HTTP/S traffic.

How to eliminate wrong answers

Option A is wrong because SSL Proxy Load Balancer is a regional load balancer that terminates SSL connections and forwards TCP traffic, but it does not support cross-region failover or global load balancing. Option C is wrong because Proxy Network Load Balancer is a regional load balancer for TCP/UDP traffic and cannot redirect traffic to a secondary region. Option D is wrong because Internal TCP/UDP Load Balancer is a regional internal load balancer used for private traffic within a VPC and cannot handle cross-region failover.

Option E is wrong because Network Load Balancer is a regional passthrough load balancer for TCP/UDP traffic and does not support global load balancing or regional failover.

Practice this question →
23
MCQmedium

The developer runs the command above and sees both instances are unhealthy. The instances are running and serving traffic on port 80 when accessed directly. What is the most likely cause?

A.Firewall rules block the health check probe IP ranges
B.The instances have been deleted
C.The instances are not running the specified health check port
D.The load balancer is misconfigured
E.The instances are out of memory and unable to respond
AnswerA

Health check probes originate from Google's health checker IP ranges; they must be allowed in firewall rules.

Why this answer

The most likely cause is that firewall rules are blocking the health check probe IP ranges. Google Cloud Platform (GCP) load balancers use specific, documented IP ranges for health check probes. If a firewall rule denies traffic from these ranges, the load balancer will mark the instances as unhealthy even though the instances are running and serving traffic on port 80 when accessed directly.

This is a common misconfiguration because the health check probes originate from these special IP ranges, not from the load balancer's frontend IP.

Exam trap

The PCD exam often tests the misconception that health checks originate from the load balancer's frontend IP or that the instance's direct accessibility implies it will pass health checks, ignoring that health check probes come from specific, separate IP ranges that must be explicitly allowed in firewall rules.

How to eliminate wrong answers

Option B is wrong because the instances are explicitly described as 'running and serving traffic on port 80 when accessed directly,' so they have not been deleted. Option C is wrong because the instances are serving traffic on port 80, which matches the specified health check port (port 80), so the port is correct. Option D is wrong because the load balancer is correctly configured to send health checks to port 80, and the instances respond on that port; the issue is that the health check probes are being blocked, not that the load balancer configuration is incorrect.

Option E is wrong because the instances are serving traffic on port 80 when accessed directly, indicating they are not out of memory and are capable of responding; memory exhaustion would prevent all responses, not just health check responses.

Practice this question →
24
MCQmedium

A developer deploys this Cloud Run service. During a load test, each incoming request starts a new container instance, even though concurrency is set to 80. What is the reason?

A.The memory limit is too low
B.The container is CPU-bound and cannot handle multiple requests concurrently
C.The CPU limit is too low
D.The concurrency setting of 80 is too high and Cloud Run ignores it
E.The container is not designed to handle multiple concurrent requests (single-threaded)
AnswerE

If the container processes one request at a time, Cloud Run will start a new instance per request.

Why this answer

Cloud Run's concurrency setting controls how many requests the runtime can send to a container instance, but the container itself must be capable of handling those requests concurrently. If the application is single-threaded or uses a blocking I/O model (e.g., a simple Flask or Express server without async workers), it can only process one request at a time. Cloud Run detects that the container is busy and starts a new instance for each incoming request, effectively ignoring the concurrency setting.

Exam trap

The PCD exam often tests the misconception that Cloud Run's concurrency setting is a hard limit that the platform enforces regardless of application design, when in reality the application must be capable of handling concurrent requests for the setting to take effect.

How to eliminate wrong answers

Option A is wrong because a low memory limit would cause out-of-memory errors or container restarts, not the creation of a new container instance per request. Option B is wrong because being CPU-bound does not prevent a container from handling multiple concurrent requests; it may slow down processing, but Cloud Run still sends multiple requests to the same instance if concurrency is set. Option C is wrong because a low CPU limit would throttle CPU usage, not force a new instance per request; the container would still receive concurrent requests, just processed more slowly.

Option D is wrong because Cloud Run does not ignore a concurrency setting of 80; it respects the setting as long as the container can handle the load, but if the container is single-threaded, it effectively becomes a bottleneck.

Practice this question →
25
Drag & Dropmedium

Drag and drop the steps to set up a Cloud Function triggered by a Cloud Storage event in the correct order.

Drag steps to the numbered slots on the right, or tap a step then tap a slot.

Steps
Order
1Step 1
2Step 2
3Step 3
4Step 4

Why this order

Cloud Functions can be triggered by Cloud Storage events; deployment includes specifying the bucket trigger.

Practice this question →
26
MCQmedium

A developer is designing a data pipeline using Pub/Sub and Dataflow. They need to guarantee at-least-once delivery with no duplicates in the sink. Which Dataflow feature should they use?

A.Exactly-once processing
B.Checkpointing
C.Idempotent writes
D.Windowing
AnswerA

Exactly-once processing ensures each record is processed once, eliminating duplicates in the sink.

Why this answer

Dataflow's exactly-once processing (also known as 'exactly-once semantics' or 'EOS') ensures that each record is processed exactly once, even if the pipeline restarts or fails. This eliminates duplicates in the sink while still guaranteeing at-least-once delivery from Pub/Sub, because Dataflow uses a combination of source-side deduplication (via Pub/Sub message IDs) and sink-side idempotent writes (via the Dataflow sink's commit protocol). The result is that no duplicate records are written to the sink, meeting the requirement of no duplicates.

Exam trap

The trap here is that candidates confuse 'idempotent writes' (a sink-side property) with Dataflow's built-in 'exactly-once processing' feature, or they mistakenly think checkpointing alone eliminates duplicates, when in fact checkpointing only saves state and does not prevent duplicate writes to the sink.

How to eliminate wrong answers

Option B is wrong because checkpointing is a mechanism for saving pipeline state (e.g., snapshots of progress) to enable recovery after failures, but it does not by itself prevent duplicates in the sink — it only ensures that processing can resume from the last checkpoint, which may still cause duplicate writes if the sink is not idempotent. Option C is wrong because idempotent writes are a property of the sink (e.g., BigQuery's insertId or Cloud Storage's generation number) that allows the same write to be applied multiple times without creating duplicates, but the question asks for a Dataflow feature, not a sink feature; Dataflow's exactly-once processing uses idempotent writes as part of its implementation, but the feature itself is exactly-once processing. Option D is wrong because windowing is a Dataflow feature that groups unbounded data into finite windows (e.g., fixed, sliding, session) for aggregation or processing, but it has no direct role in guaranteeing at-least-once delivery or preventing duplicates — it is a time-based grouping mechanism, not a delivery semantics feature.

Practice this question →
27
MCQeasy

A startup expects low and predictable traffic initially but wants to use containers with minimal operational overhead. Which compute service should they choose?

A.App Engine Flexible Environment
B.Google Kubernetes Engine (GKE)
C.Cloud Run
D.Compute Engine
E.Cloud Functions
AnswerC

Fully managed, autoscaling, no infrastructure to manage.

Why this answer

Cloud Run is the correct choice because it runs containers in a fully managed, serverless environment that automatically scales from zero, requires no cluster management, and charges only for resources used during request processing. This matches the startup's need for minimal operational overhead and low, predictable traffic, as Cloud Run abstracts away infrastructure management entirely.

Exam trap

The PCD exam often tests the distinction between serverless containers (Cloud Run) and managed Kubernetes (GKE), where candidates mistakenly choose GKE for container support without considering the operational overhead of cluster management.

How to eliminate wrong answers

Option A is wrong because App Engine Flexible Environment requires managing VM instances and has a minimum of 1 instance running, incurring cost even with no traffic, and does not offer the same zero-scaling efficiency as Cloud Run. Option B is wrong because Google Kubernetes Engine (GKE) requires managing a Kubernetes cluster, including node pools, upgrades, and networking, which adds significant operational overhead unsuitable for minimal management. Option D is wrong because Compute Engine requires full VM management, including OS patching, scaling configuration, and capacity planning, contradicting the goal of minimal operational overhead.

Option E is wrong because Cloud Functions is for event-driven, short-lived code snippets, not for running containers, and has a 9-minute timeout and limited runtime support, making it unsuitable for containerized applications.

Practice this question →
28
MCQmedium

A stateful service on GKE needs to persist data that must be accessible from any pod in the cluster, regardless of which node the pod runs on. Which volume type should they use?

A.PersistentVolumeClaim with RWX access mode
B.emptyDir
C.ConfigMap
D.hostPath
AnswerA

PersistentVolumeClaim with ReadWriteMany allows multiple pods to access the same volume concurrently, even across nodes.

Why this answer

A PersistentVolumeClaim (PVC) with RWX (ReadWriteMany) access mode is correct because it allows multiple pods across different nodes to read and write to the same persistent volume simultaneously. This is essential for a stateful service where data must be accessible from any pod in the cluster, regardless of which node the pod runs on. RWX is typically backed by network filesystems like NFS or GKE Filestore, which provide shared access across nodes.

Exam trap

The PCD exam often tests the distinction between access modes (RWO, RWM, RWX) and candidates mistakenly choose hostPath or emptyDir because they think local storage is sufficient, overlooking the requirement for cross-node accessibility.

How to eliminate wrong answers

Option B (emptyDir) is wrong because it creates a temporary directory that is tied to the lifecycle of a pod and is not shared across pods on different nodes; data is lost when the pod is deleted. Option C (ConfigMap) is wrong because it is designed for storing non-sensitive configuration data as key-value pairs, not for persistent storage of application data; it cannot be used for read/write operations by pods. Option D (hostPath) is wrong because it mounts a file or directory from the host node's filesystem into a pod, making data inaccessible from pods running on other nodes and violating the requirement for cluster-wide accessibility.

