Designing and planning a cloud solution architecture practice questions

Answer: A) Cloud Run functions

• ACloud Run functions run event-triggered code, such as on Pub/Sub messages, without managing infrastructure.
• BThat requires managing a VM and polling code.
• CThat adds a lot of infrastructure for a small function.
• DCloud Router exchanges routes; it doesn't run code.

Answer: A) Cloud Spanner

• ASpanner provides synchronously replicated, externally consistent relational storage that scales writes horizontally across nodes and regions, with multi-region configurations offering up to 99.999% availability.
• BCloud SQL scales reads with replicas but writes are bound to a single primary instance and it is regional — it can't meet global horizontal write scale.
• CFirestore is a document database without full relational SQL semantics and multi-statement cross-entity transactions at this scale profile.
• DBigtable is NoSQL wide-column with no SQL joins or relational transactions.

Answer: A) Google Kubernetes Engine

• AGKE allows DaemonSets and node-level customization that Cloud Run doesn't expose.
• BCloud Run abstracts away nodes and doesn't run DaemonSets.
• CFunctions are for event-driven code, not node-level agents.
• DApp Engine standard runs in a managed sandbox.

Answer: A) RTO is the maximum acceptable downtime before service is restored; RPO is the maximum acceptable amount of data loss measured in time

• ARTO answers "how long can we be down?" and RPO answers "how much recent data can we afford to lose?" — both drive the backup and failover architecture chosen.
• BThey measure distinct concerns — downtime duration (RTO) versus acceptable data loss (RPO) — and can require different architectural choices.
• CThis reverses the definitions — RTO is about downtime, RPO is about data loss.
• DRTO and RPO are general business-continuity concepts that apply equally to cloud architectures when designing backup and failover strategies.

Answer: A) BigQuery

• ABigQuery is a serverless, columnar data warehouse for large-scale SQL analytics; the on-demand model charges by bytes scanned and requires no cluster sizing.
• BBigtable is a low-latency wide-column NoSQL store for operational workloads, not ad-hoc analytical SQL over historical data.
• CCloud SQL is a managed OLTP relational database; it isn't built for petabyte-scale analytical scans.
• DFirestore is a document database for application data, not an analytics warehouse.

Answer: C) 12 instances, so the two surviving zones can carry the full peak

Capacity planning starts with the expected peak and per-instance capacity measured by load testing, then adds headroom for the failure domain you must survive. 8 instances are needed at peak, and with one of three zones lost, the remaining two zones must carry all 8, so the minimum is 12.

• AIncorrect. 2,000 / 250 = 8 instances covers the peak only while all three zones are healthy. If one zone fails, about a third of capacity is lost and the service can no longer handle the peak.
• BIncorrect. N+1 protects against losing a single instance, not a whole zone. With 3 instances per zone, a zone outage leaves 6 instances (1,500 RPS), which is below the peak.
• CCorrect. The peak needs 8 instances. If one of three zones is lost, the remaining two zones must hold those 8, which means 4 per zone, or 12 in total. This is planning for peak plus failure headroom, based on capacity measured by load testing.
• DIncorrect. Doubling would survive a zone outage, but the question asks for the minimum. 12 instances already meet the requirement, so 16 over-provisions and adds cost.

Answer: A) Batch, with each analysis defined as a job containing container runnables

Batch is Google Cloud's fully managed service for queuing, scheduling and running large numbers of tasks on autoscaled Compute Engine VMs. GKE, Dataflow and a plain MIG would each need a self-managed cluster, a different programming model, or a custom scheduler.

• ACorrect. Batch is a fully managed service that queues and schedules jobs, provisions and autoscales Compute Engine VMs (through a regional MIG), runs container or script runnables, and handles common errors. No third-party scheduler or cluster is needed.
• BIncorrect. Kubernetes Jobs can run containers, but with GKE Standard the team has to create and operate the cluster and its node pools. The requirement was to avoid running a cluster.
• CIncorrect. Dataflow runs Apache Beam data-processing pipelines. It is not a general-purpose service for queuing and running arbitrary containerized tasks on VMs.
• DIncorrect. A MIG scales a group of identical VMs, but it has no job queue, task scheduling or task-level retries. The team would have to build those on its own.

Answer: A) Migration Center, using its discovery client to collect data and then generating cost estimates

Migration Center is Google Cloud's tool for discovering assets, assessing utilization and estimating costs and TCO before a migration. Migrate to VMs and Database Migration Service carry out migrations, and the Pricing Calculator only prices configurations you enter by hand.

• ACorrect. Migration Center's discovery client scans the on-premises environment and collects performance data over time. From that data, Migration Center produces cost and TCO estimates based on the observed resources.
• BIncorrect. Migrate to Virtual Machines carries out the actual VM migration (replication and cutover). It is not the tool for discovery, utilization analysis or pre-migration cost estimates.
• CIncorrect. The Pricing Calculator only prices configurations you enter by hand. It does not discover servers or collect utilization data to right-size them.
• DIncorrect. Database Migration Service migrates databases to managed targets such as Cloud SQL. It does not inventory servers or estimate infrastructure costs.

Answer: A) Queue orders for later processing

• ABuffering orders lets the frontend keep working while downstream services recover.
• BTight synchronous coupling makes the frontend fail whenever inventory fails.
• CRemoving retries makes transient failures more likely to lose orders.
• DBrowser memory isn't durable and is lost when the page closes.

Answer: A) Dedicated Interconnect, with redundant 10-Gbps connections in two edge availability domains of the colocation metro

For private, SLA-backed bandwidth of several Gbps where the company can meet Google in a colocation facility, use Dedicated Interconnect with a redundant topology. HA VPN can be scaled with more tunnels, but it still runs over the public internet.

• ADedicated Interconnect is a direct physical link from the company's network to Google's network in a colocation facility, using 10-, 100- or 400-Gbps circuits. One 10-Gbps circuit covers 8 Gbps, and a redundant topology across edge availability domains qualifies for the Interconnect SLA.
• BAdding tunnels (about 3 Gbps each) could reach 8 Gbps, and HA VPN has a 99.99% SLA. But the IPsec tunnels run over the public internet, which breaks the private-path requirement.
• CDirect Peering gives access to Google's public services and APIs, not private connectivity to VPC networks, and it has no SLA.
• DCross-Cloud Interconnect connects Google Cloud to another cloud provider's network. It isn't used to connect an on-premises data center.