Bản chất Web Framework

Mở đầu

Backend hosting evolution: physical server → monolith → containerization → microservice → serverless. Mỗi stage giải pain trước, kèm new complexity. Chương này dẫn full journey.

---

1. Sao hiểu architecture evolution?

• Hiểu stack hiện tại từ đâu ra
• Tránh over-engineering (dùng K8s cho app 10 user)
• Choose tech đúng cho project + scale

---

2. Physical server era (1990s)

1 web = 1 physical server tự rack
- Mua server hardware ($5k-50k)
- Cài Linux/Windows
- Manual setup web server (Apache)
- Cài app
- Maintain — nâng cấp, security patch, backup

Pain:

• Provision lâu (vài ngày-tuần)
• Idle resource (server chạy 10%)
• Scale = mua thêm server
• DR khó

---

3. Monolith era (2000s)

App + DB + cache trong 1 server, 1 codebase, 1 deploy.

┌──────────────────────────┐
│      Single App          │
│  ┌────┐ ┌────┐ ┌────┐    │
│  │User│ │Order│ │Pay │   │  ← Module chung memory
│  └────┘ └────┘ └────┘    │
└──────────┬───────────────┘
           │
       ┌───▼───┐
       │  DB   │
       └───────┘

Lợi: đơn giản, 1 deploy, transaction qua DB rõ. Hại:

• 1 module crash → cả app down
• Scale toàn app, không chỉ module hot
• Big repo, multi-team conflict
• Tech stack lock cứng

---

4. Containerization + Microservice (2010s)

4.1 Docker xuất hiện (2013)

FROM node:20-alpine
WORKDIR /app
COPY package.json ./
RUN npm install
COPY . .
CMD ["node", "server.js"]

docker build → image bất biến → chạy mọi nơi (dev, staging, prod giống nhau).

Giải pain:

• "Works on my machine" → image consistent
• Provision giây (vs hour)
• Resource isolation
• Tech stack mix-and-match

4.2 Microservice break monolith

                  ┌────────────┐
                  │  Gateway   │
                  └─────┬──────┘
        ┌───────────┬───┴───┬──────────┐
        ▼           ▼       ▼          ▼
   ┌────────┐  ┌──────┐ ┌──────┐  ┌──────┐
   │  User  │  │Order │ │ Pay  │  │ Cart │
   │ Service│  │ Svc  │ │ Svc  │  │ Svc  │
   └───┬────┘  └──┬───┘ └──┬───┘  └──┬───┘
       │           │        │          │
       ▼           ▼        ▼          ▼
    ┌────┐      ┌────┐   ┌────┐     ┌────┐
    │ DB │      │ DB │   │ DB │     │ DB │
    └────┘      └────┘   └────┘     └────┘

Lợi:

• Mỗi service team riêng, deploy riêng
• Tech stack tự chọn (Go service cao tải, Python service AI)
• Scale riêng từng service
• Fault isolation

Hại:

• Distributed system complexity (network failure, eventual consistency)
• Cần ops mạnh (service discovery, monitoring, tracing)
• Debug khó (cross-service)
• Cost ops cao hơn

4.3 Kubernetes (2014)

Orchestrate hàng trăm container:

• Auto scaling
• Self-healing (restart container chết)
• Load balancing
• Rolling deploy

apiVersion: apps/v1
kind: Deployment
metadata:
  name: user-service
spec:
  replicas: 3
  template:
    spec:
      containers:
        - name: user
          image: myapp/user-service:1.0
          resources:
            limits:
              memory: "512Mi"
              cpu: "500m"

---

5. Serverless + Cloud-native era (2020s+)

5.1 Serverless (FaaS)

// AWS Lambda / Vercel / Cloudflare Workers
export async function handler(event) {
  const user = await db.user(event.userId)
  return { statusCode: 200, body: JSON.stringify(user) }
}

Lợi:

• No server management
• Pay per request (idle = $0)
• Auto scale 0 → infinity
• Deploy giây

Hại:

• Cold start (lần đầu chậm 100ms-1s)
• Vendor lock-in
• Time limit (vd 15 min AWS Lambda)
• Long-running connection khó (WebSocket)

5.2 BaaS (Backend as a Service)

Supabase, Firebase, AppWrite, PocketBase — backend ready-to-use:

• Auth
• DB + realtime sync
• Storage
• Edge function

→ FE dev có thể build full app không cần BE team.

5.3 Edge computing

Cloudflare Workers, Vercel Edge, AWS Lambda@Edge — chạy code ở edge node gần user (50ms latency global).

---

6. Comparison + selection

Era	Use	Pros	Cons	Cost
Physical	Legacy data center	Full control	Manual, slow scale	High initial
Monolith	MVP, internal tool	Simple	Single point of failure	Medium
Container + Monolith	Mature SaaS	Reproducible, easy deploy	Need ops skill	Medium
Microservice + K8s	Large enterprise	Scale, fault isolation	Complexity cao	High
Serverless	Event-driven, sporadic traffic	Zero ops, scale 0 → ∞	Cold start, vendor lock	Low (pay-per-use)
BaaS	MVP, no backend team	Cực nhanh ship	Limited customization	Varies

Decision flow

< 1k user, MVP? → Monolith on Vercel/Railway/Fly.io
+ Need realtime, auth, storage? → Supabase / Firebase BaaS
1k-100k user, growing? → Containerized monolith on managed K8s (GKE, EKS)
> 100k user, multi-team? → Microservice on K8s
Event-driven, sporadic? → Serverless (Lambda, Cloudflare Workers)
Global low-latency? → Edge compute

---

7. Tổng kết + roadmap

Stage	Skill
Junior	Monolith framework (Express, Django, Spring Boot)
Mid	Docker + CI/CD + 1 cloud provider
Senior	Microservice patterns + K8s + observability
Architect	Event-driven, eventual consistency, multi-region, cost optimization

2026 cho VN dev

• Serverless platform trendy: Vercel, Cloudflare Workers, Fly.io, Railway
• VN cloud: VNG Cloud (CloudVerse), Viettel Cloud, FPT Cloud — chú ý compliance (data residency)
• Hybrid: BE serverless (Vercel) + DB managed (Supabase, Neon, PlanetScale)
• AI-native architecture: API gateway layer (LiteLLM, Portkey) cho LLM cost + routing
• VN context: startup → Vercel + Supabase; enterprise → AWS/GCP với K8s + Spring Boot

8. Glossary

Term	Note
Monolith	1 codebase, 1 deploy
Microservice	Service nhỏ độc lập, communicate qua API/MQ
Container	Lightweight VM, image bất biến
Orchestration	Quản lý container scale (K8s, Nomad)
Serverless / FaaS	Function as a Service, no server mgmt
BaaS	Backend as a Service
Cold start	Lần đầu invoke serverless function, chậm
Edge compute	Chạy gần user, low latency
IaC	Infrastructure as Code (Terraform)
GitOps	Git là source of truth cho infra + deploy