Practice this question →
29
Multi-Selectmedium

A company is using Cloud Run for a stateless API. They want to ensure that the service can handle sudden traffic spikes. Which two features should they configure?

Select 2 answers
A.Enable container concurrency.
B.Use Cloud Load Balancing.
C.Enable CPU always on allocation.
D.Set max instances to a high value.
E.Set min instances to zero to save cost.
AnswersA, D

Container concurrency allows multiple requests per container, increasing throughput.

Why this answer

Enabling container concurrency allows a single Cloud Run container instance to handle multiple requests simultaneously, up to the configured concurrency limit (default 80, max 1000). This improves throughput and resource utilization during traffic spikes without requiring additional instances. Option D is correct because setting max instances to a high value ensures the service can scale out to handle sudden load by creating more container instances, up to the configured maximum, preventing cold starts and request queuing.

Exam trap

The PCD exam often tests the misconception that Cloud Load Balancing is required for scaling Cloud Run, but Cloud Run's built-in autoscaling and managed HTTPS load balancer already handle traffic spikes; the trap is that candidates confuse external load balancing with internal scaling mechanisms.

Practice this question →
30
MCQeasy

A startup is building a REST API on Cloud Run. They expect unpredictable traffic spikes and want to ensure the service can scale from 0 to many instances automatically. What scaling configuration should they use?

A.Set max instances to 1 to control costs.
B.Set min instances to 0 and max instances to 1000.
C.Use manual scaling with a fixed number of instances.
D.Set min instances to 5 and max to 100.
AnswerB

This configuration allows the service to scale from zero to a high number as needed, handling spikes while minimizing cost during idle periods.

Why this answer

Cloud Run's autoscaling allows min instances to be set to 0, enabling the service to scale down to zero when idle (cost-efficient), and max instances to 1000 to handle unpredictable traffic spikes by scaling out horizontally. This configuration ensures the service can start from zero and automatically add instances up to the maximum limit as demand increases, which is ideal for unpredictable workloads.

Exam trap

The trap here is that candidates often confuse 'min instances' with 'max instances' or assume that setting min instances to 0 will cause the service to be unavailable during cold starts, but Cloud Run handles cold starts transparently, and the question specifically asks for scaling from 0 to many instances, which requires min=0 and a high max limit.

How to eliminate wrong answers

Option A is wrong because setting max instances to 1 prevents the service from scaling out beyond a single instance, which cannot handle traffic spikes and defeats the purpose of autoscaling. Option C is wrong because manual scaling with a fixed number of instances does not allow dynamic scaling from 0 or to many instances; it requires manual intervention to adjust capacity, which is unsuitable for unpredictable spikes. Option D is wrong because setting min instances to 5 forces at least 5 instances to run continuously, incurring cost even when there is no traffic, and does not allow scaling down to zero, which contradicts the requirement to scale from 0.

Practice this question →
31
Multi-Selectmedium

A company uses Cloud Spanner for a global application. They want to improve read performance for point-reads (individual row lookups). Which TWO strategies should they adopt?

Select 2 answers
A.Use read replicas
B.Create secondary indexes
C.Partition the table by time
D.Use batch reads
E.Use interleaved tables
AnswersB, E

Secondary indexes enable efficient point reads on columns other than the primary key.

Why this answer

Secondary indexes in Cloud Spanner allow point-reads to be served directly from the index table, avoiding a full table scan and reducing latency. Interleaved tables store child rows physically adjacent to their parent row, enabling efficient single-row lookups without cross-node coordination.

Exam trap

The PCD exam often tests the misconception that read replicas or batch operations improve point-read latency, when in fact they address throughput or bulk retrieval, not the speed of individual row lookups.

Practice this question →
32
MCQmedium

A company runs a global e-commerce platform on GKE. They need to serve users with low latency from multiple regions. Which load balancing solution should they use?

A.Regional external HTTP(S) Load Balancer
B.Global external HTTP(S) Load Balancer
C.Internal TCP/UDP Load Balancer
D.SSL Proxy Load Balancer
AnswerB

Global load balancer routes users to the nearest region, minimizing latency.

Why this answer

A global external HTTP(S) Load Balancer is the correct choice because it provides a single anycast IP address that routes traffic from users worldwide to the nearest GKE backend, minimizing latency. It supports cross-regional failover and integrates with Cloud CDN for caching static content, making it ideal for a global e-commerce platform. Regional load balancers cannot serve traffic across multiple regions with a single IP, which is required for global low-latency access.

Exam trap

The PCD exam often tests the misconception that a Regional external HTTP(S) Load Balancer can be used for global traffic by simply deploying it in one region, but the trap is that it lacks anycast IP and cannot route users to the nearest region, causing higher latency for distant users.

How to eliminate wrong answers

Option A is wrong because a Regional external HTTP(S) Load Balancer only distributes traffic within a single GCP region, so it cannot serve users globally with low latency from multiple regions. Option C is wrong because an Internal TCP/UDP Load Balancer is designed for private VPC traffic within a region and does not expose a public endpoint for external users. Option D is wrong because an SSL Proxy Load Balancer terminates SSL/TLS connections but does not provide global anycast IP or HTTP(S) content-based routing; it is limited to TCP traffic and lacks the global scope needed for multi-region user distribution.

Practice this question →
33
MCQeasy

A company is designing a global e-commerce platform on Google Cloud. The application requires low-latency access for users worldwide and must be highly available. Which load balancing solution should they use?

A.External TCP/UDP Network Load Balancer
B.External HTTP(S) Load Balancer
C.Cloud CDN
D.Internal TCP/UDP Load Balancer
AnswerB

External HTTP(S) Load Balancer is a global load balancer that provides low latency and high availability for web applications.

Why this answer

The External HTTP(S) Load Balancer is the correct choice because it is a global, proxy-based Layer 7 load balancer that terminates HTTP/HTTPS traffic at Google's edge points of presence (PoPs) and routes requests to the nearest healthy backend. This provides low-latency access for users worldwide by leveraging Google's global network and anycast IPs, while also offering built-in high availability, SSL offloading, and content-based routing.

Exam trap

The PCD exam often tests the misconception that a Layer 4 load balancer (like External TCP/UDP Network Load Balancer) is sufficient for global low-latency access, but candidates must remember that only Layer 7 global load balancers provide anycast IPs and cross-region routing for worldwide users.

How to eliminate wrong answers

Option A is wrong because the External TCP/UDP Network Load Balancer is a regional, Layer 4 load balancer that does not provide global anycast IP or cross-region failover, so it cannot deliver low-latency access for users worldwide. Option C is wrong because Cloud CDN is a content delivery network that caches static content at edge locations, not a load balancer; it can be used in conjunction with a load balancer but does not itself handle traffic distribution or high availability for dynamic requests. Option D is wrong because the Internal TCP/UDP Load Balancer is a regional, private load balancer designed for internal traffic within a VPC, not for global external user access.

Practice this question →
34
MCQeasy

A company runs a batch job that processes large files from Cloud Storage every night. The job must complete within a 2-hour window. If the job fails, it should retry automatically. Which Google Cloud service should they use to orchestrate this job?

A.Compute Engine with startup script
B.Cloud Run
C.App Engine Cron
D.Cloud Composer
AnswerD

Cloud Composer is a managed workflow orchestration service that supports scheduling, retries, and complex dependencies, ideal for batch jobs.

Why this answer

Cloud Composer (D) is the correct choice because it is a fully managed workflow orchestration service built on Apache Airflow, designed to schedule, monitor, and retry batch jobs with complex dependencies. It can trigger a Cloud Storage file processing job, enforce a 2-hour execution window, and automatically retry on failure using Airflow's built-in retry mechanisms and SLA monitoring.

Exam trap

The PCD exam often tests the distinction between simple scheduling (App Engine Cron) and full orchestration with retry and dependency management (Cloud Composer), leading candidates to pick App Engine Cron because they overlook the requirement for automatic retry and time-window enforcement.

How to eliminate wrong answers

Option A is wrong because Compute Engine with a startup script is a manual, single-instance solution that lacks built-in scheduling, retry logic, and orchestration capabilities; it would require custom scripting and external cron to handle failures and time windows. Option B is wrong because Cloud Run is a serverless container platform for request-driven or event-driven workloads, not designed for long-running batch orchestration with retry policies and time-window enforcement; it lacks native workflow sequencing and retry orchestration. Option C is wrong because App Engine Cron is a simple scheduling service that triggers HTTP endpoints at fixed intervals, but it does not provide retry logic, dependency management, or execution time-window enforcement; it cannot automatically retry a failed job or ensure completion within a 2-hour window.

Practice this question →
35
Multi-Selectmedium

A developer is building an event-driven system using Cloud Pub/Sub. They need to ensure reliable message delivery and processing. Which three practices should they follow?

Select 3 answers
A.Set a minimum number of delivery attempts.
B.Use pull subscriptions with synchronous acknowledgment.
C.Use message ordering.
D.Configure a dead-letter topic.
E.Use exponential backoff for pull subscriptions.
AnswersB, D, E

Sync ack allows you to acknowledge after processing, preventing loss.

Why this answer

Pull subscriptions with synchronous acknowledgment (option B) ensure that a message is not acknowledged until the subscriber has successfully processed it. This prevents premature acknowledgment and message loss, because Cloud Pub/Sub will redeliver the message if the acknowledgment deadline expires without a synchronous ack. This is a core pattern for reliable processing.

Exam trap

The trap here is confusing reliability features with ordering or delivery attempt counts; The PCD exam often tests that reliable processing relies on synchronous acknowledgment and dead-letter topics, not on setting a minimum delivery attempts or enabling ordering.

Practice this question →
36
MCQhard

A company uses Cloud Storage for backups. They need to comply with a regulation requiring immutable storage for 7 years. Which bucket configuration should they use?

