Task 1.2: Designing a cloud solution infrastructure that meets technical requirements.

• ARegional GKE clusters and Cloud SQL for MySQL instances configured for high availability

  Correct

  A regional cluster replicates the control plane and nodes across zones, and a Cloud SQL HA instance fails over to a standby in another zone, both as managed features.

• BZonal GKE clusters with node auto-repair and a Cloud SQL read replica in the same zone

  Incorrect

  Everything sits in one zone, and a read replica does not provide failover, so a zone outage takes the system down.

• CRegional GKE clusters and MySQL on Compute Engine VMs with a custom failover script

  Incorrect

  This survives a zone failure, but running and failing over MySQL yourself adds the administration the company wants to reduce.

• DOne zonal GKE cluster per application, each restored from nightly backups after a failure

  Incorrect

  Restoring a cluster and its data after a zone failure takes far longer than a 99.9% target allows.

Regional GKE clusters and Cloud SQL high availability keep the system running through a zone failure as managed features, so the design meets the availability requirement without adding operational work.

• ACloud Architecture Center reference architectures

  Incorrect

  Reference architectures are fixed documents; they do not draft a design for your application or converse about changes.

• BMigration Center with its asset discovery client

  Incorrect

  Migration Center inventories and assesses existing infrastructure; it does not design a new application.

• CGemini Cloud Assist in Application Design Center

  Correct

  Gemini Cloud Assist lets you design, configure and provision architectures through natural language in App Design Center and generate Terraform for them.

• DInfrastructure Manager with a Terraform blueprint

  Incorrect

  Infrastructure Manager deploys Terraform you already have; it does not produce a design from a description.

The guide lists Gemini Cloud Assist under technical design. Its design features work inside Application Design Center: describe the requirements, review the generated architecture, iterate in conversation and export the infrastructure as code.

• ABackup and DR Service backup plans applied across the projects, with its reports

  Correct

  Backup and DR manages backup policy from one interface, can back up instances that sit in other projects, and its prebuilt reports cover backup jobs and compliance.

• BSnapshot schedules that each team creates and attaches to the disks in its own project

  Incorrect

  Schedules work per project and per disk, so the policy is repeated 40 times and there is still no single view of what is protected.

• CMachine images of each VM, created nightly by a Cloud Scheduler job in each project

  Incorrect

  Every project keeps its own jobs and images, with no shared policy and no compliance report across the projects.

• DA backup=daily label that each team adds to the VMs it wants to protect

  Incorrect

  A label is metadata for organizing and filtering resources; on its own it creates no backup.

The requirement is central backup governance across many projects. Backup and DR Service applies backup plans to resources across the organization from one console and reports on backup jobs and compliance, which per-project snapshots, images or labels cannot provide.

• AA regional external Application Load Balancer in the region that has the most customers

  Incorrect

  A single region leaves distant customers with long round trips and gives no regional failover.

• BCloud DNS records that send each country to a separate regional load balancer address

  Incorrect

  This creates several entry points and relies on DNS caching, which the single-entry-point requirement rules out.

• CA global external Application Load Balancer with backends in regions near the customers

  Correct

  The global load balancer serves one anycast IP from Google's edge and sends each request to the closest healthy backend region.

• DStandard Tier networking on every backend so traffic travels over the public internet

  Incorrect

  Standard Tier hands traffic to the internet near the region, which usually adds latency for far-away users.

For global users with one entry point, the global external Application Load Balancer on Premium Tier is the latency and availability design: users enter Google's network close to them and are routed to the nearest backends.

• AA multi-region location for the bucket that stores the objects

  Incorrect

  A multi-region bucket improves availability, but every request is still served from the bucket rather than an edge cache.

• BMemorystore for Redis instances in each region to hold the files

  Incorrect

  Memorystore is an in-memory store for applications, not a cache that serves web content to browsers.

• CA Cloud Armor security policy with rate limits on the load balancer

  Incorrect

  Cloud Armor protects the site from attacks; it does not speed up content delivery.

• DCloud CDN enabled on the backend bucket behind the load balancer

  Correct

  Cloud CDN caches the content at Google's edge close to readers, which cuts latency and the traffic that must come from the bucket.

Performance and latency for static, widely shared content is a caching problem. Cloud CDN uses Google's edge network with the external Application Load Balancer to serve content closer to users.

• AA fixed number of nodes sized for the highest evening peak

  Incorrect

  This is today's setup: it covers the peak and leaves capacity idle the rest of the day.

• BA Cloud Scheduler job that resizes the instance twice a day

  Incorrect

  A schedule follows the clock, not the load, so unplanned spikes still find the instance at its smaller size.

• CA multi-region instance configuration for the database

  Incorrect

  More regions raise availability and cost; they do not make the compute capacity grow and shrink with load.

• DThe Spanner managed autoscaler with limits and a CPU target

  Correct

  The managed autoscaler adds or removes compute capacity as the workload changes, within minimum and maximum limits, based on CPU and storage targets.

Elasticity is a technical requirement. Spanner's managed autoscaler scales an instance's nodes or processing units up and down with its workload, so capacity tracks demand instead of the peak.

• ARun point-in-time recovery to create a new instance as of 14:04

  Correct

  Point-in-time recovery always creates a new instance at the chosen timestamp, so the data before the DELETE comes back without touching production.

• BRestore last night's automated backup over the production instance

  Incorrect

  This loses every change since the backup and overwrites the instance the team wants to keep.

• CPromote the cross-region read replica to a standalone instance

  Incorrect

  The replica has already applied the DELETE, so promoting it does not bring the rows back.

• DFail over the high-availability instance to its standby in another zone

  Incorrect

  The standby holds the same data as the primary, including the DELETE.

Backup and recovery design should cover logical errors, not just infrastructure failures. Replication and HA copy mistakes instantly; point-in-time recovery from backups and binary logs undoes them.