High-Level Design (HLD / Distributed Systems)

The canonical 4-step approach (requirements → estimation → high-level design → deep dive), capacity-estimation method, time budgeting, and how FAANG / EU / regional styles differ.

CAP stated correctly, the PACELC extension, the consistency spectrum (linearizable → eventual), ACID vs BASE, and quorum reads/writes.

Single-leader / multi-leader / leaderless replication, sync vs async, replication lag, sharding strategies, consistent hashing, resharding, and hot partitions.

Cache layers, the four caching patterns, eviction policies, cache stampede mitigation, invalidation, Redis vs Memcached, and two-tier (L1/L2) caching.

Queue vs log, delivery guarantees, ordering, backpressure, dead-letter queues, the outbox pattern, and CDC. RabbitMQ vs Kafka.

When to choose relational, document, key-value, wide-column, graph, search, or time-series stores, and the cost of polyglot persistence.

REST principles and idempotency, REST vs gRPC vs GraphQL, real-time options (WebSockets/SSE/long polling), versioning, and pagination.

Distributed rate limiting with Redis, per-user/IP/key granularity, edge vs application limits, the API-gateway role, and 429 + Retry-After.

L4 vs L7, balancing algorithms, consistent hashing for cache-aware routing, sticky sessions, health checks, and LB redundancy.

Monolith vs microservices vs modular monolith, service discovery, API gateway, service mesh, distributed tracing, and the Saga + outbox patterns.

The storage hierarchy and latency numbers, B-trees vs LSM-trees, row vs columnar storage, object/block/file storage, and tiered hot/warm/cold data.

What a CDN does, edge cache invalidation, edge compute, anycast vs DNS geo-routing, HTTP/2-3, and TLS termination at the edge.

The three pillars, structured logging, metric types, distributed tracing with OpenTelemetry, and SLI/SLO/SLA with error budgets.

The most common HLD warm-up. ID-generation strategies, SQL vs NoSQL choice, caching, custom URLs, and analytics — with FAANG / EU / regional variants.

Timeline generation (pull / fan-out-on-write / hybrid), feed ranking, the celebrity (hot-user) problem, media storage, and caching layers.

1:1 and group chat, online presence, delivery/read receipts, media, WebSocket vs polling, and sharding by chat ID.

Geospatial indexing (geohash / S2 / H3), driver-rider matching, surge pricing, real-time location tracking, and payment integration.

Feed generation, image storage and delivery (S3 + CDN), thumbnails, stories (24h TTL), likes/comments at scale, and hashtag search.

Video encoding pipeline, adaptive bitrate streaming (HLS/DASH), CDN delivery, recommendations, and view counting at scale.

Channel abstraction, template service, user preferences, a delivery queue with retries and DLQ, per-channel rate limiting, and provider failover.

URL frontier (priority + dedup), politeness (robots.txt, per-domain rate limit), DNS caching, content storage, duplicate detection, and distributed coordination.

Trie data structure, prefix caching, top-K suggestions, ranking by popularity/personalization, and the trend-update pipeline.

Consistent hashing, virtual nodes, replication, vector clocks, read repair, Merkle trees, gossip, and sloppy quorum.

Job queue, worker assignment, leader election (Zookeeper/etcd), missed-run handling, distributed locks, idempotency, and monitoring.

Operational Transformation vs CRDTs, real-time sync over WebSockets, presence, cursor sharing, version history, and offline editing.

Idempotency keys, distributed transactions via Saga, reconciliation, double-entry bookkeeping, fraud hooks, and processor integration (Stripe/Click/Payme model).

Exam session management, real-time scoring, leaderboards, payment integration, content delivery, and anti-cheat/proctoring — an end-to-end real-system case.

How many problems to solve, the internalizable template, per-section time-boxing, handling "I don't know" gracefully, and how to signal senior.