A.Use a bucket with a retention policy (not locked)
B.Set a lifecycle rule to archive to Coldline
C.Enable Object Versioning
D.Set a retention policy and lock the bucket
AnswerD

Locking the retention policy makes it permanent, ensuring objects cannot be deleted or overwritten for the specified duration.

Why this answer

Locking a retention policy in Cloud Storage enforces immutable storage for the specified duration (7 years). Once locked, the retention policy cannot be removed or shortened, ensuring compliance with regulations that require data to be preserved in its original state and not modifiable or deletable until the retention period expires.

Exam trap

The trap here is that candidates confuse a simple retention policy (which can be removed) with a locked retention policy (which is immutable), or they assume Object Versioning alone provides sufficient protection against deletion.

How to eliminate wrong answers

Option A is wrong because a retention policy that is not locked can be removed or shortened, which does not provide the immutable guarantee required by regulation. Option B is wrong because a lifecycle rule to archive to Coldline only moves data to a lower-cost storage class; it does not prevent deletion or modification of objects. Option C is wrong because Object Versioning alone does not prevent deletion of object versions; it only preserves previous versions when objects are overwritten or deleted, but versions can still be deleted manually or by lifecycle rules.

Practice this question →
37
MCQeasy

A startup is deploying a stateless web app on Compute Engine. They expect traffic spikes. What is the most cost-effective way to handle scaling?

A.Use App Engine Standard.
B.Use a single large VM with more cores.
C.Use managed instance groups with autoscaling based on CPU utilization.
D.Use Cloud Functions.
AnswerC

MIG with autoscaling scales horizontally and cost-effectively.

Why this answer

Managed instance groups (MIGs) with autoscaling based on CPU utilization are the most cost-effective solution for a stateless web app with traffic spikes because they automatically add or remove VM instances in response to real-time CPU load, ensuring you only pay for the compute resources you actually use. This approach directly matches the stateless nature of the app, allowing instances to be created and destroyed without data loss, and avoids over-provisioning or under-utilizing resources.

Exam trap

The PCD exam often tests the misconception that serverless options like App Engine or Cloud Functions are always the most cost-effective for any web app, but the trap here is that for a stateless web app with traffic spikes, managed instance groups with autoscaling provide finer control over scaling behavior and can be more cost-effective than paying for always-on App Engine instances or the per-invocation cost of Cloud Functions for sustained HTTP traffic.

How to eliminate wrong answers

Option A is wrong because App Engine Standard, while autoscaling, is a fully managed platform that can be more expensive for sustained traffic spikes due to its pricing model (per-instance-hour plus resource usage) and may introduce vendor lock-in or scaling limits (e.g., 10 concurrent requests per instance by default). Option B is wrong because a single large VM with more cores is a vertical scaling approach that has a hard upper limit (maximum machine size), creates a single point of failure, and is not cost-effective as you pay for idle capacity during low traffic. Option D is wrong because Cloud Functions is a serverless compute service designed for event-driven, short-lived tasks (max 9 minutes execution time, 60 minutes for HTTP functions) and is not suitable for running a persistent web app that requires continuous HTTP serving and state management across requests.

Practice this question →
38
MCQmedium

A company is designing a real-time leaderboard for a mobile gaming application. The leaderboard must support millions of concurrent users updating their scores and querying rankings with low latency (under 100ms). Scores change frequently and require strong consistency for reads. The development team is evaluating Cloud SQL and Cloud Spanner. They estimate they need to handle 100,000 writes per second. Which database should they choose and why?

A.Cloud Firestore because it offers real-time synchronization and is serverless.
B.Cloud Bigtable because it's optimized for high write throughput and time-series data.
C.Cloud SQL with read replicas because it's cost-effective and supports ACID transactions.
D.Cloud Spanner because it provides horizontal scaling, strong consistency, and high write throughput.
AnswerD

Spanner is built for high-throughput, strongly consistent global workloads.

Why this answer

Cloud Spanner is the correct choice because it provides horizontal scaling with strong consistency and can handle 100,000 writes per second while maintaining ACID transactions and low-latency reads. Unlike Cloud SQL, Spanner scales horizontally across nodes without sacrificing consistency, making it ideal for a real-time leaderboard with millions of concurrent users.

Exam trap

The PCD exam often tests the misconception that Cloud SQL can scale writes via read replicas, but read replicas only offload read traffic, not write throughput, and Cloud SQL's single-primary architecture cannot handle 100,000 writes per second.

How to eliminate wrong answers

Option A is wrong because Cloud Firestore is a NoSQL document database optimized for mobile and web apps with real-time sync, but it does not support the required 100,000 writes per second with strong consistency for reads—it offers eventual consistency by default and has a write limit of 10,000 writes per second per database. Option B is wrong because Cloud Bigtable is optimized for high write throughput and time-series data but does not support strong consistency for reads (it provides eventual consistency) and lacks ACID transactions, which are required for a leaderboard with frequent score updates. Option C is wrong because Cloud SQL with read replicas cannot horizontally scale to 100,000 writes per second—it is limited by the primary instance's write capacity (typically up to tens of thousands of writes per second) and read replicas do not improve write throughput; additionally, strong consistency for reads would require reading from the primary, increasing latency.

Practice this question →
39
MCQhard

A company running a high-traffic e-commerce platform on Google Cloud experiences occasional data loss in their Cloud SQL database during failover events. The database is configured with a failover replica in a different zone. What is the most likely cause of the data loss?

A.Automated backups are not enabled.
B.The database is using asynchronous replication to the failover replica.
C.The failover replica is configured as a read replica instead of a failover replica.
D.The database is not using regional persistent disks.
AnswerB

Asynchronous replication may not have replicated the most recent transactions before failover.

Why this answer

Cloud SQL uses synchronous replication for failover replicas by default, ensuring that transactions are committed on both the primary and the replica before acknowledging the write. If asynchronous replication is configured, the replica may lag behind the primary, and during a failover, any transactions not yet replicated are lost. This is the most likely cause of data loss during failover events.

Exam trap

The PCD exam often tests the distinction between synchronous and asynchronous replication in the context of failover replicas, where candidates mistakenly assume all replicas are synchronous by default or confuse failover replicas with read replicas.

How to eliminate wrong answers

Option A is wrong because automated backups are for point-in-time recovery and do not affect data loss during failover events; they are unrelated to replication consistency. Option C is wrong because a read replica cannot be promoted to a primary during failover; the question specifies a failover replica is configured, so this misconfiguration would prevent failover entirely, not cause data loss. Option D is wrong because regional persistent disks provide zonal redundancy for storage, but Cloud SQL failover replicas already use separate zones; the data loss is due to replication lag, not disk durability.

Practice this question →
40
MCQeasy

A development team is deploying a new application on Cloud Run. They anticipate unpredictable traffic patterns and want to minimize cold start latency. They also need to ensure that the application can handle sudden spikes without request drops. Which configuration should they use?

A.Use App Engine Standard Environment with automatic scaling.
B.Set min-instances to a non-zero value to keep some instances warm, and enable CPU always-on.
C.Set min-instances to 0 and max-instances to a high number to allow scaling from zero.
D.Use Cloud Functions instead of Cloud Run for better cold start performance.
AnswerB

Min-instances keeps containers warm; CPU always-on prevents cold start latency.

Why this answer

Setting min-instances to a non-zero value ensures that Cloud Run always keeps at least that many instances warm, eliminating cold starts for baseline traffic. Enabling CPU always-on prevents the instance's CPU from being throttled to zero when idle, allowing the instance to handle incoming requests immediately without a cold start penalty. This combination minimizes latency for unpredictable traffic and ensures capacity to absorb sudden spikes without dropping requests.

Exam trap

The PCD exam often tests the misconception that setting min-instances to 0 is acceptable for minimizing cold starts, or that switching to a different serverless product like Cloud Functions inherently solves cold start issues, when in fact the correct approach is to keep instances warm with min-instances and CPU always-on.

How to eliminate wrong answers

Option A is wrong because App Engine Standard Environment with automatic scaling does not provide the same fine-grained control over minimum instances and CPU always-on as Cloud Run, and it can still experience cold starts when scaling from zero. Option C is wrong because setting min-instances to 0 allows instances to scale down to zero, which guarantees cold starts on every new request after idle periods, directly contradicting the requirement to minimize cold start latency. Option D is wrong because Cloud Functions also suffers from cold starts (often worse than Cloud Run) and does not offer a min-instances or CPU always-on feature to keep instances warm; the recommendation to switch to Cloud Functions would not solve the cold start problem.

Practice this question →
41
MCQmedium

A developer is building a Cloud Pub/Sub-based event-driven system. They need to ensure that messages are processed at least once, and they want to handle processing failures. What should they do?

A.Use pull subscriptions with auto-acknowledgment
B.Configure max delivery attempts on the subscription
C.Use Cloud Tasks instead of Pub/Sub
D.Use push subscriptions with a dead-letter topic
AnswerD

Push subscriptions with a dead-letter topic provide retries and failure handling.

Why this answer

Using push subscriptions with a dead-letter topic ensures at-least-once delivery and provides a mechanism to handle processing failures. When a push subscription fails to deliver a message (e.g., due to a downstream error), Pub/Sub automatically retries delivery. After exhausting the maximum delivery attempts (default 5), the message is forwarded to a dead-letter topic, where it can be analyzed or reprocessed without losing the message.

This guarantees that every message is either processed successfully or stored for manual intervention, satisfying the at-least-once requirement.

Exam trap

The PCD exam often tests the misconception that simply increasing delivery attempts (Option B) is sufficient for failure handling, but the trap is that without a dead-letter topic, messages are permanently lost after the final attempt, violating the at-least-once requirement.

