afrexai-api-architect（afrexai API 架构师）

API Architect — Full Lifecycle API Development

Design, build, test, document, secure, and monitor production-grade APIs. Not just curl commands — a complete engineering methodology.

When to Use

• Designing a new API (REST, GraphQL, or gRPC)
• Reviewing an existing API for quality, consistency, or security
• Generating or validating OpenAPI/Swagger specs
• Building comprehensive test suites (unit, integration, contract, load)
• Debugging production API issues
• Planning API versioning and deprecation
• Setting up monitoring, rate limiting, and error handling

Phase 1: API Design

Design-First Approach

Always design before coding. The spec IS the contract.

Resource Modeling

Map your domain to resources using this template:

# api-design.yaml
service: order-management
base_path: /api/v1
resources:
  - name: orders
    path: /orders
    description: Customer purchase orders
    identifier: order_id (UUID)
    parent: null
    operations: [list, create, get, update, cancel]
    sub_resources:
      - name: line_items
        path: /orders/{order_id}/items
        operations: [list, add, update, remove]
      - name: payments
        path: /orders/{order_id}/payments
        operations: [list, create, get, refund]
    states: [draft, confirmed, processing, shipped, delivered, cancelled]
    transitions:
      - from: draft → to: confirmed (action: confirm)
      - from: confirmed → to: processing (action: process)
      - from: processing → to: shipped (action: ship)
      - from: shipped → to: delivered (action: deliver)
      - from: [draft, confirmed] → to: cancelled (action: cancel)

Naming Conventions Checklist

HTTP Methods — Decision Matrix

Need to...                          → Method   Idempotent?  Safe?
Get a resource or collection        → GET      Yes          Yes
Create a new resource               → POST     No           No
Full replace of a resource          → PUT      Yes          No
Partial update of a resource        → PATCH    No*          No
Remove a resource                   → DELETE   Yes          No
Check if resource exists            → HEAD     Yes          Yes
List allowed methods                → OPTIONS  Yes          Yes

* PATCH can be idempotent if using JSON Merge Patch

Status Code Decision Tree

Success?
├── Created something new? → 201 Created (Location header)
├── Accepted for async processing? → 202 Accepted (include status URL)
├── No body to return? → 204 No Content
└── Returning data? → 200 OK

Client error?
├── Malformed request syntax? → 400 Bad Request
├── No/invalid credentials? → 401 Unauthorized
├── Valid credentials but insufficient permissions? → 403 Forbidden
├── Resource doesn't exist? → 404 Not Found
├── Method not allowed on resource? → 405 Method Not Allowed
├── Conflict with current state? → 409 Conflict
├── Resource permanently gone? → 410 Gone
├── Validation failed? → 422 Unprocessable Entity
├── Too many requests? → 429 Too Many Requests (Retry-After header)
└── Precondition failed (etag mismatch)? → 412 Precondition Failed

Server error?
├── Unexpected failure? → 500 Internal Server Error
├── Upstream dependency failed? → 502 Bad Gateway
├── Temporarily overloaded? → 503 Service Unavailable (Retry-After)
└── Upstream timeout? → 504 Gateway Timeout

Request/Response Design

Standard Response Envelope

// Success (single resource)
{
  "data": { "id": "ord_abc123", "status": "confirmed", ... },
  "meta": { "request_id": "req_xyz789" }
}

// Success (collection)
{
  "data": [ ... ],
  "meta": { "request_id": "req_xyz789" },
  "pagination": {
    "total": 142,
    "page": 2,
    "per_page": 20,
    "total_pages": 8,
    "next": "/api/v1/orders?page=3&per_page=20",
    "prev": "/api/v1/orders?page=1&per_page=20"
  }
}

// Error
{
  "error": {
    "code": "VALIDATION_FAILED",
    "message": "Request validation failed",
    "details": [
      { "field": "email", "message": "Must be a valid email address", "code": "INVALID_FORMAT" },
      { "field": "age", "message": "Must be at least 18", "code": "MIN_VALUE", "min": 18 }
    ]
  },
  "meta": { "request_id": "req_xyz789" }
}

Pagination Patterns — When to Use Which

Filtering, Sorting, Field Selection

# Filtering
GET /orders?status=active&created_after=2024-01-01&total_min=100

# Sorting (prefix - for descending)
GET /orders?sort=-created_at,total

# Field selection (reduce payload)
GET /orders?fields=id,status,total,customer.name

# Search
GET /products?q=wireless+headphones

# Combined
GET /orders?status=active&sort=-created_at&fields=id,status,total&page=1&per_page=10

Phase 2: OpenAPI Specification

Generate OpenAPI 3.1 Spec

For each resource in your design, generate a complete spec:

openapi: 3.1.0
info:
  title: Order Management API
  version: 1.0.0
  description: |
    Order lifecycle management.
    
