Cloud architecture

1.You are asked to design the cloud setup for a clinic booking system that now runs on one server. Before drawing anything, what do you ask first?

2.A product lead says the new ticket sale platform must never go down. What is the strongest senior response?

3.A team asks for a five minute recovery time objective (RTO) for an internal reporting tool used once a week. What do you do?

4.You are planning a move to the cloud for a sensor dashboard used by utilities in several countries. Which early question most often reveals hard constraints?

5.An interviewer says a recipe app has 50,000 daily users. To size peak compute, which extra figure matters most?

6.A team spreads a news feed across three availability zones in one region. What does this mainly protect against?

7.Which failure does a multi-region design protect against that a well-built multi-zone design does not?

8.A code review tool runs in two regions for resilience. One pipeline deploys to both at the same moment, and a bad release breaks both. What would have contained it?

9.You consider synchronous database replication between two regions with a 70 millisecond round trip. What is the main effect?

10.A service needs 12 instances at peak. It runs in three zones and must survive losing one zone at peak with no slowdown. How many instances should run?

11.Why do many teams prefer three availability zones over two for a stateless web tier?

12.A self-run consensus cluster has three nodes: two in zone A and one in zone B. Zone A fails. What happens?

13.A ride dispatch map's zone failover plan launches new instances through the provider's API only after a zone fails. What is the main concern?

14.A grant application portal keeps a database replica running in a second region but no app servers. During a disaster it launches app servers from prebuilt images. Which pattern is this?

15.An internal wiki can accept a 24 hour recovery point objective (RPO) and a 48 hour recovery time objective (RTO). What is the cheapest design that meets both?

16.A ledger for a small lender needs a 15 minute recovery time and at most one minute of data loss in a regional outage. Which design fits best?

17.Cross-region replication lag averages 2 seconds but reaches 90 seconds during nightly batch loads. The recovery point objective (RPO) is 30 seconds. What do you conclude?

18.Why does active-active across two regions often cost about twice a single-region setup, or more?

19.A shared shopping list app goes active-active, accepting writes in both regions. What must the design address that a single-writer design does not?

20.For a news site's regional failover, which approach to the failover decision is usually safest?

21.A team has failed over to its second region. The primary region is healthy again. What is most often underestimated?

22.How do you show that a warm standby meets a 30 minute recovery time objective?

23.A wedding photo gallery has a warm standby in region B, but users sign in through an identity service that runs only in region A. What is the problem?

24.Downtime costs a ticket sale company $20,000 per hour. A warm standby costs $15,000 a month more than backup and restore and would cut expected outage time by 6 hours a year. Is it justified on cost alone?

25.Backups are copied to a second region, but they are encrypted with a key that exists only in the primary region. What is the risk?

26.A chat app asks for a zero recovery point objective across regions. What is the honest answer?

27.You are designing a landing zone for 40 product teams. Which account structure scales best?

28.In a multi-account landing zone, where should the shared transit hub and private DNS live?

29.Why should an organization's management account hold almost no workloads?

30.Product teams wait three weeks for each new account because a platform engineer sets them up by hand. What is the best fix?

31.Which control belongs at the organization level instead of inside each workload account?

32.Developers want somewhere to try new cloud services freely. What design fits?

33.A security team writes a new organization-wide deny policy that might break running workloads. How should it roll out?

34.Why does splitting workloads across many accounts help with service quotas?

35.Twenty-five virtual networks must all reach each other. How many peering links would a full mesh need?

36.A plan gives each of 200 future virtual networks a /16 block from 10.0.0.0/8. What is the main concern?

37.Security wants all traffic between spoke networks inspected. In a hub and spoke design, where does inspection go?

38.A hub's inspection firewall is a single appliance in one zone, and all spoke traffic passes through it. What do you change?

39.A factory sends a steady 2 Gbps of sensor data to the cloud over a site-to-site virtual private network (VPN) on the internet. Latency and throughput swing widely. What helps most?

40.A data center reaches the cloud over one dedicated private circuit. What is the main design gap?

41.Workloads in 60 accounts need internet egress, and security wants outbound traffic limited to approved domains. What fits best?

42.What is a real downside of sending all accounts' internet egress through one central hub?

43.A service in one account must expose an API to 40 consumer accounts, several of which use overlapping IP ranges. What fits best?

44.On-premises servers must resolve private cloud names, and cloud workloads must resolve on-premises names. What design fits?

45.How do you stop a compromised workload from copying data to an attacker's storage bucket at the same cloud provider?

46.A team wants to run Kafka on virtual machines to save 30 percent over the managed offering. What should weigh most in the decision?

47.When does running a database yourself instead of using the managed service usually make sense?

48.A five-person team building a booking app proposes running its own Kubernetes control plane on virtual machines. What do you suggest?

49.A team avoids every managed service so it can move clouds later. What is the more balanced approach?

50.A managed database tops out at 64 TB of storage. The data grows 1.5 TB a month and holds 20 TB today. What do you do?

51.A webhook receiver for a payroll export tool is idle most of the day, then gets bursts of 500 requests per second for a few minutes. What runs it well?

52.A team picks a managed database but chooses the single-zone option to save money, believing managed means highly available. What is wrong?

53.A photo backup service receives 3 million uploads a day, averaging 4 MB each. About how much new storage does it need?

54.A news feed has 8 million daily active users who each make 40 requests a day. Peak is three times the average. What are the rates?