How to eliminate wrong answers

Option A is wrong because auto-acknowledgment (i.e., acknowledging immediately upon receipt) can cause messages to be lost if processing fails after acknowledgment, violating the at-least-once guarantee. Option B is wrong because configuring max delivery attempts on the subscription alone does not handle failures; without a dead-letter topic, messages that exceed the max attempts are simply dropped and lost. Option C is wrong because Cloud Tasks is designed for HTTP-based task execution with at-least-once delivery, but it lacks the native dead-lettering and pub-sub decoupling that Pub/Sub provides for event-driven systems; using Cloud Tasks would introduce unnecessary complexity and not directly address the failure handling requirement as effectively as a dead-letter topic.

Practice this question →
42
MCQhard

A company uses Cloud SQL for MySQL and wants to achieve high availability with automatic failover across zones while minimizing data loss. Which configuration should they use?

A.Enable read replicas in different zones
B.Use external read replicas with a failover script
C.Use Cloud SQL Enterprise Plus edition
D.Configure a regional Cloud SQL instance with automatic failover
E.Enable point-in-time recovery
AnswerD

Provides zone-level failover with synchronous replication, minimal data loss.

Why this answer

A regional Cloud SQL instance with automatic failover provides synchronous replication of data between two zones within the same region, ensuring zero data loss (RPO=0) and automatic failover with minimal downtime (RTO typically under 60 seconds). This meets the requirement for high availability with automatic failover across zones while minimizing data loss.

Exam trap

The PCD exam often tests the distinction between read replicas (asynchronous, for scaling) and regional instances (synchronous, for HA), leading candidates to mistakenly choose read replicas for high availability.

How to eliminate wrong answers

Option A is wrong because read replicas are asynchronous and do not provide automatic failover; they are designed for read scaling, not high availability with automatic failover. Option B is wrong because external read replicas require manual failover scripting and introduce latency and complexity, and they cannot guarantee minimal data loss due to asynchronous replication. Option C is wrong because Cloud SQL Enterprise Plus edition is a pricing tier that offers improved performance and availability features, but it does not itself enable regional failover; you must still configure a regional instance.

Option E is wrong because point-in-time recovery (PITR) is a backup feature for recovering to a specific timestamp, not a mechanism for automatic failover or high availability.

Practice this question →
43
MCQhard

A multi-region application uses Cloud Spanner. The team needs to ensure that a write is immediately visible to all subsequent reads, even those performed in different regions. Which consistency mode should they use?

A.Eventual consistency
B.Global consistency
C.Bounded staleness
D.Strong consistency
AnswerD

Cloud Spanner offers strong consistency by default, ensuring all reads reflect the most recent write.

Why this answer

Strong consistency (D) ensures that once a write is acknowledged, any subsequent read, regardless of region, will reflect that write. Cloud Spanner uses the TrueTime API and Paxos-based replication to provide external consistency (a form of strong consistency) across regions, making it the correct choice for immediate global visibility.

Exam trap

The PCD exam often tests the distinction between 'strong consistency' and 'global consistency' to trap candidates who assume 'global' is a valid Spanner mode, when in fact Spanner uses 'strong' or 'external' consistency for cross-region reads.

How to eliminate wrong answers

Option A is wrong because eventual consistency allows a delay before writes are visible to all readers, which violates the requirement for immediate visibility. Option B is wrong because 'Global consistency' is not a defined consistency mode in Cloud Spanner; the correct term is 'strong consistency' or 'external consistency'. Option C is wrong because bounded staleness allows reads to see data that is up to a specified time in the past, which does not guarantee immediate visibility of the most recent write.

Practice this question →
44
MCQhard

An online gaming platform uses Cloud Spanner as its globally distributed database. They notice that write latency increases significantly during peak hours. The application performs many single-row writes with high consistency requirements. Which design change would most effectively reduce write latency?

A.Increase the number of nodes in the Spanner instance.
B.Use interleaved tables to colocate related rows.
C.Switch to eventual consistency mode for writes.
D.Split the table into multiple smaller tables.
AnswerB

Interleaved tables store parent and child rows in the same split, reducing the number of participants in a transaction and decreasing write latency.

Why this answer

Interleaved tables in Cloud Spanner physically colocate parent and child rows, reducing the number of splits and cross-node round trips for related single-row writes. This minimizes distributed transaction overhead and write latency, especially under high consistency requirements, without requiring additional nodes or sacrificing consistency.

Exam trap

The trap here is that candidates often assume scaling nodes (Option A) is the universal fix for latency, but Cloud Spanner's write latency is dominated by distributed coordination, not node count, making interleaved tables a more targeted solution.

How to eliminate wrong answers

Option A is wrong because increasing nodes primarily improves read throughput and storage capacity, not write latency; in fact, more nodes can increase distributed coordination overhead for single-row writes. Option C is wrong because Cloud Spanner does not support eventual consistency for writes—it always provides strong external consistency via the TrueTime API, and switching consistency models is not a valid design change. Option D is wrong because splitting a table into multiple smaller tables does not reduce write latency; it can increase the number of distributed transactions and cross-node coordination, worsening latency.

Practice this question →
45
MCQmedium

A company runs a stateful application on Compute Engine instances with persistent disks. The application must be highly available and be able to recover from a zonal failure with minimal data loss. The current architecture uses a single instance in one zone. Which design should the team implement?

A.Use a standard persistent disk and configure a global load balancer to failover.
B.Create a snapshot schedule and restore the snapshot to a new instance in another zone on failure.
C.Use a regional persistent disk attached to a managed instance group across two zones.
D.Migrate to Cloud Filestore for shared file storage across zones.
AnswerC

Regional persistent disks replicate synchronously across zones, enabling fast failover.

Why this answer

A regional persistent disk synchronously replicates data across two zones, and when attached to a managed instance group (MIG) spanning those zones, it provides automatic failover with minimal data loss. This design ensures that if one zone fails, the MIG can detach the disk from the failed instance and attach it to a healthy instance in the surviving zone, preserving state with near-zero RPO.

Exam trap

The trap here is that candidates often confuse high availability with backup strategies (snapshots) or assume that a load balancer alone can handle storage failover, failing to recognize that stateful applications require synchronous data replication across zones to achieve minimal data loss.

How to eliminate wrong answers

Option A is wrong because a standard persistent disk is zonal, not regional, and a global load balancer alone cannot failover the disk or its data; the load balancer handles traffic but the disk remains tied to the original zone, so a zonal failure still causes data loss. Option B is wrong because snapshot schedules are asynchronous and point-in-time, meaning any data written between the last snapshot and the failure is lost, resulting in higher RPO than the minimal data loss requirement. Option D is wrong because Cloud Filestore is a managed NFS service designed for shared file storage, not for block-level persistent disks; it introduces network latency and does not provide the same low-level synchronous replication as a regional persistent disk, and it is not directly attachable to Compute Engine instances as a boot disk.

Practice this question →
46
MCQeasy

A developer is designing a serverless event-driven application that processes messages from Pub/Sub and writes results to BigQuery. The workload is unpredictable but must scale to zero when idle. Which compute option should they choose?

A.Cloud Run with Pub/Sub push subscription
B.Cloud Functions with Pub/Sub trigger
C.Compute Engine with managed instance groups
D.Google Kubernetes Engine with Horizontal Pod Autoscaler
AnswerB

Cloud Functions is serverless, scales to zero, and has native Pub/Sub integration.

Why this answer

Cloud Functions with a Pub/Sub trigger is the correct choice because it is purpose-built for event-driven, serverless workloads that scale to zero when idle. It automatically scales from zero to thousands of concurrent invocations based on the volume of Pub/Sub messages, and it natively integrates with Pub/Sub via a background function that is invoked for each message, making it ideal for unpredictable, bursty workloads that must process messages and write results to BigQuery.

Exam trap

The PCD exam often tests the misconception that Cloud Run is always the best serverless option, but the trap here is that Cloud Functions is the native, simpler choice for pure event-driven Pub/Sub processing, while Cloud Run is better suited for HTTP request-driven workloads or when you need longer request timeouts or custom runtimes.

How to eliminate wrong answers

Option A is wrong because Cloud Run with a Pub/Sub push subscription requires a running container instance to receive push requests, and while it can scale to zero, it introduces additional latency and complexity compared to a native Pub/Sub trigger, and it is not the simplest or most cost-effective choice for a purely event-driven, message-processing workload. Option C is wrong because Compute Engine with managed instance groups requires provisioning and maintaining VMs, does not scale to zero (minimum 1 VM), and incurs costs even when idle, making it unsuitable for a serverless, scale-to-zero requirement. Option D is wrong because Google Kubernetes Engine with Horizontal Pod Autoscaler requires a running cluster with node pools, does not scale to zero (minimum 1 node), and adds operational overhead for managing Kubernetes infrastructure, which is unnecessary for a simple Pub/Sub-to-BigQuery pipeline.

Practice this question →
47
MCQeasy

A media company wants to serve video content globally with low latency and high throughput. Which Google Cloud service is best suited?

A.Cloud CDN
B.Cloud Load Balancer
C.Cloud Storage with public bucket
D.App Engine
AnswerA

Cloud CDN provides global content caching at edge locations, ensuring low latency and high throughput.

Why this answer

Cloud CDN leverages Google's global edge cache network to deliver video content from locations closest to end users, minimizing latency and offloading origin servers. It integrates with Cloud Load Balancer and Cloud Storage to provide high-throughput, low-latency streaming without requiring users to manage caching infrastructure.

Exam trap

The trap here is confusing load balancing (traffic distribution) with content delivery (caching at edge), leading candidates to choose Cloud Load Balancer when the question explicitly asks for low latency and high throughput for global video serving.

How to eliminate wrong answers

Option B is wrong because Cloud Load Balancer distributes traffic across backends but does not cache content; it alone cannot reduce latency for repeated requests or offload origin servers. Option C is wrong because Cloud Storage with a public bucket serves content directly from a single regional bucket, resulting in higher latency for global users and no edge caching to improve throughput. Option D is wrong because App Engine is a compute platform for hosting applications, not a content delivery service; it lacks built-in edge caching and global distribution optimized for video streaming.