    ## Authentication
    All endpoints require Bearer token authentication.
    
    ## Rate Limits
    - Standard: 100 req/min
    - Bulk operations: 10 req/min
  contact:
    name: API Support
    email: api@example.com
  license:
    name: MIT
servers:
  - url: https://api.example.com/v1
    description: Production
  - url: https://staging-api.example.com/v1
    description: Staging

paths:
  /orders:
    get:
      operationId: listOrders
      summary: List orders
      tags: [Orders]
      parameters:
        - $ref: '#/components/parameters/PageParam'
        - $ref: '#/components/parameters/PerPageParam'
        - name: status
          in: query
          schema:
            $ref: '#/components/schemas/OrderStatus'
        - name: created_after
          in: query
          schema:
            type: string
            format: date-time
      responses:
        '200':
          description: Order list
          content:
            application/json:
              schema:
                $ref: '#/components/schemas/OrderListResponse'
        '401':
          $ref: '#/components/responses/Unauthorized'

    post:
      operationId: createOrder
      summary: Create an order
      tags: [Orders]
      requestBody:
        required: true
        content:
          application/json:
            schema:
              $ref: '#/components/schemas/CreateOrderRequest'
            examples:
              basic:
                summary: Basic order
                value:
                  customer_id: "cust_abc"
                  items:
                    - product_id: "prod_xyz"
                      quantity: 2
      responses:
        '201':
          description: Order created
          headers:
            Location:
              schema:
                type: string
              description: URL of created order
          content:
            application/json:
              schema:
                $ref: '#/components/schemas/OrderResponse'
        '422':
          $ref: '#/components/responses/ValidationError'

components:
  securitySchemes:
    BearerAuth:
      type: http
      scheme: bearer
      bearerFormat: JWT

  parameters:
    PageParam:
      name: page
      in: query
      schema: { type: integer, minimum: 1, default: 1 }
    PerPageParam:
      name: per_page
      in: query
      schema: { type: integer, minimum: 1, maximum: 100, default: 20 }

  responses:
    Unauthorized:
      description: Missing or invalid authentication
      content:
        application/json:
          schema:
            $ref: '#/components/schemas/ErrorResponse'
    ValidationError:
      description: Request validation failed
      content:
        application/json:
          schema:
            $ref: '#/components/schemas/ErrorResponse'

security:
  - BearerAuth: []

Spec Quality Checklist (score each 0-2)

Score: ___/20 (Target: 16+)

Phase 3: Implementation Patterns

Request Validation Layer

Every endpoint MUST validate before processing:

Validation Order:
1. Content-Type header (reject non-JSON early)
2. Authentication (401 before wasting cycles)
3. Authorization (403 - does this user have access?)
4. Path parameters (404 - does the resource exist?)
5. Query parameters (400 - valid types/ranges?)
6. Request body schema (422 - valid structure?)
7. Business rules (422 - valid state transition?)

Error Handling — Standard Error Codes

Define a consistent error code enum for your API:

# Authentication & Authorization
AUTH_REQUIRED          — No credentials provided
AUTH_INVALID           — Invalid/expired credentials
AUTH_INSUFFICIENT      — Valid credentials, wrong permissions
AUTH_RATE_LIMITED       — Too many auth attempts

# Validation
VALIDATION_FAILED      — Generic validation error (see details array)
INVALID_FORMAT         — Field format wrong (email, UUID, etc.)
REQUIRED_FIELD         — Required field missing
OUT_OF_RANGE           — Value outside allowed range
INVALID_ENUM           — Value not in allowed set

# Resource
NOT_FOUND              — Resource doesn't exist
ALREADY_EXISTS         — Duplicate (unique constraint)
CONFLICT               — State conflict (e.g., already cancelled)
GONE                   — Resource permanently deleted

# Business Logic
INSUFFICIENT_FUNDS     — Payment-related
QUOTA_EXCEEDED         — Usage limit reached
FEATURE_DISABLED       — Feature flag off
DEPENDENCY_FAILED      — Upstream service error

