Kafka Interview Questions 2026

Design Swiggy's real-time order tracking system. A customer should see their delivery agent's location update every 5 seconds on a map. Cover: location ingestion from agents, WebSocket vs SSE for pushing updates to clients, data store choices, and how you scale to 100k concurrent orders during peak hours.

Design Swiggy's delivery partner assignment algorithm. When an order is placed, how do you assign the optimal nearby delivery partner within 2 minutes considering distance, current load, and restaurant prep time?

Design Swiggy's notification system. When an order status changes (Confirmed, Picked Up, Arriving), send push notification, SMS, and in-app notification reliably with at-least-once delivery. Cover fan-out and handling inactive devices.

Design Facebook's News Feed ranking system. How do you score posts from thousands of friends and pages? Cover: candidate generation, feature extraction, ML ranking model, how you handle near-real-time freshness, and edge vs server-side ranking for mobile clients.

Design Facebook Live - a system to stream live video from one user to potentially millions of concurrent viewers with low latency. Cover: ingest pipeline, transcoding, CDN edge caching, adaptive bitrate, and how you handle a celebrity going live.

Design WhatsApp's end-to-end encrypted messaging system. How do you exchange keys, store messages server-side without reading them, and handle multi-device sync? Cover the Signal Protocol basics.

Design Instagram Stories. A story disappears after 24 hours and is shown in a ring around profile pictures. Cover: upload pipeline, storage with TTL, efficient delivery to followers, and view count tracking at scale.

You have a Kafka topic with 6 partitions and a consumer group with 10 consumer instances. What happens to the 4 extra consumers, and why is over-provisioning consumers like this a mistake?

Design Zomato's surge pricing engine. When demand spikes (rain, IPL final), delivery fees increase automatically. Cover: demand signals, pricing formula, fairness constraints, A/B testing pricing strategies, and how to avoid customer backlash.

Design Zomato's delivery time prediction system. Before placing an order, show "Delivers in 35 minutes". The estimate must account for restaurant preparation time, distance, traffic, and number of parallel orders the partner is handling.

Design Zomato Pro (subscription). Members get free delivery, discounts, and priority support. Cover: subscription lifecycle, prorated upgrades/downgrades, cohort-based analytics to measure retention, and how you handle payment failures.

Explain Zomato's restaurant onboarding data pipeline. A new restaurant submits menus, photos, and operating hours. How do you validate, normalise, and make this data searchable within minutes of submission?

Explain the difference between at-most-once, at-least-once, and exactly-once delivery semantics in Kafka. Which one requires the consumer to handle duplicate processing?

Design Razorpay's payment retry system. When a payment fails due to a transient error (bank timeout, network blip), how do you retry safely without double-charging the customer? Cover: idempotency keys, exponential backoff with jitter, dead-letter queues, and reconciliation with bank statements.

Design Razorpay's payout system (RazorpayX). Businesses send bulk payouts to thousands of vendor accounts simultaneously. Cover: batching, bank API rate limits, real-time status tracking, and failure handling.

How do you implement a distributed ledger for financial transactions? Explain double-entry bookkeeping in code, why you never update a balance directly, and how you use event sourcing to reconstruct the current balance from a transaction log.

Design Razorpay's dashboard for merchants. A merchant wants to see real-time revenue, transaction success rates, and refund trends. How do you aggregate millions of transactions into live charts with under 5-second latency?

Kafka guarantees message order within a partition but not across partitions. For a payments system, why does this matter, and how would you key messages to preserve per-account ordering?

Design PhonePe's UPI transaction system to handle 1 billion transactions per day. Cover: request routing to the correct bank, NPCI integration, handling duplicate transactions, consistency vs availability trade-offs, and how you build a reliable audit trail for regulatory compliance.

Design PhonePe's switch - the internal routing layer that forwards UPI requests to the correct bank. It must be highly available (99.999%), handle 50k TPS, and complete each request within 30 seconds (NPCI timeout). Cover: circuit breakers, retries, and bank-specific timeouts.

Design PhonePe's merchant settlement system. Thousands of merchants receive daily settlements of their collected payments minus fees. How do you compute net payables, batch into bank transfers, and provide a reconciliation report?

Design LinkedIn's "People You May Know" feature. How do you compute second and third-degree connections for 900 million users efficiently? Discuss graph storage (adjacency list vs matrix), batch vs real-time computation, and how to refresh recommendations when someone adds a new connection.

Design LinkedIn's job recommendation engine. Given a user's profile, skills, and activity, rank the most relevant open positions. Cover candidate retrieval, feature engineering (skills overlap, company affinity, recency), and how you A/B test ranking changes at scale.

Design LinkedIn's feed. Each user sees posts from their connections, pages they follow, and sponsored content. Posts must be ranked by relevance. Cover: fan-out-on-write vs fan-out-on-read, feed generation for users with 30k connections, and freshness.

LinkedIn uses Kafka extensively. Explain how you would design the Kafka topic structure for LinkedIn's activity events (profile view, connection request, message sent). Cover: topic naming, partitioning strategy, consumer group design, and exactly-once semantics.

Design LinkedIn's InMail spam filtering system. Premium users can message strangers. How do you detect and block spam campaigns while allowing legitimate cold outreach? Cover: sender reputation, content analysis, and feedback loops.

Design Uber's driver-rider matching system. How do you match the nearest available driver to a rider within 200ms at global scale? Cover: geospatial indexing (H3/S2), the dispatch algorithm, how you handle supply-demand imbalances, and what happens when a matched driver rejects the trip.