Practice this question →
48
MCQhard

A developer finds the JSON key shown in the exhibit in a Cloud Storage bucket that is publicly accessible. Which security best practice was violated?

A.The key is not rotated regularly.
B.The key was created as a user-managed key instead of a Google-managed key.
C.The key was not encrypted using Cloud KMS.
D.The key was stored in a publicly accessible Cloud Storage bucket.
AnswerD

Service account keys must be kept confidential and never exposed publicly.

Why this answer

Storing a JSON key (a service account private key) in a publicly accessible Cloud Storage bucket directly violates the principle of least privilege and exposes sensitive credentials to unauthorized users. Any entity with read access to the bucket can retrieve the key and impersonate the service account, potentially gaining unauthorized access to Google Cloud resources.

Exam trap

The PCD exam often tests the distinction between encryption (which protects data at rest) and access control (which governs who can read the data), leading candidates to mistakenly choose an encryption-related option when the real issue is public exposure.

How to eliminate wrong answers

Option A is wrong because while key rotation is a security best practice, the violation here is the public exposure of the key, not the lack of rotation. Option B is wrong because the key type (user-managed vs. Google-managed) is irrelevant to the immediate security breach; the issue is the public accessibility of the bucket, not the key's management origin.

Option C is wrong because Cloud KMS encryption protects data at rest, but the key is already exposed by being in a public bucket; encryption does not prevent unauthorized access if the bucket permissions are misconfigured.

Practice this question →
49
MCQmedium

A company is designing a microservices application. They want to ensure that if one service fails, it does not cascade to other services. Which pattern should they implement?

A.Auto-scaling
B.Retry with exponential backoff
C.Load shedding
D.Circuit Breaker
AnswerD

Circuit breaker stops calls to a failing service, preventing cascade.

Why this answer

The Circuit Breaker pattern is the correct choice because it prevents cascading failures by monitoring service calls and opening the circuit when failures exceed a threshold, allowing the system to fail fast and avoid resource exhaustion. This pattern directly addresses the requirement to isolate failures between microservices, ensuring that a failure in one service does not propagate to others.

Exam trap

The PCD exam often tests the misconception that retry mechanisms or load shedding are sufficient for failure isolation, but they do not prevent cascading failures because they lack the stateful tripping and fast-fail behavior of the Circuit Breaker pattern.

How to eliminate wrong answers

Option A is wrong because Auto-scaling handles increased load by adding instances but does not prevent failure propagation between services. Option B is wrong because Retry with exponential backoff can actually worsen cascading failures by repeatedly attempting calls to a failing service, potentially overwhelming it further. Option C is wrong because Load shedding drops excess requests to protect a service from overload but does not isolate failures from propagating to dependent services.

Practice this question →
50
MCQmedium

A company runs a critical financial application on Google Cloud using Compute Engine instances in a managed instance group (MIG) with auto-scaling based on CPU utilization. The application stores state in a local SSD and relies on sticky sessions (session affinity). Recently, during a traffic spike, the MIG scaled out new instances, but some users lost their sessions because the load balancer routed them to a different instance. The team needs to maintain session persistence without sacrificing scalability. What should they do?

A.Implement a shared session store using Cloud Memorystore for Redis.
B.Increase the instance group's cooldown period to reduce scaling frequency.
C.Use a global HTTPS Load Balancer with cookie-based session affinity.
D.Use Cloud NAT for consistent source IP routing.
AnswerA

External session store makes sessions available to all instances.

Why this answer

Using Cloud Memorystore for Redis provides a centralized, external session store that decouples session state from individual Compute Engine instances. This ensures that any instance in the managed instance group can serve any user's request, maintaining session persistence even as the MIG scales out or in based on CPU utilization. It preserves scalability because the session store is independent of instance lifecycle, and Redis offers low-latency reads and writes suitable for session data.

Exam trap

The PCD exam often tests the misconception that session affinity alone is sufficient for session persistence, but the trap here is that candidates overlook the need for a shared external store when instances are ephemeral or can be terminated, as local SSD state is lost on instance stop/termination.

How to eliminate wrong answers

Option B is wrong because increasing the cooldown period only delays the scaling of new instances, which does not solve the fundamental problem of session state being stored locally on instances; users will still lose sessions if they are routed to a different instance after scaling. Option C is wrong because while a global HTTPS Load Balancer with cookie-based session affinity can route a user to the same instance, it does not address the issue that the session data is stored on a local SSD; if the instance is terminated or scaled down, the session is lost, and session affinity cannot guarantee persistence across instance failures or scaling events. Option D is wrong because Cloud NAT provides outbound internet connectivity with a consistent source IP for instances, but it does not affect how the load balancer routes incoming traffic or how session state is stored; it is irrelevant to session persistence.

Practice this question →
51
Matchingmedium

Match each Cloud SQL database engine to its description.

Drag a concept onto its matching description — or click a concept then click the description.

Concepts
Matches

Open-source relational database

Advanced open-source relational database

Microsoft relational database with Windows integration

PostgreSQL-compatible with high performance for transactions

Globally distributed, strongly consistent relational database

Why these pairings

Cloud SQL offers managed relational databases; AlloyDB and Spanner are for higher scale.

Practice this question →
52
MCQmedium

A company runs a stateful microservice that requires read-after-write consistency but can tolerate some latency for writes. They are currently using a single Cloud SQL instance and want to scale read traffic. Which approach should they take?

A.Use Cloud Memorystore to cache reads
B.Shard the database manually
C.Enable Cloud SQL read replicas
D.Use Cloud Bigtable
E.Migrate to Cloud Spanner
AnswerC

Scales read capacity with eventual consistency, good for the described needs.

Why this answer

Cloud SQL read replicas are the correct choice because they provide an asynchronous read-only copy of the primary instance, which can scale read traffic without compromising the read-after-write consistency required by the stateful microservice. The primary instance handles all writes, ensuring strong consistency for writes, while replicas serve stale reads that eventually become consistent, which aligns with the tolerance for write latency.

Exam trap

The PCD exam often tests the misconception that caching (Memorystore) is the default solution for scaling reads, but the trap here is that caching does not guarantee read-after-write consistency, whereas read replicas can be configured to serve stale reads while the primary maintains strong consistency for writes.

How to eliminate wrong answers

Option A is wrong because Cloud Memorystore (Redis/Memcached) caches data in memory, but it does not guarantee read-after-write consistency — a write to Cloud SQL may not be immediately reflected in the cache, leading to stale reads. Option B is wrong because manual sharding distributes data across multiple databases, which complicates consistency guarantees and requires application-level logic to maintain read-after-write consistency, increasing complexity and risk. Option D is wrong because Cloud Bigtable is a NoSQL wide-column store optimized for high-throughput, low-latency analytics, not for transactional workloads requiring strong read-after-write consistency.

Option E is wrong because Cloud Spanner provides strong global consistency and horizontal scaling, but it is overkill for this scenario — it introduces higher cost and complexity when a simpler read replica solution suffices.

Practice this question →
53
MCQeasy

What is the primary benefit of using Cloud Load Balancing with global anycast IP?

A.Provides DDoS protection
B.Supports WebSocket
C.Reduces latency for users worldwide
D.Enables cross-zone failover
AnswerC

Anycast directs traffic to the closest region, minimizing network hops and latency.

Why this answer

Cloud Load Balancing with a global anycast IP directs user traffic to the nearest available backend instance based on network topology and latency. This minimizes the number of network hops and reduces round-trip time, providing lower latency for users worldwide compared to a single-region deployment.

Exam trap

The PCD exam often tests the misconception that global anycast IP is primarily for DDoS protection or that it provides cross-zone failover, when in fact its core benefit is latency reduction via proximity-based routing.

How to eliminate wrong answers

Option A is wrong because while Cloud Load Balancing can absorb some volumetric attacks due to its scale, its primary benefit is not DDoS protection; dedicated services like Cloud Armor or third-party DDoS mitigation are designed for that purpose. Option B is wrong because WebSocket support is a feature of the load balancer's protocol handling (e.g., HTTP/2 or TCP proxy), not a benefit specific to global anycast IP. Option D is wrong because cross-zone failover is a regional capability that ensures high availability within a single region; global anycast IP enables multi-region failover and traffic steering, not cross-zone failover.

Practice this question →
54
MCQeasy

A company wants to run a batch job every hour that processes files from Cloud Storage. The job takes about 10 minutes. Which serverless option should they use?

A.Cloud Run jobs
B.Cloud Functions with Cloud Scheduler
C.Compute Engine with cron
D.App Engine Cron Service with Cloud Tasks
AnswerB

Cloud Functions triggered by Cloud Scheduler is serverless and simple for periodic tasks.

Why this answer

Cloud Functions triggered by Cloud Scheduler is ideal for periodic, short-lived batch jobs that process files. Cloud Run Jobs is also suitable but less event-driven. Compute Engine requires manual setup.

App Engine Cron Service is possible but more complex.

Practice this question →
55
Multi-Selecthard

A team is building a serverless event-driven application using Cloud Functions and Cloud Pub/Sub. The function processes messages from a Pub/Sub subscription and writes results to Firestore. During peak hours, the function experiences high latency and some messages are being retried multiple times. Which three steps should the team take to improve reliability and scalability? (Choose three.)

Select 3 answers
A.Enable retry policy on the Pub/Sub subscription to automatically retry failed messages.
B.Batch multiple Pub/Sub messages into a single Cloud Function invocation.
C.Configure the Cloud Function with a min instance count and increase max instances.
D.Increase the Cloud Function timeout to the maximum allowed value.
E.Set a longer acknowledgement deadline for the subscription to allow more processing time.
AnswersA, C, E

Retry policy ensures messages are not lost and are retried until successful.

