scalability-playbook（可扩展性手册）

Scalability Playbook

Systematic approach to identifying and resolving scalability bottlenecks.

Bottleneck Analysis

Current System Profile

Traffic: 1,000 req/min
Users: 10,000 active
Data: 100GB database
Response time: p95 = 500ms

Identified Bottlenecks

1. Database Queries

Symptom: Slow page loads (2-3s) Measurement: Query time p95 = 800ms Impact: HIGH - affects all reads Trigger: When p95 >500ms

2. Single Server

Symptom: High CPU (>80%) Measurement: Load average >4 Impact: MEDIUM - intermittent slowdowns Trigger: When CPU >70%

3. No Caching

Symptom: Repeated DB queries Measurement: Cache hit rate = 0% Impact: MEDIUM - unnecessary load Trigger: When query volume >10k/min

Scaling Strategies (Ordered)

Level 1: Quick Wins (Days)

1.1 Add Database Indexes

Problem: Slow queries Solution:

CREATE INDEX idx_users_email ON users(email);
CREATE INDEX idx_orders_user_created ON orders(user_id, created_at);

Expected Impact: 80% faster queries Cost: $0 Effort: 1 day

1.2 Enable Query Caching

Problem: Repeated queries Solution: Redis cache layer

const cached = await redis.get(`user:${userId}`);
if (cached) return JSON.parse(cached);

const user = await db.users.findById(userId);
await redis.setex(`user:${userId}`, 3600, JSON.stringify(user));

Expected Impact: 60% reduction in DB load Cost: $50/month Effort: 2 days

Level 2: Horizontal Scaling (Weeks)

2.1 Add Read Replicas

Problem: Read-heavy workload Solution: Route reads to replicas

Write Load: Primary DB
Read Load: 3x Read Replicas

Expected Impact: 3x read capacity Cost: $300/month Effort: 1 week

2.2 Load Balancer + Multiple Servers

Problem: Single point of failure Solution:

ALB
 ├── Server 1
 ├── Server 2
 └── Server 3

Expected Impact: 3x throughput Cost: $400/month Effort: 1 week

Level 3: Architecture Changes (Months)

3.1 CDN for Static Assets

Problem: Slow asset delivery Solution: CloudFront CDN Expected Impact: 90% faster asset loads Cost: $100/month Effort: 1 week

3.2 Async Processing

Problem: Slow sync operations Solution: Background job queues

// Before: Sync
await sendEmail(user);
await processPayment(order);
await updateAnalytics(event);
return response; // Waits 5+ seconds

// After: Async
await queue.add("send-email", { userId });
await queue.add("process-payment", { orderId });
await queue.add("update-analytics", { event });
return response; // Returns immediately

Expected Impact: 80% faster responses Cost: $50/month (SQS) Effort: 2 weeks

Level 4: Data Layer Optimization (Months)

4.1 Database Sharding

Problem: Single DB too large Solution: Shard by user_id

Shard 1: user_id 0-24999
Shard 2: user_id 25000-49999
Shard 3: user_id 50000-74999
Shard 4: user_id 75000-99999

Expected Impact: 4x capacity Cost: $1,200/month Effort: 2 months

4.2 Event-Driven Architecture

Problem: Tight coupling, cascading failures Solution: Message broker (Kafka)

Service A → Kafka → Service B
          ↘        ↗ Service C

Expected Impact: Better isolation, resilience Cost: $500/month Effort: 3 months

Scaling Triggers

| Metric           | Current | Warning | Critical | Action                  |
| ---------------- | ------- | ------- | -------- | ----------------------- |
| CPU              | 40%     | 70%     | 85%      | Add servers             |
| Memory           | 50%     | 75%     | 90%      | Upgrade instances       |
| DB Connections   | 20      | 40      | 50       | Add read replicas       |
| Query Time (p95) | 200ms   | 500ms   | 1000ms   | Add indexes             |
| Queue Depth      | 100     | 1000    | 5000     | Add workers             |
| Error Rate       | 0.1%    | 1%      | 5%       | Investigate immediately |

Phased Scaling Plan

Phase 1: Current → 10x (0-3 months)

Target: 10,000 req/min, 100K users

Actions:

1. Add database indexes (Week 1)
2. Implement Redis caching (Week 2)
3. Add 3x read replicas (Week 4)
4. Horizontal scale app servers (Week 6)
5. CDN for static assets (Week 8)

Cost: $500 → $1,000/month

Phase 2: 10x → 100x (3-12 months)

Target: 100,000 req/min, 1M users

Actions:

1. Database sharding (Month 4-6)
2. Multi-region deployment (Month 6-8)
3. Microservices extraction (Month 8-12)
4. Event-driven architecture (Month 10-12)

Cost: $1,000 → $10,000/month

Phase 3: 100x → 1000x (12-24 months)

Target: 1M req/min, 10M users

Actions:

1. Global CDN (Month 13)
2. Advanced caching (L1/L2) (Month 14-15)
3. Custom DB solutions (Month 16-18)
4. Edge computing (Month 18-20)

Cost: $10,000 → $100,000/month

Load Testing Plan

# Current baseline
hey -n 10000 -c 100 https://api.example.com/users

# Target 10x
hey -n 100000 -c 1000 https://api.example.com/users

# Measure:
# - Requests/sec
# - p50, p95, p99 latency
# - Error rate
# - Resource utilization

Cost-Benefit Analysis

| Strategy      | Cost/Month | Expected Impact    | ROI | Priority |
| ------------- | ---------- | ------------------ | --- | -------- |
| DB Indexes    | $0         | 80% faster queries | ∞   | HIGH     |
| Redis Cache   | $50        | 60% less DB load   | 12x | HIGH     |
| Read Replicas | $300       | 3x capacity        | 10x | MEDIUM   |
| Load Balancer | $400       | 3x throughput      | 7x  | MEDIUM   |
| DB Sharding   | $1,200     | 4x capacity        | 3x  | LOW      |

Best Practices

1. Measure first: Don't optimize blindly
2. Low-hanging fruit: Start with easy wins
3. Load test: Validate before production
4. Monitor continuously: Set up alerts
5. Plan ahead: Scale before hitting limits
6. Cost-conscious: ROI-driven decisions
7. Incremental: Small, safe changes

Output Checklist

• Current system profile
• Bottlenecks identified and measured
• Scaling strategies ordered by effort
• Triggers defined for each action
• Phased plan (1x → 10x → 100x)
• Cost estimates per phase
• Load testing plan
• Monitoring dashboard
• Rollback procedures