How We Migrated a High‑Traffic Video Service to Node FaaS in 4 Months

Background

Wow Video is a short‑video shopping feature on the Taobao homepage, aiming to create warm, user‑centric videos, encourage merchant‑generated content, and improve homepage distribution efficiency.

Goal: Build short videos centered on "people using items" with a personal touch, drive more merchant videos, and make it easy for users to find the content they want.

The service faces three high‑traffic characteristics: massive traffic, high stability requirements, and rapid iteration cycles.

Taobao Shopping Guide R&D Model

Middle‑Platformization

Most capabilities are provided by the middle platform, allowing rapid assembly of features and reducing duplicate development.

Workflow

Typical development involves a front‑end and a back‑end working together through development, integration, and testing.

Analysis of recent tasks shows back‑end effort is lower due to middle‑platform reuse, while front‑end effort is higher because UI changes often require rebuilding.

R&D Pain Points

High Integration Cost

Integration between front‑end and back‑end consumes about 30% of development effort, leading to fatigue and slower iteration.

Front‑End Resource‑Lock

Front‑end developers focus on UI and interaction, limiting their exposure to core business logic and hindering deeper skill growth.

Encountering FaaS

In April, the author joined the Node FaaS initiative and selected Wow Video as the first pilot because it has many slots, strong algorithmic recommendation, multiple data sources, and a stateless service model—making it well‑suited for FaaS.

Analysis: high slot count + algorithm dependence = moderate complexity; many data sources = extensive upstream integration; statelessness = ideal for horizontal scaling on FaaS.

Migration in Progress

Three measures were taken to ensure a smooth migration without affecting the business:

Quickly copy existing Java code to the new platform.

Use Java as a fallback to guarantee stability.

Carefully evaluate subsequent requirements for feasibility.

Copy Java Code

Rather than rewriting from scratch, the team copied ~5,000 lines of well‑documented Java code, preserving error handling and boundary logic, resulting in a zero‑bug release.

Use Java as Fallback

The original architecture used a CDN fallback for failed requests. The new flow added a Java fallback between the FaaS endpoint and the CDN, ensuring user experience remains intact even when the FaaS layer is unstable.

Previous: Java API → CDN fallback Now: FaaS API → Java API → CDN fallback

Post‑Migration

The rollout proceeded smoothly with gradual gray‑release and scaling. However, taking over the backend introduced new challenges, requiring the author to learn deeper business logic, middleware, databases, and operations.

Learning Curve

By actively seeking knowledge from backend teammates and dissecting requirements, the author transitioned from “how do I do this?” to confidently designing solutions, illustrating personal growth beyond pure technical skills.

Conclusion

Serverless and FaaS are not just programming models; they provide opportunities for developers to own business logic and accelerate skill upgrades. The migration proved that FaaS can be applied in high‑traffic scenarios and helps front‑end engineers broaden both technical and business perspectives.