Design Uber's dynamic pricing (surge) system. When supply is low and demand is high in a geofenced zone, prices increase automatically. Explain the data pipeline, the pricing model, how you prevent oscillations (price going up → drivers rush in → price drops → drivers leave), and regulatory constraints.

Design Uber Eats' order batching system. A single delivery partner picks up orders from multiple nearby restaurants and delivers to multiple customers in one trip. How do you optimise the route and assignment?

Design Uber's payment reconciliation system. Uber collects payments in 70+ countries with different currencies, payment methods, and regulations. How do you reconcile driver payouts against rider charges and detect discrepancies?

Your pipeline SLA requires data to be available in the warehouse by 6 AM. Yesterday it missed at 7:30 AM. Walk me through your incident response.

Trip fares are occasionally being calculated twice, double-charging riders. The fare calculation service is stateless and horizontally scaled behind a queue. Where would you look first, and what's the likely root cause?

Design Stripe's webhook delivery system. Stripe sends millions of webhooks per day to customer endpoints that can be slow, down, or buggy. Cover: at-least-once delivery guarantees, retry with exponential backoff, ordering guarantees (or lack thereof), how to handle a customer endpoint that is down for 24h, and the dead-letter strategy.

Design Stripe's fraud detection system. For every payment, you must decide accept/decline in under 100ms using hundreds of signals. Cover: feature engineering, real-time ML scoring, rule engine, and how you minimise false positives (declined legitimate payments).

Design Stripe's subscription billing system (Stripe Billing). Customers are charged on recurring intervals (monthly/annual) in different currencies, with proration for plan upgrades. Cover: invoice generation, dunning for failed payments, and SCA (Strong Customer Authentication) compliance.

Design a webhook delivery system that must guarantee at-least-once delivery to customer endpoints, with retries and backoff, without letting one slow customer endpoint block delivery to others.

A Kafka consumer processing payment events crashes and restarts, re-reading the last uncommitted batch. How do you make the payment-processing logic itself idempotent, not just rely on Kafka?

Design iCloud Photo Library — 2 billion photos synced across iPhone, iPad, and Mac. Cover: deduplication (perceptual hashing), delta sync (only changed chunks), offline-first with conflict resolution, and privacy — how Apple performs server-side processing without seeing unencrypted photos.

Design Netflix's video recommendation system. Netflix serves 250 million subscribers. How do you generate personalised top-10 rows for each user in real time? Cover: collaborative filtering, content-based signals, contextual signals (time of day, device), and how you handle the cold-start problem for new users.

Design Netflix's content delivery pipeline. A new movie is added. Walk through transcoding into 50+ formats, uploading to CDN edge nodes globally, and ensuring playback starts within 200ms for any device anywhere.

Design Netflix's A/B testing platform. Engineers run 100+ experiments simultaneously on different features. How do you assign users to buckets consistently, collect metrics, measure statistical significance, and avoid interaction effects?

Design a data pipeline to ingest CDC (Change Data Capture) events from a MySQL database into a Snowflake data warehouse in near real-time.

Kafka topics have a retention period, but consumer offsets also have their own retention. What happens if a consumer group is offline longer than the offset retention period?

Design Amazon's recommendation engine (customers who bought X also bought Y). Cover: collaborative filtering vs content-based, item-item similarity at scale, real-time vs batch computation, and how you handle the cold-start problem for new products.

You have a Spark job that ingests 500GB of clickstream data daily. It is failing with OOM errors on the executor. How do you debug and fix it?

Design a notification system (push/email/SMS) that sends 10 million notifications per day to users across time zones with delivery guarantees.

Design Google Maps ETA calculation. How do you compute the fastest route for 1B+ users in real time? Cover: graph representation of road network, Dijkstra vs A* vs Contraction Hierarchies, live traffic data ingestion, and how you handle road closures.

Design Flipkart's product search and ranking system. A user searches "wireless earphones under 2000". How do you retrieve, rank, and filter 10,000+ matching products in under 200ms?

Design Flipkart's order management system. From "Place Order" to "Delivered", walk through every state transition, the microservices involved, how you handle partial failures (payment succeeds but inventory deduct fails), and eventual consistency.

Design Airbnb's trust and safety system. How do you detect fraudulent listings, scam messages, and fake reviews at scale without blocking legitimate hosts and guests?

Design Airbnb's review system. Guests and hosts review each other after a stay. Reviews are revealed simultaneously to prevent bias. Cover: the blind reveal mechanism, preventing fake reviews, and abuse detection.

Design Airbnb's host payout system. Hosts in 220+ countries receive payouts in local currency via bank transfer, PayPal, or Payoneer. Cover: currency conversion, scheduled payouts, failed payout handling, and tax compliance.

Design Paytm's cashback system. When a user completes a transaction, they may receive cashback based on complex rules (merchant, time, amount, user segment). The cashback is credited to their wallet. Cover: rule engine, deduplication, and fraud prevention.

Design Paytm's KYC (Know Your Customer) pipeline. Users must submit Aadhaar/PAN documents for verification. How do you store documents securely (encrypted at rest), verify them (ML + manual review), and handle status callbacks?

Design Adobe Experience Cloud's customer data platform. Unify customer profiles from website visits, email clicks, and CRM data across devices and sessions. Cover identity resolution (same person on mobile and desktop), real-time segment computation, and privacy (GDPR right-to-erasure).

Design Adobe Sign (e-signature platform). A contract is sent to 5 signatories who must sign in a specific order. Cover: document integrity (hash verification), audit trail, legal compliance (e-SIGN Act), and what happens if a signatory refuses.

You have a streaming PySpark job reading from Kafka that processes IoT sensor events. Events can arrive up to 10 minutes late. How do you handle late data correctly?