# System
INTERNAL_ERROR         — Unexpected server error
SERVICE_UNAVAILABLE    — Temporarily down
TIMEOUT                — Request took too long

Idempotency

For non-idempotent operations (POST), require an idempotency key:

Request:
POST /orders
Idempotency-Key: ord_req_abc123

Server behavior:
1. Check if Idempotency-Key was seen before
2. If yes → return cached response (same status, same body)
3. If no → process request, cache response for 24h
4. Key format: client-generated UUID or meaningful string

Rate Limiting

Standard headers to include:

X-RateLimit-Limit: 100          # Max requests per window
X-RateLimit-Remaining: 67       # Remaining in current window
X-RateLimit-Reset: 1706886400   # Unix timestamp when window resets
Retry-After: 30                 # Seconds to wait (on 429)

Rate limit tiers:

Phase 4: Testing Strategy

Test Pyramid for APIs

        /  E2E  \          — 5-10 critical user flows
       / Contract \        — Schema validation, backward compat
      / Integration \      — Database, external services, auth
     /    Unit Tests  \    — Business logic, validation, transforms
    ‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾

Test Checklist Per Endpoint

For EVERY endpoint, test these scenarios:

endpoint: POST /orders
tests:
  happy_path:
    - Creates order with valid data → 201
    - Returns created resource with ID
    - Location header points to new resource
    - Timestamps are set (created_at, updated_at)
  
  validation:
    - Missing required fields → 422 with field-level errors
    - Invalid field types (string where int expected) → 422
    - Empty body → 400
    - Invalid Content-Type → 415
    - Extra unknown fields → ignored or 422 (pick one, be consistent)
    - Boundary values (min/max length, 0, negative, empty string vs null)
  
  authentication:
    - No token → 401
    - Expired token → 401
    - Invalid token → 401
    - Valid token, wrong scope → 403
  
  authorization:
    - User accessing own resource → 200
    - User accessing other's resource → 403 or 404 (security choice)
    - Admin accessing any resource → 200
  
  edge_cases:
    - Duplicate creation (same idempotency key) → same 201 response
    - Concurrent creation race condition → one wins, one gets 409
    - Resource at max relationships → 422
    - Unicode in text fields → handled correctly
    - Very long strings → 422 with max length error
    - SQL injection in params → no effect (parameterized queries)
    - XSS in text fields → stored safely, escaped on output
  
  performance:
    - Response time < 200ms (p95)
    - List endpoint with 10K records → paginated, < 500ms
    - Bulk operation timeout handling

curl Testing Recipes

# === Setup ===
BASE="https://api.example.com/v1"
TOKEN="your_bearer_token"
alias api='curl -s -H "Authorization: Bearer $TOKEN" -H "Content-Type: application/json"'

# === CRUD Lifecycle Test ===
# Create
ORDER=$(api -X POST "$BASE/orders" -d '{"customer_id":"cust_1","items":[{"product_id":"prod_1","qty":2}]}')
ORDER_ID=$(echo "$ORDER" | jq -r '.data.id')
echo "Created: $ORDER_ID"

# Read
api "$BASE/orders/$ORDER_ID" | jq .

# Update
api -X PATCH "$BASE/orders/$ORDER_ID" -d '{"notes":"Rush order"}' | jq .

# List with filters
api "$BASE/orders?status=draft&sort=-created_at&per_page=5" | jq .

# Action (state transition)
api -X POST "$BASE/orders/$ORDER_ID/confirm" | jq .

# Delete
curl -s -o /dev/null -w "%{http_code}" -X DELETE -H "Authorization: Bearer $TOKEN" "$BASE/orders/$ORDER_ID"

# === Error Testing ===
# No auth
curl -s "$BASE/orders" | jq .error

# Invalid body
api -X POST "$BASE/orders" -d '{"invalid": true}' | jq .error

# Not found
api "$BASE/orders/nonexistent" | jq .error

# === Performance ===
# Timing breakdown
curl -s -o /dev/null -w "DNS:%{time_namelookup} TCP:%{time_connect} TLS:%{time_appconnect} TTFB:%{time_starttransfer} Total:%{time_total}\
" -H "Authorization: Bearer $TOKEN" "$BASE/orders"

# Quick load test (50 requests, 10 concurrent)
seq 50 | xargs -P10 -I{} curl -s -o /dev/null -w "%{http_code} %{time_total}s\
" -H "Authorization: Bearer $TOKEN" "$BASE/orders"

