System Design
Keep the architecture muscles warm: scaling tradeoffs, caches, queues, consistency, API boundaries, and reliability habits that matter under load.
40 concepts248 live tasks
Start practicing freeWhat's inside
Scalability & Load
Capacity estimates turn vague scale into constraints
Latency, throughput, and utilization are different levers
Vertical scaling buys simplicity; horizontal scaling buys headroom
Stateless services are easier to replicate
Backpressure makes overload visible before collapse
Graceful degradation preserves the most valuable behavior
Averages hide hot keys and skew
Load shedding rejects work to protect the system
Caching & CDNs
Cache-aside keeps cache misses explicit
Write-through and write-behind move durability risk
Invalidation is where caching becomes a correctness problem
TTL jitter prevents synchronized expirations
CDNs cache close to users but obey HTTP semantics
Request coalescing lets one miss fill many waiters
A cache stampede turns one expiration into a traffic spike
A cache is not the source of truth unless you design it that way
Queues & Async Processing
Queues decouple producers from consumer timing
At-least-once delivery means duplicates are normal
Idempotency keys make retries safe
Ordering is usually scoped to a partition
Retries need backoff and a place for poison messages
Sagas coordinate distributed work with compensation
The transactional outbox closes the database-message gap
A stream records events; a work queue assigns tasks
Data, Consistency & Storage
Replication lag creates stale reads
The shard key decides both scale and pain
CAP is about behavior during partitions
Quorums trade latency for stronger replica agreement
Indexes are built for access patterns, not tables
Read replicas scale reads but not write bottlenecks
Denormalization precomputes reads and complicates writes
Consistency models are user-visible promises
APIs, Reliability & Observability
API versioning is contract management
Rate limits protect shared capacity and product rules
Authentication identifies; authorization decides
Circuit breakers stop repeated calls to a failing dependency
Every remote call needs a timeout and a deadline budget
SLOs turn reliability into an explicit product target
Metrics, logs, and traces answer different incident questions
RPO and RTO define disaster-recovery promises