Shelf · System Design

System Design

Scale, caching, queues, consistency and the interview framework, always with numbers.

11 chapters · 69 cards · 0 chapters written · #5 in your reading order

1

The method

Planned
Difficulty
Priority
  1. Design interview method: the eight-step flowMediumP5Know it exists
  2. Clarify the user journey and scopeEasyP2Talk about it
  3. Turn nonfunctional goals into decisionsMediumP2Talk about it
  4. Back-of-envelope estimates with unitsMediumP2Know cold
  5. Draw and narrate one complete requestEasyP2Talk about it
  6. Manage a 45-minute design conversationMediumP5Know it exists
2

Building blocks

Planned
Difficulty
Priority
  1. DNS and CDN on the request pathEasyP2Talk about it
  2. L4/L7 load balancers and reverse proxiesMediumP2Talk about it
  3. App tiers and resource saturationEasyP2Talk about it
  4. Cache and queue as supporting componentsEasyP2Talk about it
  5. Database and object store responsibilitiesMediumP2Talk about it
  6. Search as a rebuildable read modelMediumP3Talk about it
3

Scaling

Planned
Difficulty
Priority
  1. Vertical versus horizontal scalingEasyP2Talk about it
  2. Sharding and hot keysHardP3Talk about it
  3. Stateless services still have stateMediumP2Talk about it
  4. Replication is not sharding or backupMediumP2Talk about it
  5. Read replicas and replica lagMediumP2Talk about it
  6. Find the next bottleneck before adding nodesMediumP2Talk about it
4

Caching

Planned
Difficulty
Priority
  1. Cache patterns and invalidation contractsMediumP2Know cold
  2. TTL and eviction solve different problemsEasyP2Know cold
  3. Invalidation under concurrent reads and writesHardP3Talk about it
  4. Prevent a thundering herd or stampedeMediumP3Talk about it
  5. Match Redis structures to the access patternMediumP3Talk about it
  6. What not to cacheEasyP2Talk about it
5

Async, queues & streams

Planned
Difficulty
Priority
  1. Messaging semantics: queue versus logMediumP2Talk about it
  2. Ordering, partitions and consumer progressHardP3Talk about it
  3. End-to-end idempotency and effect boundariesHardP2Know cold
  4. DLQs and controlled redriveMediumP3Talk about it
  5. Transactional outbox and relay failureHardP3Talk about it
  6. Backpressure and fan-out with bounded workMediumP2Talk about it
6

Distributed fundamentals

Planned
Difficulty
Priority
  1. Distributed consistency: CAP and PACELCHardP3Talk about it
  2. Consistency models and quorumsHardP3Talk about it
  3. Leader election and split-brain preventionHardP4Talk about it
  4. Distributed locks: leases and fencingHardP4Talk about it
  5. Clock skew and ordering without wall-time trustMediumP4Talk about it
  6. Two-phase commit versus sagaHardP3Talk about it
7

Reliability engineering

Planned
Difficulty
Priority
  1. Retry and timeout budgetsHardP2Know cold
  2. Circuit breakers and recovery probesMediumP3Talk about it
  3. Bulkheads and graceful degradationMediumP3Talk about it
  4. Rate limiting and overload enforcementMediumP2Talk about it
  5. Load shedding before queues become failureMediumP3Talk about it
  6. Health checks that support recoveryMediumP2Talk about it
8

APIs & service boundaries

Planned
Difficulty
Priority
  1. API contract design across REST, gRPC and GraphQLMediumP2Know cold
  2. Pagination styles with stable orderingMediumP2Know cold
  3. Versioning and compatibility contractsMediumP2Talk about it
  4. Service boundaries: monolith or microservicesHardP2Talk about it
  5. Decomposition without a distributed monolithHardP3Talk about it
  6. BFF and gateway responsibilitiesMediumP3Talk about it
9

Data-intensive system design

Planned
Difficulty
Priority
  1. Batch and streaming designMediumP3Talk about it
  2. Orchestration as durable run stateMediumP2Talk about it
  3. Data quality gates before publishingMediumP2Know cold
  4. Backfill and replay with late dataHardP2Talk about it
  5. Event time, windows and late-result policyHardP3Talk about it
  6. Exactly-once in pipelines has boundariesHardP3Talk about it
10

Case studies, end to end

Planned
Difficulty
Priority
  1. URL shortener, end to endMediumP3Talk about it
  2. Rate limiter, end to endHardP3Talk about it
  3. Notification service, end to endHardP3Talk about it
  4. S3 upload/processing service, end to endHardP2Talk about it
  5. Job scheduler, end to endHardP3Talk about it
  6. Chat, end to endHardP3Talk about it
  7. News feed, end to endHardP3Talk about it
  8. Metrics/logging system, end to endHardP3Talk about it
  9. Multi-tenant ETL platform, end to endHardP2Talk about it
11

Security & operability in design + the review checklist

Planned
Difficulty
Priority
  1. Tenant isolation and PII boundariesHardP3Talk about it
  2. Place authn/authz at every trust boundaryMediumP2Know cold
  3. Follow sensitive data through its lifecycleMediumP2Talk about it
  4. Cost as a design constraintMediumP4Talk about it
  5. Design the operating and recovery pathMediumP2Talk about it
  6. The what-did-I-forget design reviewMediumP5Know it exists