Clarify the goal & success metrics Users, workload mix (read/write %, real‑time vs. async), target geos, SLIs (P50/P99, availability), cost guardrails.
Constraints & back‑of‑the‑envelope Requests/sec, data size/day, retention, peak factors; call out assumptions explicitly.
API & data model (v0) Public API (request/response), core entities, access patterns → drive storage choice.
High‑level architecture Client → Edge (CDN/WAF) → Gateway → Services → Datastores → Async (queue/stream) → Analytics → Admin.
Deep dives & trade‑offs
- Storage: SQL vs NoSQL; sharding keys; secondary indexes.
- Caching: CDN/edge, read‑through/write‑through, TTLs, eviction, stampede protection.
- Consistency: Quorum math (R+W > N), read‑your‑writes strategies.
- Scalability: Stateless services, idempotent handlers, bulkheads, circuit breakers, retries with jitter.
- Search/Feeds: Inverted indexes, fan‑out vs fan‑in, push vs pull, precompute vs on‑read.
- Streaming: Exactly‑once semantics (effectively‑once), DLQs, replays.
- Security/Privacy: AuthN/Z, PII segregation, KMS, audit logs, GDPR/retention.
- Operability: SLIs/SLOs, dashboards, alerting, runbooks, chaos drills.
Bottlenecks & failure modes Single shard hot‑keys, thundering herds, split‑brain, cascading retries, slow consumers. Offer mitigations.
Evolution plan Phased rollout, A/B & feature flags, multi‑region strategy, cost projection, what you’d do with ×10 traffic.