Contract Testing

Validate your API hasn't broken backward compatibility:

# contract-tests.yaml
contract:
  name: Order API Contract
  version: 1.0.0
  
  rules:
    # These changes are SAFE (non-breaking)
    safe:
      - Adding new optional fields to responses
      - Adding new endpoints
      - Adding new optional query parameters
      - Adding new enum values (if clients handle unknown)
      - Widening a constraint (min: 5 → min: 1)
    
    # These changes are BREAKING
    breaking:
      - Removing a response field
      - Renaming a response field
      - Changing a field type
      - Adding a new required request field
      - Removing an endpoint
      - Narrowing a constraint (max: 100 → max: 50)
      - Changing error response format
      - Removing an enum value
    
    # Verify after every change
    checks:
      - All existing fields still present in responses
      - All existing field types unchanged
      - All existing required fields still required (no more, no fewer)
      - Default values unchanged
      - Error format unchanged

Phase 5: Security

Security Checklist (audit every API)

authentication:
  - [ ] All endpoints require auth (except /health, /docs, public webhooks)
  - [ ] Tokens expire (short-lived access + long-lived refresh)
  - [ ] Token rotation supported
  - [ ] Failed auth returns 401 with no info leakage
  - [ ] API keys are hashed in storage (never plain text)

authorization:
  - [ ] Resource-level checks (user can only access their data)
  - [ ] Endpoint-level checks (role-based access)
  - [ ] No IDOR vulnerabilities (can't guess other users' resource IDs)
  - [ ] Admin endpoints separately protected
  - [ ] Webhook endpoints verify signatures

input_validation:
  - [ ] All inputs validated server-side (never trust client)
  - [ ] SQL injection prevented (parameterized queries only)
  - [ ] NoSQL injection prevented
  - [ ] Path traversal prevented
  - [ ] Request size limited (body, headers, URL length)
  - [ ] File upload types restricted and scanned

output_security:
  - [ ] No sensitive data in responses (passwords, tokens, internal IDs)
  - [ ] No stack traces in production errors
  - [ ] Consistent error format (no info leakage in different error types)
  - [ ] PII redacted in logs

transport:
  - [ ] HTTPS only (HTTP redirects to HTTPS)
  - [ ] HSTS header set
  - [ ] TLS 1.2+ required
  - [ ] CORS configured restrictively (specific origins, not *)
  
headers:
  - [ ] X-Content-Type-Options: nosniff
  - [ ] X-Frame-Options: DENY
  - [ ] Content-Security-Policy set
  - [ ] No Server version header
  - [ ] Cache-Control: no-store for sensitive endpoints

CORS Configuration

# Restrictive (recommended)
cors:
  origins:
    - https://app.example.com
    - https://admin.example.com
  methods: [GET, POST, PUT, PATCH, DELETE]
  headers: [Authorization, Content-Type, X-Request-ID]
  credentials: true
  max_age: 3600

# Common mistakes to avoid:
# ❌ Access-Control-Allow-Origin: *  (with credentials)
# ❌ Reflecting Origin header without validation
# ❌ Allowing all methods/headers

Phase 6: Versioning & Deprecation

Versioning Strategy Decision

Recommended: URL path for major versions, header for minor variations.

Deprecation Playbook

Timeline:
1. T+0: Announce deprecation (docs, changelog, email)
2. T+0: Add Deprecation + Sunset headers to old endpoints
3. T+30d: Log warnings for old endpoint usage
4. T+60d: Email heavy users of old endpoint directly
5. T+90d: Return 299 warning header
6. T+180d: Shut down old endpoint (410 Gone)

Headers:
Deprecation: true
Sunset: Sat, 01 Jun 2025 00:00:00 GMT
Link: <https://api.example.com/v2/orders>; rel="successor-version"

Migration Guide Template

# Migrating from v1 to v2

## Breaking Changes
1. `user.name` split into `user.first_name` + `user.last_name`
2. Pagination changed from offset to cursor-based
3. Error format updated (see new schema)

## Step-by-Step Migration
1. Update your client SDK to v2 (`npm install @example/sdk@2`)
2. Update response parsing for split name fields
3. Replace `?page=N` with `?after=cursor` pagination
4. Update error handling for new error format

## Compatibility Mode
Set `X-Compat-Mode: v1` header to get v1-style responses from v2 endpoints.
Available until 2025-06-01.

