Google Cloud Revision and Interview Questions

Cheat sheet

Hierarchy: organization → folders → projects → resources; billing account linked to projects; labels for cost. IAM: principal + role + resource, inherited downward; prefer predefined roles and groups; dedicated service accounts, no keys; Workload Identity Federation for outside workloads. Compute: Compute Engine + MIGs (VMs), Cloud Run (serverless containers, concurrency, min instances), GKE Autopilot/Standard, Cloud Run functions + Eventarc. Messaging: Pub/Sub (fan-out, at-least-once, ack deadline, dead-letter), Cloud Tasks (rate-limited calls), Cloud Scheduler, Workflows. Data: Cloud Storage classes Standard/Nearline/Coldline/Archive (30/90/365-day minimums, all millisecond access); Cloud SQL HA, AlloyDB, Spanner; Firestore, Bigtable, Memorystore; BigQuery (columnar, partition and cluster, avoid SELECT *), Dataflow, Dataproc. Network: global VPC, regional subnets, firewall priorities, Cloud NAT, global Application Load Balancer, Cloud Armor, Private Service Connect. Ops: Cloud Build, Cloud Deploy, Logging/Monitoring, Secret Manager, org policies, budgets, CUDs.

Common interview questions

Answer each in two or three sentences, including a trade-off.

1. How do projects, folders and the organization node relate to IAM inheritance?
2. Basic vs predefined vs custom roles: which would you use and why?
3. Why avoid service account keys, and what replaces them?
4. Cloud Run vs GKE vs Compute Engine for a new API?
5. Pub/Sub vs Cloud Tasks: when each?
6. How do Cloud Storage classes differ, and what is a minimum storage duration?
7. Cloud SQL vs Spanner vs Firestore for an e-commerce app?
8. How do you reduce the cost of a BigQuery query?
9. Why is a Google Cloud VPC called global?
10. How would GitHub Actions deploy to Google Cloud without secrets?

Lead with the decision

Say "I would use Cloud Run because the service is stateless HTTP and traffic is spiky" before listing features. Interviewers are checking judgement, so name the alternative you rejected.

त्वरित जाँच: A BigQuery query uses SELECT * with LIMIT 10 on a 2 TB unpartitioned table. Roughly how much data does an on-demand query scan?

• Only 10 rows
• Nothing, because results are cached by default
• Exactly 10 MB
• About the full size of the referenced columns, close to 2 TB