Why this answer

Enabling a retry policy on the Pub/Sub subscription ensures that messages that fail to be processed (e.g., due to transient errors or timeouts) are automatically retried. This prevents message loss and improves reliability by allowing the Cloud Function to reprocess messages without manual intervention. The retry policy works with the subscription's acknowledgement deadline, so messages are redelivered if not acknowledged in time.

Exam trap

The PCD exam often tests the misconception that increasing timeout or batching messages are universal fixes for latency, when in fact serverless scaling and proper acknowledgement handling are the correct levers for reliability and scalability.

Practice this question →
56
MCQmedium

A team runs a microservice on Compute Engine behind a regional external HTTP load balancer. They want to automatically replace unhealthy instances without manual intervention. Which feature should they use?

A.Unmanaged instance group with health check
B.Instance template with manual replacement
C.Load balancer backend service health check only
D.Managed instance group with autoscaling and health check
AnswerD

Managed instance groups support autohealing, which automatically recreates instances based on health check results.

Why this answer

A managed instance group (MIG) with autoscaling and a health check is the correct choice because it automatically replaces unhealthy instances based on the health check results. The MIG uses the health check to detect failed instances, then automatically recreates them from the instance template, ensuring high availability without manual intervention. Autoscaling further adjusts the number of instances based on load, but the core replacement mechanism is driven by the MIG's health check and autohealing feature.

Exam trap

The trap here is that candidates often confuse the load balancer's health check (which only affects traffic routing) with the managed instance group's health check (which triggers automatic instance replacement), leading them to choose option C instead of D.

How to eliminate wrong answers

Option A is wrong because an unmanaged instance group does not support automatic replacement of unhealthy instances; it requires manual intervention to remove and add instances. Option B is wrong because an instance template is a configuration resource, not a mechanism for automatic replacement; it defines the VM configuration but does not provide any health-check-driven autohealing. Option C is wrong because a load balancer backend service health check alone only marks instances as unhealthy for traffic routing; it does not trigger instance replacement, which requires a managed instance group with autohealing.

Practice this question →
57
MCQeasy

An e-commerce company relies on a Compute Engine backend serving content to global users. They notice high latency for users outside the primary region. Which service should they add to reduce latency by caching content at edge locations?

A.Cloud Armor
B.Cloud NAT
C.Cloud CDN
D.Cloud Endpoints
E.Cloud Load Balancing
AnswerC

Caches static content globally near users, reducing latency.

Why this answer

Cloud CDN (Content Delivery Network) uses Google's global edge cache locations to serve cached content closer to users, reducing latency for requests that would otherwise travel to the origin Compute Engine backend in a single region. By caching static or dynamic content at edge nodes, Cloud CDN minimizes round-trip time and offloads traffic from the backend instance.

Exam trap

The PCD exam often tests the distinction between load balancing (which distributes traffic but does not cache) and CDN (which caches at edge locations), leading candidates to mistakenly choose Cloud Load Balancing because they associate it with global performance improvements.

How to eliminate wrong answers

Option A is wrong because Cloud Armor is a web application firewall (WAF) and DDoS protection service that filters traffic based on security rules, not a caching or content delivery service. Option B is wrong because Cloud NAT provides outbound internet connectivity for private instances via network address translation, it does not cache content or reduce latency for inbound user requests. Option D is wrong because Cloud Endpoints is an API management service that handles authentication, quotas, and monitoring for APIs, not a content caching or edge delivery solution.

Option E is wrong because Cloud Load Balancing distributes traffic across backend instances for high availability and scalability, but it does not cache content at edge locations; it still requires the request to reach the origin region.

Practice this question →
58
Multi-Selecthard

A team is deploying a critical application on Google Kubernetes Engine (GKE) and needs to ensure high availability and disaster recovery. Which THREE actions should they take?

Select 3 answers
A.Deploy all pods in a single zone for simplicity.
B.Use a regional cluster with control plane replicated across zones.
C.Distribute workloads across multiple zones using node affinity and anti-affinity.
D.Use a zonal cluster to reduce costs.
E.Configure PodDisruptionBudgets to ensure minimum pod availability.
AnswersB, C, E

Regional clusters replicate the control plane across multiple zones, providing high availability.

Why this answer

A regional cluster in GKE replicates the control plane across multiple zones within a region, ensuring that if one zone fails, the control plane remains available. This is essential for high availability and disaster recovery, as it eliminates a single point of failure for cluster management operations.

Exam trap

The PCD exam often tests the misconception that a zonal cluster is sufficient for disaster recovery because it is cheaper, but the trap is that a zonal cluster's control plane is not replicated, making it vulnerable to zonal failures, whereas a regional cluster provides the necessary redundancy for both control plane and workloads.

Practice this question →
59
Multi-Selecteasy

A company is designing a scalable web application on Google Cloud. They expect variable traffic and want to automatically scale resources based on load. Which two services can automatically scale? (Choose two.)

Select 2 answers
A.Compute Engine unmanaged instance group
B.Cloud Run
C.Compute Engine managed instance group
D.Cloud Dataproc
E.Cloud SQL
AnswersB, C

Cloud Run automatically scales container instances from zero to a maximum based on incoming request volume.

Why this answer

Cloud Run is a fully managed serverless compute platform that automatically scales your containerized applications based on incoming traffic, including scaling to zero when there is no traffic. This autoscaling is handled by the Knative serving layer, which adjusts the number of container instances based on request concurrency and CPU utilization.

Exam trap

The trap here is that candidates often confuse unmanaged instance groups with managed instance groups, assuming both support autoscaling, but only managed instance groups have built-in autoscalers.

Practice this question →
60
Multi-Selectmedium

A team uses Google Kubernetes Engine (GKE) with Node Auto-Provisioning. They want to optimize cost while maintaining high availability across zones. Which two strategies should they implement? (Select exactly 2.)

Select 2 answers
A.Use cluster autoscaler with appropriate min and max node counts
B.Spread node pools across multiple zones
C.Use preemptible VMs for all node pools
D.Disable cluster autoscaler to prevent scaling
E.Use sole-tenant nodes for high availability
AnswersA, B

The cluster autoscaler automatically adjusts node count based on demand, optimizing cost.

Why this answer

Node Auto-Provisioning (NAP) in GKE works in conjunction with the cluster autoscaler to automatically create and delete node pools based on workload demands. By setting appropriate minimum and maximum node counts, you ensure the cluster can scale down to zero when idle (saving cost) and scale up to handle peak load, while avoiding runaway scaling that could increase costs unexpectedly.

Exam trap

The trap here is that candidates often confuse preemptible VMs (cost-saving but low availability) with high availability, or assume that disabling the autoscaler prevents cost spikes, when in fact it leads to either over-provisioning or under-provisioning, both of which harm the dual goal of cost optimization and high availability.

Practice this question →
61
MCQhard

You are troubleshooting a web application deployed on Compute Engine instances behind a target pool. Users report intermittent timeouts when accessing the application via the forwarding rule's IP address. Based on the exhibit, what is the most likely cause of the issue?

A.The forwarding rule is missing a backend service.
B.The target pool lacks health checks, causing traffic to be sent to unhealthy instances.
C.The port range is set to 80-80, which restricts traffic to port 80 only.
D.The forwarding rule should use a backend service instead of a target pool for HTTP traffic.
AnswerB

Target pools rely on health checks to stop routing to unhealthy instances; without them, traffic may be routed to failed instances.

Why this answer

The target pool in a legacy HTTP(S) load balancer does not automatically perform health checks unless they are explicitly configured. Without health checks, the load balancer continues to send traffic to all instances in the pool, including those that are unhealthy or unresponsive. This causes intermittent timeouts when users hit an unhealthy instance, as the forwarding rule distributes connections across the entire pool without verifying instance health.

Exam trap

The PCD exam often tests the misconception that a forwarding rule's port range or the use of a target pool versus a backend service is the root cause of intermittent timeouts, when in fact the absence of health checks is the critical missing component.

How to eliminate wrong answers

Option A is wrong because the forwarding rule in this legacy setup is correctly configured with a target pool; a backend service is used only with the newer HTTP(S) load balancer (using instance groups), not with target pools. Option C is wrong because setting the port range to 80-80 is a valid configuration that restricts traffic to port 80, which is the intended behavior for an HTTP application, and does not cause intermittent timeouts. Option D is wrong because while using a backend service is a modern approach, the question describes a target pool configuration which is still valid for legacy HTTP load balancing; the issue is not the type of load balancer but the missing health checks.

Practice this question →
62
MCQhard

An administrator runs the above command to create a Compute Engine instance. However, the nginx service does not start. What is the most likely cause?

A.The instance has no external IP address and cannot reach the internet to download packages.
B.The metadata key is misspelled; it should be 'startup-script-url'.
C.The instance does not have the compute.instance.update permission.
D.The startup script runs before the boot disk is fully mounted.
AnswerA

By default, instances are created without an external IP unless --no-address is not specified. The command does not specify --no-address, but if the project's default is to not assign external IPs, the instance may lack internet access. However, in newer GCP projects, the default is to assign an ephemeral external IP. Actually, the default behavior depends on the project's VPC configuration. Without an external IP and without Cloud NAT, the instance cannot access the internet, causing apt-get to fail.

Why this answer

The command likely creates a Compute Engine instance without specifying an external IP address (e.g., using `--no-address` or omitting `--address`). Without an external IP, the instance cannot reach the internet to download the nginx package from repositories, causing the startup script that installs and starts nginx to fail. This is the most direct cause of the nginx service not starting.

Exam trap

The PCD exam often tests the nuance that startup scripts execute after the boot disk is mounted and that missing external IP prevents internet-dependent operations, leading candidates to incorrectly blame script syntax or permissions.