Phase 7: Monitoring & Observability

Key Metrics Dashboard

availability:
  - Uptime percentage (target: 99.9% = 8.7h downtime/year)
  - Health check status (/health endpoint)
  - Error rate (5xx / total requests)

performance:
  - p50 latency (target: < 100ms)
  - p95 latency (target: < 500ms)
  - p99 latency (target: < 1000ms)
  - Throughput (requests/second)
  - Time to first byte (TTFB)

business:
  - Requests per endpoint (usage patterns)
  - Unique API consumers/day
  - Error rate by endpoint
  - Rate limit hits/day
  - Authentication failures/day

infrastructure:
  - Database query time (p95)
  - Connection pool utilization
  - Memory/CPU per instance
  - Queue depth (async operations)

Structured Logging

Every request should log:

{
  "timestamp": "2024-01-15T10:30:00.000Z",
  "level": "info",
  "request_id": "req_abc123",
  "method": "POST",
  "path": "/api/v1/orders",
  "status": 201,
  "duration_ms": 45,
  "user_id": "usr_xyz",
  "ip": "203.0.113.1",
  "user_agent": "MyApp/2.0",
  "request_size": 256,
  "response_size": 1024
}

Health Check Endpoint

// GET /health — for load balancers (simple)
{ "status": "ok" }

// GET /health/detailed — for monitoring (authenticated)
{
  "status": "degraded",
  "version": "1.5.2",
  "uptime_seconds": 86400,
  "checks": {
    "database": { "status": "ok", "latency_ms": 5 },
    "redis": { "status": "ok", "latency_ms": 2 },
    "external_payment_api": { "status": "degraded", "latency_ms": 2500, "error": "timeout" },
    "disk": { "status": "ok", "free_gb": 45.2 }
  }
}

Phase 8: API Review Scoring

When reviewing an existing API, score across these dimensions:

API Quality Rubric (0-100)

Score: ___/100

Review Output Template

## API Review: [Service Name]

**Date:** YYYY-MM-DD
**Reviewer:** [Agent]
**Score:** XX/100 (Rating)

### Summary
[2-3 sentence overview of API quality]

### Scores by Dimension
- Design Consistency: X/20 — [key finding]
- Documentation: X/15 — [key finding]
- Error Handling: X/15 — [key finding]
- Security: X/20 — [key finding]
- Performance: X/15 — [key finding]
- Developer Experience: X/15 — [key finding]

### Critical Issues (fix immediately)
1. [Issue + recommendation]

### High Priority (fix this sprint)
1. [Issue + recommendation]

### Nice to Have (backlog)
1. [Issue + recommendation]

### Positive Highlights
- [What's working well]

GraphQL-Specific Guidance

Schema Design Principles

# Good: clear types, nullable where appropriate, connections for lists
type Order {
  id: ID!
  status: OrderStatus!
  customer: Customer!
  items(first: Int, after: String): ItemConnection!
  total: Money!
  createdAt: DateTime!
  updatedAt: DateTime!
}

type Money {
  amount: Int!       # cents, not dollars (avoid float)
  currency: Currency!
}

enum OrderStatus {
  DRAFT
  CONFIRMED
  PROCESSING
  SHIPPED
  DELIVERED
  CANCELLED
}

# Mutations return the modified resource + errors
type CreateOrderPayload {
  order: Order
  errors: [UserError!]!
}

type UserError {
  field: [String!]
  message: String!
  code: ErrorCode!
}

GraphQL Anti-Patterns

Edge Cases & Gotchas

Timezone Handling

• Always store and return UTC (ISO 8601: 2024-01-15T10:30:00Z)
• Accept timezone in input, convert to UTC immediately
• Never use local server time

Large Payloads

• Set Content-Length limits (e.g., 1MB default, 10MB for uploads)
• Use streaming for file uploads (multipart/form-data)
• Compress responses (Accept-Encoding: gzip)
• For very large exports → return 202 + poll status endpoint

Eventual Consistency

• If using async processing, always return 202 + status URL
• Include estimated completion time when possible
• Client should poll with exponential backoff

Concurrent Updates

• Use ETags for optimistic concurrency:

- GET returns ETag: "v1" header - PUT/PATCH sends If-Match: "v1" header - Server returns 412 if resource changed since

Webhook Design

• Include event type, timestamp, and full resource in payload
• Sign payloads (HMAC-SHA256)
• Expect retries (make handlers idempotent)
• Return 200 quickly, process async
• Include webhook ID for deduplication

Quick Commands