System Design — Senior

1Consistency & correctness

0/6▶

Senior rounds probe tradeoffs under distribution. Reach for consistency vocabulary naturally and know when eventual is acceptable.

Get the models straight

CAP and PACELC in plain words; why "CA" isn't real for a distributed store.

Wikipedia — PACELC.

Strong vs eventual consistency; where eventual is fine (feeds) vs not (balances).

Designing Data-Intensive Applications — ch. 5 & 9.

DB isolation levels & the anomalies they prevent.

PostgreSQL — Isolation.

Idempotency keys and exactly-once effects (vs impossible exactly-once delivery).

Stripe — Idempotency.

The outbox pattern / CDC to keep a DB and a bus in sync.

microservices.io — Outbox.

Quorums, leader/follower, and consensus (Raft) at a glance.

Raft — visual.

2Data at scale

0/5▶

You own the data model and its failure modes: partitioning, hot keys, and cache correctness.

Partitioning & caching

Sharding strategies: hash vs range vs directory; resharding; consistent hashing.

Consistent hashing.

Hot keys / celebrity problem — detection and mitigation.

ByteByteGo — search "hot partition".

Cache invalidation & stampede protection (locks, request coalescing, TTL jitter).

AWS Builders' Library.

Choosing a store per access pattern: KV, document, wide-column, search, time-series.

DDIA — ch. 3.

Bloom filters & other probabilistic structures for scale.

Bloom filter.

3Designing for failure

0/5▶

The senior differentiator: assume everything fails and keep the blast radius small.

Resilience patterns

Retries with exponential backoff + jitter; retry budgets to avoid retry storms.

AWS — Backoff with jitter.

Circuit breakers & bulkheads to isolate a failing dependency.

Fowler — Circuit Breaker.

Backpressure and load shedding — drop or queue rather than topple.

AWS — Load shedding.

Graceful degradation, fallbacks, and the thundering-herd problem.

AWS Builders' Library.

Single points of failure & how you'd remove each.

System Design Primer.

4Observability & delivery

0/4▶

You make systems operable: measurable, safely deployable, capacity-planned.

Run it in production

Metrics, logs, traces; RED/USE method for dashboards.

SRE Book — Monitoring.

SLI / SLO / error budgets — how they drive decisions.

SRE Book — SLOs.

Safe deploys: canary, blue-green, feature flags, fast rollback.

Fowler — Blue-green.

Capacity planning & back-of-envelope headroom.

Primer — Estimates.

5Networking & performance

0/3▶

Senior designs account for the physics: latency, connections, and payload size.

The wire

Latency numbers everyone should know (RAM vs SSD vs network vs cross-region).

Latency numbers.

Connection reuse, keep-alive, HTTP/2, and connection pools.

MDN — Connection management.

Websockets vs SSE vs long polling for real-time.

MDN — WebSockets.

6Harder prompts

0/8▶

Open-ended prompts that combine the ideas above. Drive them; state assumptions; defend tradeoffs.

Design these end to end

News feed / timeline — fan-out on write vs read; the celebrity problem.

ByteByteGo — search "news feed".

Chat / messaging — delivery guarantees, ordering, presence, websockets at scale.

ByteByteGo — search "chat system".

Distributed rate limiter — shared counters, clock skew, sliding window at scale.

ByteByteGo — search "rate limiter".

Payment / ledger — idempotency, double-entry, reconciliation.

Stripe — Idempotency.

Distributed cache — sharding, eviction, replication; consistent hashing.

ByteByteGo — search "distributed cache".

Proximity / geo search — geohashing, quadtrees, sharding by region.

ByteByteGo — search "proximity service".

Distributed lock / leader election — fencing tokens; why a naive lock is unsafe.

Kleppmann — Distributed locking.

Ad-click / event aggregation at scale — dedup, exactly-once effects.

ByteByteGo — search "ad click aggregation".