Core Database
Store Choice
Pick one core DB
Use one DB you know; most interviewers do not need an explicit SQL vs NoSQL comparison.
Use Postgres for product
Recommended relational default for product designs; ACID properties help data integrity.
Use DynamoDB for infra
Recommended NoSQL default for infrastructure designs; widely accepted with broad features.
Relational Tools
- SQL Joins: Combine tables for queries; minimize them because joins can be a major bottleneck.
- Indexes: Speed queries; relational DBs support many, multi-column, geospatial, and full-text indexes.
- RDBMS Transactions: Group operations atomically so either all succeed or all fail.
NoSQL Scaling
- Consistent Hashing: Distribute NoSQL data across many servers using consistent hashing.
- Sharding: Distribute NoSQL data across many servers using sharding.
Blob and Search
Use S3 for large blobs
Default if no experience; store images, videos, and files outside the database.
Keep metadata in DB
Store URL pointers in Postgres or DynamoDB; avoid S3 as the primary database.
Use presigned URLs
Client uploads or downloads directly with temporary access; server records status.
Chunk large uploads
Multipart upload resumes failures and can upload chunks in parallel.
Edge Routing
API Gateway
Default for nearly all product designs; routes requests and handles auth, rate limits, logging.
Load Balancer
Use instead of an API Gateway for persistent connections like websockets, or for flexible routing and lower downstream connection load.
L4 Load Balancer
Use for persistent connections like websockets.
L7 Load Balancer
Use otherwise for flexible routing and lower downstream connection load.
Async Processing
Async Fit
Use queues for async work
Buffer bursts and distribute work to worker pools at their own pace.
Don't queue low-latency sync
Strong latency constraints like < 500ms will likely break if a queue is added.
Use streams for replay
Retain data for a configurable period so consumers can re-read by position or time.
Use streams for event sourcing
Store changes as events; replay for audit, rollback, or reconstruction.
Delivery Controls
- Backpressure: Slow or reject producers when queues are overwhelmed; capacity shortfalls cannot be hidden forever.
- Dead Letter Queues: Store messages that cannot be processed for debugging and auditing.
- Scaling with Partitions: Partition queues or streams by key so related messages/events stay together.
Coordination
Lock Usage
Distributed Lock
Hold a resource across systems for a short time.
Ticket During Checkout
Use it for a resource such as a ticket during checkout.
Lock Expiry
Lock Expiry
Set locks to expire.
Crashes
Use expiry, e.g. 10 minutes, so crashes do not leave resources locked.
Deadlock Prevention
Deadlocks
Avoid lock acquisition from far-flung code.
Infrastructure
Avoid lock acquisition from infrastructure so prevention is recognizable.
Caching
Cache Placement
- Distributed Cache: Store expensive data in memory to reduce DB queries and lower latency.
- CDN: Cache content near users; works for static assets, dynamic content, and API responses.
Cache Consistency
Expire Cached Data
Expire cached data when the database changes, like an event venue update.
Update Cached Data
Update cached data when the database changes, like an event venue update.
Write-Through Cache
Write through to the datastore rather than only writing the cache first.
Write-Back Cache
Write to cache first and datastore asynchronously; faster writes but possible data loss.
Your account is free and you can post anonymously if you choose.