How to eliminate wrong answers

Option B is wrong because the metadata key 'startup-script-url' is valid for specifying a startup script stored in Cloud Storage; the question does not indicate a misspelling, and the script itself could be correct. Option C is wrong because the instance does not need the 'compute.instance.update' permission to run startup scripts; that permission is for modifying the instance metadata, not for executing scripts. Option D is wrong because the boot disk is fully mounted before the startup script runs; Compute Engine ensures the root filesystem is available before executing startup scripts.

Practice this question →
63
MCQeasy

A developer deployed the above Cloud Run service YAML. The service deploys successfully but any request fails with a 503 error. What is the most likely cause?

A.The container is not listening on the expected port.
B.The service has no ingress setting.
C.The container image has a different entrypoint.
D.containerConcurrency is set too high.
AnswerA

Cloud Run requires the container to listen on the port specified by the PORT environment variable (default 8080). If the container listens on a different port, requests time out or fail.

Why this answer

A 503 error from Cloud Run indicates that the service is failing to respond to health checks or requests. The most common cause is that the container is not listening on the port specified in the `containerPort` field of the YAML (default 8080). Cloud Run sends requests to that port, and if the application is bound to a different port (e.g., 3000 or 80), the request never reaches the application, resulting in a 503.

Exam trap

The PCD exam often tests the distinction between a container that fails to start (which would show a different error) and a container that runs but is unreachable on the expected port (which causes 503 errors).

How to eliminate wrong answers

Option B is wrong because Cloud Run services have a default ingress setting of 'all' (allowing all traffic) when not explicitly set, so missing ingress does not cause a 503. Option C is wrong because a different entrypoint would cause the container to fail to start or crash, resulting in a different error (e.g., 'Container failed to start' or 'CrashLoopBackOff'), not a 503 response. Option D is wrong because setting `containerConcurrency` too high (e.g., 80 or more) could cause performance degradation or timeouts under load, but it would not cause every request to fail with a 503; the service would still respond to some requests.

Practice this question →
64
Multi-Selecthard

A company runs a stateful application on Compute Engine. They need to achieve an RPO of less than 15 minutes and an RTO of less than 30 minutes for a regional disaster. Which three steps should they include in their disaster recovery plan? (Select exactly 3.)

Select 3 answers
A.Use a managed instance group in multiple zones within the same region
B.Develop custom scripts to replicate application data asynchronously to another region
C.Configure persistent disk snapshots to a different region
D.Use regional persistent disks to replicate data within the region
E.Configure Cloud DNS with geo-routing to direct traffic to a healthy region
AnswersB, C, E

Asynchronous replication to another region can meet RPO and allow failover to that region.

Why this answer

Asynchronous replication of application data to another region can achieve an RPO of less than 15 minutes and an RTO of less than 30 minutes, as it allows the application to fail over to a secondary region with minimal data loss. Custom scripts can control replication frequency and ensure data consistency, meeting the strict RPO requirement.

Exam trap

The PCD exam often tests the distinction between regional and multi-region disaster recovery, where candidates mistakenly choose intra-region solutions like regional persistent disks or multi-zone instance groups for a regional disaster scenario.

Practice this question →
65
MCQmedium

A company is running a global application on Cloud Spanner. They notice high write latency on a specific table because a frequently updated row is being accessed by many clients simultaneously. Which design pattern should they implement to distribute writes across multiple nodes and reduce contention?

A.Increase the number of nodes in the Cloud Spanner instance.
B.Use interleaved tables to co-locate related data.
C.Add a hash suffix to the primary key of the hot row to split it into multiple rows.
D.Migrate the table to Cloud Bigtable which handles hotspots better.
AnswerC

This distributes writes across multiple splits.

Why this answer

Adding a hash suffix to the primary key of the hot row splits the single heavily contended row into multiple logical rows, each with a different primary key. This distributes the write load across multiple Cloud Spanner splits and nodes, reducing lock contention and write latency. Cloud Spanner uses a distributed, synchronous replication architecture where a single row is managed by a single split; splitting the hot row into multiple rows allows parallel writes to different splits.

Exam trap

The PCD exam often tests the misconception that scaling up the instance (adding nodes) solves single-row contention, but the trap here is that Cloud Spanner's architecture requires splitting the hot row's key to distribute writes across splits, not just adding more nodes.

How to eliminate wrong answers

Option A is wrong because increasing the number of nodes in Cloud Spanner increases overall throughput and storage capacity, but does not resolve contention on a single hot row—that row is still managed by one split and one leader, so write latency remains high. Option B is wrong because interleaved tables co-locate parent and child rows for efficient joins and strong consistency, but they do not help with write contention on a single frequently updated row; they actually increase the likelihood of contention if the parent row is the hot row. Option D is wrong because migrating to Cloud Bigtable is not a recommended design pattern for this scenario; Bigtable handles hotspots via automatic sharding, but the question asks for a design pattern within Cloud Spanner, and Bigtable does not support global, strongly consistent transactions or SQL queries, which the application likely requires.

Practice this question →
66
MCQmedium

Refer to the exhibit. The Cloud Run service is experiencing high tail latency under moderate load. Which change would most effectively reduce latency?

A.Increase CPU limit to 2.
B.Increase containerConcurrency to 250.
C.Increase timeoutSeconds to 600.
D.Reduce containerConcurrency to 10.
AnswerD

Lower concurrency reduces request queuing per container, improving tail latency under load.

Why this answer

High tail latency under moderate load often indicates that requests are queuing behind each other due to excessive concurrency. Reducing `containerConcurrency` to 10 limits the number of simultaneous requests each container instance handles, which reduces queueing delay and improves per-request response time. This is the most effective change because it directly controls the request multiplexing level, preventing a single instance from being overwhelmed.

Exam trap

The PCD exam often tests the misconception that increasing resources (CPU/memory) or timeouts always improves performance, when in fact controlling concurrency is the key to reducing tail latency in serverless platforms like Cloud Run.

How to eliminate wrong answers

Option A is wrong because increasing the CPU limit to 2 does not address the root cause of tail latency; it may reduce compute-bound delays but does not control request queuing or concurrency pressure. Option B is wrong because increasing `containerConcurrency` to 250 would exacerbate the problem by allowing more simultaneous requests per instance, increasing queueing and tail latency. Option C is wrong because increasing `timeoutSeconds` to 600 only extends the maximum request duration, which does not reduce latency; it may even mask underlying performance issues by allowing slow requests to linger longer.

Practice this question →
67
MCQhard

A large e-commerce platform uses Cloud Bigtable to store user session data and product recommendations. They have a single cluster in a single zone. During a recent zone outage, the application became unavailable for 30 minutes because Cloud Bigtable was unreachable. The team needs to ensure high availability for the session data with a Recovery Time Objective (RTO) of less than 5 minutes and a Recovery Point Objective (RPO) of zero (no data loss). What should they do?

A.Migrate the session data to Cloud Memorystore for Redis with persistence and replication.
B.Add a second cluster in a different zone within the same region and use multi-cluster routing to automatically failover.
C.Configure replication to a second cluster in a different region and use global routing to failover.
D.Use Cloud Bigtable's single-cluster replication to a different zone.
AnswerB

Multi-cluster within region provides zone-level HA with fast replication.

Why this answer

Adding a second Cloud Bigtable cluster in a different zone within the same region and enabling multi-cluster routing provides automatic failover with an RTO of under 5 minutes and an RPO of zero. Multi-cluster routing directs read and write requests to the nearest healthy cluster, and replication between clusters is synchronous within a region, ensuring no data loss during a zone outage.

Exam trap

The PCD exam often tests the misconception that cross-region replication can achieve an RPO of zero, but candidates must remember that only intra-region replication is synchronous, while cross-region replication is asynchronous and introduces data loss risk.

How to eliminate wrong answers

Option A is wrong because Cloud Memorystore for Redis with persistence and replication does not guarantee an RPO of zero; asynchronous replication can lose recent writes during a failover, and it is not designed for the same throughput and latency characteristics as Cloud Bigtable for session data. Option C is wrong because configuring replication to a second cluster in a different region uses asynchronous replication, which cannot achieve an RPO of zero due to cross-region replication lag, and global routing introduces higher latency and potential data inconsistency. Option D is wrong because Cloud Bigtable does not support single-cluster replication; replication is always between two or more clusters, and the term 'single-cluster replication' is a misnomer that does not exist in Cloud Bigtable's architecture.

Practice this question →
68
MCQmedium

A development team is using Cloud Build to deploy containerized applications to GKE. They want to ensure that only containers that have passed security scans and unit tests are deployed to production. Which approach should they use?

A.Deploy to a staging cluster first, then manually promote to production using kubectl.
B.Use Cloud Build with a multi-step pipeline that includes test and security scan steps, and only promote to production after successful completion.
C.Use Cloud Deploy to automate delivery with approval gates.
D.Configure Cloud Build triggers to deploy directly to production on every push.
AnswerB

This ensures that only containers that pass all checks are deployed, maintaining quality and security.

Why this answer

Using a multi-step Cloud Build pipeline with test and security scan steps, and then promoting to production only after success, ensures only validated containers are deployed. Direct deployment to production on every push is risky. Manual promotion defeats automation.

Cloud Deploy adds unnecessary complexity for this simple requirement. Thus, option B is correct.

Practice this question →
69
MCQeasy

A developer needs to store session state for a user in a cloud-native application. Which storage solution is most appropriate?

A.Cloud SQL
B.Memorystore
C.Cloud Storage
D.Bigtable
AnswerB

Memorystore provides fast, in-memory caching for session data.

Why this answer

Memorystore (Redis) is the most appropriate solution for storing session state in a cloud-native application because it provides an in-memory data store with sub-millisecond latency, which is critical for fast session reads and writes. Session state is ephemeral, key-value data that requires high throughput and low latency, and Memorystore supports features like TTL (time-to-live) for automatic session expiration and persistence options for durability. This aligns with the cloud-native principle of stateless application tiers offloading state to a managed caching layer.

