Practice Prompts

Build GET /users?page=1&limit=20 that is JWT-protected, validates query params, and returns a paginated result. Show the controller, the query DTO, and how the guard/pipe are wired.

Implement a @Roles('admin') decorator and a RolesGuard that reads required roles from route metadata and compares them to request.user.roles. Make method-level roles override class-level.

Write retry(fn, { retries, baseMs }) that retries a failing async function with exponential backoff and jitter, only retrying transient errors, and gives up after N attempts.

Implement an LRU cache with O(1) get and set and a max capacity that evicts the least-recently-used entry.

Process 10,000 items by running an async worker(item) with at most N in flight at once (a bounded Promise.all). Why not just Promise.all(items.map(worker))?

Write an interceptor that wraps every successful response in { data, timestamp } without each handler knowing about it.

Serve a multi-GB file as a gzipped download without loading it into memory, and make sure errors and resources are handled.

Implement a distributed token-bucket limiter (refill rate R, capacity C) that's correct across multiple app instances. Why must it be atomic?

A Post.author field resolver runs one query per post over a list of 50 posts (1+50 queries). Fix it with DataLoader and explain the two classic bugs.

Implement graceful shutdown so a SIGTERM (k8s rollout) drains in-flight requests and closes resources before exit.

Design the backend for a B2B SaaS where each tenant's data must be isolated. Cover the isolation model, tenant resolution, and how you prevent cross-tenant leakage.

An order flows through payment, inventory, and shipping — separate services. Design it for reliability: no double charges, no lost orders, consistent state.

You have a gateway and several backend services. Where do you authenticate and authorize, and how do downstream services trust the caller?

A product-detail endpoint gets 50k RPS, 99% reads, data changes a few times a day. Design the caching so it's fast, fresh enough, and survives a cache miss storm.

Clients retry on timeout, so a “charge card” request can arrive twice. Design the API + storage so a customer is never double-charged.

You're on call for a Nest API and latency just spiked. Design the observability you'd want in place to diagnose it in minutes, not hours.

Return the length of the longest substring without repeating characters. Backend tie-in: the sliding-window pattern underlies rate limiting and log-window analytics.

Given an array, return the K most frequent elements. Backend tie-in: leaderboards, top-N metrics, merging counts from shards.

Merge all overlapping intervals. Backend tie-in: scheduling, calendar/booking conflicts, compacting time ranges.

Order tasks so every dependency comes first; detect cycles. Backend tie-in: build/job DAGs, module init order, dependency resolution, deadlock detection.

Return indices of two numbers that add to a target. Backend tie-in: the hashmap-for-O(1)-lookup instinct behind dedup, joins, and caching.

Find the shortest path length from top-left to bottom-right in a 0/1 grid (0 = open). Backend tie-in: graph traversal for recommendations, routing, and reachability.

Implement debounce(fn, ms) (fire after quiet period) and throttle(fn, ms) (at most once per interval). Backend tie-in: rate-limiting calls, coalescing bursts, batching writes.

Validate that brackets ()[]{} are balanced and correctly nested. Backend tie-in: parsing, tokenizing, expression evaluation.

Merge K sorted sequences into one sorted output. Backend tie-in: merging sorted results from K database shards or K ordered partitions.

Implement a counter that reports how many requests happened in the last W ms. Backend tie-in: the data structure inside a sliding-window rate limiter.

p99 latency on one endpoint jumped from 80ms to 2s. You have logs, metrics, and traces. Walk through how you find and fix the cause.

RSS climbs steadily over hours until k8s OOM-kills the pod. How do you find the leak?

A list endpoint that returns 100 orders fires ~400 queries. Diagnose and fix it.

An endpoint generates a PDF / resizes an image synchronously and tanks throughput under load. Make it non-blocking.

The service takes 8s to become ready, hurting deploys and serverless cold starts. How do you cut it?