Skip to content
Tech Interview Prep
Technical interview guide

Vendor & Technology Selection Tradeoffs

Evaluating build-vs-buy and vendor options against a customer's actual constraints, not feature checklists.

Read
38 min
Practice MCQs
25
Interview Q&A
25
Edition
v4
Editorial status
Reviewed

Scope: AWS, Azure, Google Cloud, and NIST technology-selection and supply-chain guidance current 2026-08-31.

Curated: · Written: · Reviewed:

25 questions

MCQ-1Your team is shortlisting managed message brokers for a new order-events pipeline. The documented requirements are: the cluster must run in eu-central-1; sustained throughput of at least 40,000 messages/second; p99 publish-to-consume latency no worse than 250 ms at that rate; at least 30 days of message retention. Five vendors were benchmarked: | Vendor | Regions | Sustained throughput | p99 latency at 40k msg/s | Max retention | | --- | --- | --- | --- | --- | | Northwind Streams | eu-central-1, us-east-1 | 60,000 msg/s | 180 ms | 7 days | | Kestrel Bus | eu-central-1 | 45,000 msg/s | 210 ms | 90 days | | Ardent MQ | us-east-1, ap-southeast-1 | 80,000 msg/s | 90 ms | 365 days | | Halcyon Queue | eu-central-1, eu-west-1 | 52,000 msg/s | 240 ms | 45 days | | Meridian Relay | eu-central-1 | 41,000 msg/s | 245 ms | 60 days | Which vendors belong on the shortlist?
MCQ-2A platform team is choosing a managed message broker. Their scorecard has 9 mandatory requirements (pass/fail gates, no waivers) and 30 weighted desired points. Three finalists remain: - Broker A: 9/9 mandatory; 22/30 desired. Fails one desired criterion (no native Kafka protocol support, worth 3 points). - Broker B: 8/9 mandatory; 29/30 desired. Fails the EU data-residency gate: its only EU endpoint is a control plane that routes message payloads through us-east-1. - Broker C: 9/9 mandatory; 27/30 desired. Which broker should the team select?
MCQ-7A six-engineer platform team spends about 15 engineer-hours a week keeping its self-managed Kubernetes control plane current. It is weighing a managed offering where the provider operates the control plane and the team keeps its worker nodes in its own cloud account. Which operational saving can the team book into the migration plan?
MCQ-3A team is choosing a managed PostgreSQL service for a workload that needs 8 vCPU, 32 GB RAM, a 99.95% uptime SLA, and 24/7 support with a 1-hour response. Both vendors bill per hour, and the team runs the service 720 hours per month. Vendor A: Business tier $1.20/hr (8 vCPU/32 GB, 99.9% SLA, business-hours support); Enterprise tier $2.00/hr (8 vCPU/32 GB, 99.95% SLA, 24/7 support, 1-hour response). Vendor B: Standard tier $1.85/hr (8 vCPU/32 GB, 99.95% SLA, 24/7 support, 1-hour response); Premium tier $2.60/hr (8 vCPU/32 GB, 99.99% SLA, 24/7 support, 15-minute response). Which monthly cost comparison correctly prices each vendor at a tier that meets the workload's requirements?
MCQ-4A team is evaluating a managed time-series database. The vendor's feature matrix lists: max series 10 million; max write throughput 500,000 points/s; max retention 2 years; max disk 100 TB. The workload has 2 million active series, writes 300,000 points/s on average with 2× peaks, needs 1-year retention, and stores 8 bytes per point. Which checklist limit is the binding constraint for this workload?
MCQ-24A team runs an API serving 12 M requests a month today, with traffic projected to reach 40 M requests a month within a year. Plan A charges a $3,000 monthly platform fee plus $0.35 per 1,000 requests. Plan B charges $0.60 per 1,000 requests with no fee. Which conclusion does the arithmetic support?
MCQ-9A team is comparing two 3-year storage contracts for a steady 500,000 GB workload. Vendor A lists $0.10/GB-month and offers a 6-month 100% usage credit at signing; list price applies for the remaining 30 months. Vendor B lists $0.075/GB-month with no credits. The team's policy is to compare vendors on the monthly cost that will still apply after all promotional pricing has ended. What monthly cost should the team use for Vendor A in that comparison?
MCQ-6A payments service peaks at 500 req/s today. Two 18-month capacity plans are on the table: a reserved fleet at a flat $2,000/month, and autoscaling at $0.25 per node-hour with each node serving 100 req/s. Marketing's 18-month peak forecast spans 1.5× to 5× today's peak. Which statement about the decision is correct?
MCQ-5A payments team is deciding whether to build an internal rate-limiting service or buy a vendor's. Building takes 3 engineers for 5 months, then 0.2 FTE to maintain. Buying costs $8,000/month, needs 2 engineer-months to integrate, then 0.5 FTE to operate (data sync, upgrades, support escalations). A fully loaded engineer-month is $20,000. Over three years, which option is cheaper and by how much?
MCQ-10An engineering team is choosing between building an internal service and buying a vendor product, and is comparing three-year total cost of ownership. Building takes two engineers three months at a fully loaded $25k per engineer-month, then 0.5 engineer-months per month of maintenance for 36 months. Buying costs $5k per month of subscription for 36 months, one engineer for two months of integration at $25k per engineer-month, $10k of one-time training, 0.1 engineer-months per month of incident support for 36 months, and a $20k exit fee at the end of the term. Which conclusion do the three-year totals support?
MCQ-16A vendor's release notes claim version 8 is 2.1x faster than version 7: 84k vs 40k ops/s on a 32-core instance. Both runs were performed by the vendor's performance team. Version 8's run enabled a new `io_uring` backend; version 7's run used the default `epoll` backend. The report does not mention the backend flag. Your team reruns both versions on the same instance class with default backends and gets 44k (v8) and 40k (v7). What does the published 2.1x figure actually measure?
MCQ-19A team is deciding whether to migrate a service to a new JSON codec. They run 40 paired trials, alternating codecs on the same host. Old codec: mean 84 ms. New codec: mean 79 ms. The paired difference has a mean of 5 ms and a 95% confidence interval of [1 ms, 9 ms]. The team's adoption rule: migrate only if the lower bound of the 95% CI exceeds 3 ms. What should they conclude?