Exam trap

The PCD exam often tests the misconception that any managed database (like Cloud SQL or Bigtable) can handle session state, but the trap is that session state requires in-memory speed and automatic expiration, which only a caching solution like Memorystore provides, not disk-based or analytical databases.

How to eliminate wrong answers

Option A is wrong because Cloud SQL is a relational database designed for structured, transactional data with ACID compliance, not for high-speed ephemeral session state; its disk-based storage and connection overhead introduce latency unsuitable for frequent session lookups. Option C is wrong because Cloud Storage is an object store for blobs and files, not a low-latency key-value store; it lacks the sub-millisecond read/write performance and TTL-based expiration needed for session management. Option D is wrong because Bigtable is a wide-column NoSQL database optimized for analytical workloads with high throughput on large datasets, not for small, transient session records; its design for batch and streaming analytics makes it overkill and inefficient for per-request session operations.

Practice this question →
70
Multi-Selecteasy

A company deploys a microservice on Cloud Run and wants to minimize cold starts during traffic spikes. Which two steps should they take? (Select exactly 2.)

Select 2 answers
A.Enable CPU always allocated
B.Use Cloud CDN
C.Set max_instances to a high value
D.Set concurrency to 1
E.Set min_instances to a value greater than 0
AnswersA, E

CPU always allocated ensures instances are active and ready to serve requests immediately.

Why this answer

Enabling CPU always allocated (option A) prevents the CPU from being throttled when the container is not handling requests, which reduces cold start latency because the runtime environment remains warm and ready to process incoming traffic immediately. This is particularly effective for minimizing cold starts during traffic spikes because the container's CPU is always active, eliminating the need to spin up resources from a cold state.

Exam trap

The PCD exam often tests the distinction between scaling limits (max_instances) and proactive instance provisioning (min_instances), so the trap here is that candidates mistakenly think setting a high max_instances prevents cold starts, when in fact it only caps the maximum scale and does nothing to keep instances warm.

Practice this question →
71
Multi-Selectmedium

An organization is migrating a critical application to Google Cloud and needs to ensure high availability and disaster recovery. The application runs on Compute Engine and uses a stateful database. Which three design choices should they make? (Choose three.)

Select 3 answers
A.Use managed instance groups distributed across multiple zones.
B.Use regional persistent disks for the database.
C.Use a global load balancer to route traffic to the closest healthy region.
D.Use preemptible VMs to reduce costs for the database layer.
E.Deploy all instances in a single zone and use snapshots for backup.
AnswersA, B, C

MIGs across zones provide auto-healing and high availability.

Why this answer

Managed instance groups (MIGs) distributed across multiple zones provide automatic failover and self-healing for the Compute Engine instances. If a zone fails, the MIG automatically recreates instances in healthy zones, ensuring high availability for the application layer. This aligns with Google Cloud's best practices for regional resilience.

Exam trap

The PCD exam often tests the misconception that cost-saving measures like preemptible VMs can be applied to stateful workloads, but the trap here is that preemptible VMs are not guaranteed to run and thus cannot support a stateful database requiring persistent uptime and data integrity.

Practice this question →
72
MCQeasy

A company is designing a cloud-native application on Google Cloud that requires low-latency access to a global user base. The application serves static content and dynamic APIs. Which strategy best minimizes latency while maintaining high availability?

A.Deploy the application in a single region and use Cloud Interconnect for global access.
B.Use Cloud CDN to cache static content and deploy the API across multiple regions with global load balancing.
C.Use Cloud Armor to protect the application and rely on Google's backbone for low latency.
D.Store all content in Cloud Storage and serve directly from there.
AnswerB

Cloud CDN caches at edge locations, and multi-region deployment with global load balancing reduces latency for dynamic content.

Why this answer

It combines Cloud CDN for caching static content at edge locations worldwide, reducing latency for static assets, and deploys the dynamic API across multiple regions with global load balancing (using Google Cloud's global external HTTP(S) load balancer) to route users to the nearest healthy backend, minimizing latency for dynamic requests while ensuring high availability through regional redundancy.

Exam trap

The trap here is that candidates may think Cloud Interconnect or Cloud Armor alone can solve global latency, overlooking the need for edge caching and multi-region deployment to reduce physical distance and provide redundancy.

How to eliminate wrong answers

Option A is wrong because deploying in a single region forces all traffic to traverse potentially long distances, and Cloud Interconnect provides dedicated connectivity to Google's network but does not reduce geographic latency for global users; it also creates a single point of failure. Option C is wrong because Cloud Armor provides DDoS protection and WAF capabilities but does not reduce latency; relying solely on Google's backbone does not address the need for edge caching or multi-region distribution. Option D is wrong because serving all content directly from Cloud Storage without a CDN or global load balancing results in higher latency for users far from the storage region, and Cloud Storage alone does not provide dynamic API serving or global traffic management.

Practice this question →
73
MCQhard

A company runs a stateful application on Compute Engine with regional persistent disks. They want to achieve high availability with automatic failover in case of a zone failure. Which architecture meets these requirements?

A.Use Cloud Storage FUSE to mount a multi-regional bucket as a filesystem
B.Use standard persistent disks with scheduled snapshots to a multi-region bucket
C.Use zonal persistent disks with a managed instance group in a single zone
D.Use regional persistent disks with a managed instance group spanning two zones
AnswerD

Regional disks provide synchronous replication; instance group autohealing restarts VMs on failure.

Why this answer

Regional persistent disks synchronously replicate data across two zones, and when combined with a managed instance group (MIG) spanning those same two zones, the application can automatically fail over to the healthy zone if one zone fails. The MIG's autohealing and health-check mechanisms detect the failure and recreate instances in the surviving zone, while the regional PD remains accessible from either zone, ensuring high availability without manual intervention.

Exam trap

The trap here is that candidates confuse zonal persistent disks with regional persistent disks, or assume that snapshots (Option B) provide automatic failover, when in reality snapshots are for backup/DR and require manual recovery steps, not instant zone-failure recovery.

How to eliminate wrong answers

Option A is wrong because Cloud Storage FUSE mounts a multi-regional bucket as a filesystem, but Cloud Storage is an object store with eventual consistency (not POSIX-compliant), and it does not support synchronous replication or automatic failover for stateful applications requiring persistent disk semantics. Option B is wrong because standard persistent disks are zonal resources; scheduled snapshots to a multi-region bucket provide backup and disaster recovery, not automatic failover within minutes, and restoring from a snapshot requires manual or scripted steps, not instant failover. Option C is wrong because zonal persistent disks are tied to a single zone, and a managed instance group in a single zone cannot survive a zone failure—the instances and disks become unavailable simultaneously.

Practice this question →
74
MCQmedium

A developer runs the command shown in the exhibit. They need to ensure that the application running on instance-3 can be restored quickly if it fails. What should they do?

A.Add instance-3 to an unmanaged instance group.
B.Configure Cloud DNS with a health check pointing to instance-3.
C.Create a managed instance group using an instance template based on instance-3's boot disk.
D.Take a snapshot of instance-3's disk and use it to create a new instance manually.
AnswerC

Managed instance groups automatically recreate failed instances.

Why this answer

A managed instance group (MIG) with an instance template based on instance-3's boot disk provides automated health checking, auto-healing, and managed instance replacement. If instance-3 fails, the MIG automatically recreates it from the template, ensuring rapid restoration without manual intervention. This aligns with the goal of designing highly available and reliable applications on Google Cloud.

Exam trap

The PCD exam often tests the distinction between health-check-based traffic routing (Cloud DNS) and health-check-based instance recovery (MIG auto-healing), leading candidates to confuse DNS failover with automated instance restoration.

How to eliminate wrong answers

Option A is wrong because an unmanaged instance group does not provide auto-healing or automated instance replacement; it is a static collection of instances that must be managed manually, so it cannot restore instance-3 quickly on failure. Option B is wrong because Cloud DNS with a health check only routes traffic away from an unhealthy instance but does not recreate or restore the instance itself; it is a traffic management solution, not a recovery mechanism. Option D is wrong because taking a snapshot and manually creating a new instance is a manual, slow process that does not meet the requirement for rapid, automated restoration; it lacks the automation and health-based recovery of a managed instance group.

Practice this question →
75
Multi-Selecteasy

A company is designing a globally distributed application using Cloud Spanner. The application requires strong consistency and the ability to handle high read/write throughput. The team is concerned about inter-continental latency. Which two design choices would optimize performance while maintaining strong consistency? (Choose two.)

Select 2 answers
A.Enable leader-optimized routing to direct reads to the nearest leader region.
B.Use read-only replicas in each continent to serve reads locally.
C.Place a multi-region Spanner instance in geographic locations close to users.
D.Implement client-side caching with a short TTL for frequently accessed data.
E.Increase the number of nodes in the Spanner instance to improve throughput.
AnswersA, C

Leader-optimized routing reduces read latency while maintaining strong consistency.

Why this answer

A is correct because leader-optimized routing directs read requests to the nearest region that contains the leader replica for the requested data, reducing inter-continental latency while still reading from the leader to ensure strong consistency. C is correct because placing a multi-region Spanner instance in geographic locations close to users minimizes network round-trip time, and Spanner's synchronous replication across regions maintains strong consistency even with high read/write throughput.

Exam trap

The PCD exam often tests the misconception that read-only replicas or caching can provide strong consistency, but in Spanner, only leader replicas guarantee strong consistency, and any form of caching or stale replica reads breaks that guarantee.

Practice this question →
Page 1 of 2 · 114 questions totalNext →

Ready to test yourself?

Try a timed practice session using only Designing highly scalable, available, and reliable cloud-native applications questions.

Start 20-question session