55.Each instance serves 250 requests per second at its target CPU. Peak is 9,000 per second. The tier runs in three zones and must survive losing one zone at peak. How many instances?

56.A video app serves 2 MB clips 5,000 times per second. Roughly what outbound bandwidth is that?

57.A session cache must hold 50 million sessions of 2 KB each. About how much memory do the raw values need?

58.A dispatch API spends most of each request waiting on downstream calls. Under load, CPU stays at 20 percent while latency climbs. What should drive autoscaling?

59.A queue holds 120,000 messages. Each worker processes 20 messages per second. Ignoring new arrivals, how many workers clear it in 10 minutes?

60.When a video transcoding fleet scales in, workers are killed mid-job and the work is lost. What is the fix?

61.An app tier can scale to 200 instances, each with a pool of 20 database connections. The database allows 2,000 connections. What is the problem?

62.A service handles 2,000 requests per second with an average latency of 150 milliseconds. On average, how many requests are in flight?

63.Why do many teams set an autoscaling CPU target near 60 percent instead of 90 percent?

64.A clinic booking system for patients in the European Union must keep personal data there. Which design fits?

65.A team pins its database to an in-country region for residency. Which data flow is most often overlooked?

66.A residency-bound tax filing app sits behind a global content delivery network. What needs care?

67.A software company sells to customers in three jurisdictions, each requiring local storage. Which architecture fits?

68.In a design with separate deployments per jurisdiction, what should the global routing layer store?

69.A team argues that encrypting data with a key held in the right country satisfies a law requiring data to be stored there. What is the better view?

70.Leadership wants one global dashboard for a product with data pinned in several jurisdictions. What design fits?

71.A sensor platform serves 500 TB a month to the internet at $0.05 per GB. About what is the monthly egress bill?

72.A service makes 10 calls across zones per request, each 50 KB, at 2,000 requests per second. Roughly how much data crosses zones each month?

73.Which cost is most often left out of a first cloud migration estimate?

74.A fleet runs a steady 40 instances all day and rises to 100 for two hours each evening. How should the team buy capacity?

75.Which workload is the best fit for spot or preemptible capacity?

76.An archive holds 200 TB. Ninety percent is never read after 30 days. Hot storage costs $0.023 per GB-month and archive costs $0.004. Roughly what does tiering save?

77.An app logs 2 KB per request at 5,000 requests per second. About how much log data is that per month?

78.A media site serves 300 TB a month from its origin. A content delivery network with a 90 percent cache hit ratio is placed in front. What changes?

79.In a design review for a new service, which cost question tells you the most?

80.A write-heavy database replicates 20 MB per second to two other regions. Transfer between regions costs $0.02 per GB. About what is the monthly transfer?

81.A company has 500 engineers and 150 cloud accounts. How should human access work?

82.Why does federating cloud access from a central directory help most when someone leaves the company?

83.The company's identity provider is down, and engineers cannot sign in to the cloud during an incident. What should the design include?

84.A platform team wants product teams to create roles for their own services without being able to grant themselves more power. What fits?

85.An organization has 300 projects, and one role per project per job type has grown into thousands of roles. What scales better?

86.A central deployment pipeline must deploy into 80 workload accounts. Which design limits risk?

87.Reviewing 2,000 permission policies by hand is not possible. What keeps least privilege workable at scale?

88.Four hundred services each hold a long-lived database password in config files. What is the better pattern at scale?

89.One infrastructure as code state holds the network, all databases, and 60 services. Plans take 25 minutes, and any change can touch anything. What do you change?

90.How do you catch drift across hundreds of infrastructure as code stacks?

91.During an incident, an engineer raised a load balancer timeout in the console. What should happen afterward?

92.Service stacks hardcode subnet IDs copied from the network stack. What is the better link between stacks?

93.Staging and production infrastructure drift apart because engineers edit each environment's code separately. What fixes this?

94.Renaming a resource in infrastructure code would make the tool destroy and recreate a production queue. What is the right way to refactor?

95.A team wants to block public buckets and untagged resources before they ever reach the cloud. Where does the check fit best?

96.A deploy pipeline breaks one morning because the infrastructure tool's provider plugin upgraded itself. What prevents this?

97.Engineers apply infrastructure changes from their own laptops with personal admin roles. What is the better model?

98.A 3 TB Postgres database in a data center must move to the cloud with under 10 minutes of downtime. What plan fits?

99.A data center holds 80 TB to move over a 1 Gbps link that can run at 80 percent. About how long will the copy take?

100.A company will migrate 120 applications. Which one should go first?

101.Before planning migration waves, what is the most reliable way to learn which applications depend on each other?

102.An app moves to the cloud while its database stays in the data center with a 10 ms round trip. A page makes 200 queries one after another. What happens?

103.A data center lease ends in six months, and 80 applications must leave. What strategy fits?

104.A large monolith for a museum ticketing site will move to the cloud gradually. Which approach lowers risk most?

105.What makes a database cutover to the cloud reversible if problems appear after go-live?

106.A team plans to give each migrated server the same CPU and memory it had in the data center. What should it do instead?

107.A service has a 99.9 percent monthly availability service level objective (SLO). About how much downtime does a 30 day month allow?

108.A request must pass through three services in series, each 99.9 percent available. What is the best estimate for the whole path?

109.Which signal should trigger an alert to consider failing a booking system over to another region?

110.Two regions each offer 99.5 percent availability, and either can serve all traffic alone. What is the theoretical availability of the pair?