MCQ-15A team must pick one of two managed queue services in three weeks. Both vendors' published benchmarks clear the 20k msg/s target on uniform load, but the team's real ingest is heavily skewed to a few hot keys, and the SLO is p99 under 50 ms. What should the proof of concept test?
MCQ-12A security team is onboarding a vendor that holds a current SOC 2 Type II report covering the vendor's US production environment. The team's workload will run in the vendor's EU subprocessor region and handle regulated data. What is the correct use of the report?
MCQ-11A vendor's SOC 2 Type II report names one production environment, in Ireland. Its subprocessor list names two further hosting providers, in the US and Singapore, that store customer data. The contract processes EU customer PII and runs 24 months. What does the review need before signing?
MCQ-22A payments platform is replacing its in-house feature-flag service with a small SaaS vendor and wants to decide within two weeks. Flags are evaluated on every checkout request (9,000 rps at peak) and one flag gates a fraud-scoring rule. The contract is $18k/year, month-to-month, and the vendor receives internal experiment keys and pseudonymous user IDs but no card data or PII. Which review set is proportionate for the adoption decision?
MCQ-21A bank is negotiating the exit clause of a three-year SaaS contract. The vendor's draft says customer data is deleted 90 days after termination, migration assistance is billed at then-current rates for 30 days, and the export format may change at will. The vendor's backups run on a 35-day cycle. The bank's regulator requires confirmed deletion of customer personal data within 30 days of contract end. Which change to the draft does that requirement force?
MCQ-13A payments team runs a 3,000 GB event archive on a managed queue. The vendor SDK is called from 40 services, identity is tied to the vendor's IAM, and export costs $0.09/GB egress. A migration spike estimates 6 engineer-months to replace the SDK and IAM, plus $4k/month dual-run for 3 months. The managed service saves $18k/month versus self-managed. The team must decide whether to renew. What makes the lock-in analysis actionable for that renewal decision?
MCQ-14A team's checkout service runs on a single cloud provider. A forced migration, a price shock or a provider exit, would cost an estimated $6.0M in engineering and lost revenue, and finance puts that risk at 35% over the next 18 months. A portability layer would cost $400K to build and $150K/year to maintain, and would cut a forced migration to $1.5M. Over that 18-month horizon, what does the comparison support?
MCQ-18A platform team is choosing among three observability vendors. The RFP fixes the weights before the bake-off (Performance 50%, Cost 30%, Support 20%) and lists a SOC 2 Type II report as a mandatory gate. Scores on a 0–10 scale: Vendor A 6 / 9 / 7, Vendor B 9 / 8 / 9, Vendor C 7 / 7 / 7. Vendor B has no SOC 2 Type II report; A and C both have one. Which vendor does the agreed model select?
MCQ-17Fourteen months ago an ADR selected queue A over B and C, recording the weighted scores (8.4 / 7.9 / 6.1), the approvers, and the rejected alternatives. A's price has since risen from $0.40 to $1.20 per million messages and its batch-publish API is deprecated. The team must decide whether to stay or migrate, and has no budget to re-benchmark. Which addition to the original record would have made the stay-or-migrate decision possible?
MCQ-8A five-engineer team has six weeks to ship a service that will page them. Two client libraries are candidates: one the team has run in production for three years, and one in Rust that benchmarks 30% lower p99 at their projected load. Nobody on the team has written Rust, and the Rust client's API shape would leak into the service's own interfaces. Which argument for choosing the Go library is sound?
MCQ-20A payments team runs its Postgres cluster on self-managed VMs. The cluster is on Postgres 11, which reached end-of-life on November 9, 2023 and no longer receives security fixes. The current cluster costs $7,000/month all-in (instances $4,800, storage $900, 0.1625 FTE ops at $8,000/FTE-month = $1,300). Two migration candidates are on the table, both running Postgres 16: lift-and-shift to managed VMs at $8,200/month plus $20,000 one-time, and re-platform to a managed service at $9,500/month plus $60,000 one-time. The team's business case compares only the two migration candidates and recommends lift-and-shift as the cheaper of the two. After adding the status quo as a candidate, what is the incremental monthly cost of the recommended lift-and-shift over the status quo, and what does it imply for the business case?
MCQ-23A platform team selected a managed message broker 18 months ago. The business case assumed 5k msg/s of included throughput at $0.40 per million messages; the selection scored three options and rejected a self-hosted broker and a second managed vendor. Measured peak is 6k msg/s sustained and the blended cost is $1.10 per million. The vendor's roadmap lists exactly-once delivery for next quarter, two quarters after the original date. Renewal is in 30 days. What should the team do?
MCQ-25A team must choose a message broker for a new event pipeline. Requirements: sustain 80k msg/s peak, p99 publish latency under 30 ms, 7-day retention, and a vendor-supported release for at least 18 months. They benchmarked two candidates on identical 6-broker hardware. Broker A: 120k msg/s at p99 18 ms; the team has 4 years of production experience; current release 3.4 reaches end-of-life in 8 months, and the vendor confirmed no upgrade path for the team's custom auth plugin. Broker B: 95k msg/s at p99 22 ms; the team has no production experience; current release is supported for 24 months; tiered storage to object storage cuts 7-day retention cost by 40%. Which final recommendation is most credible?