How Alibaba’s FaaS Platform Boosts Content Delivery with Serverless Tech

Background

Youku's content distribution business spans homepage, channel pages, and secondary pages, originally built on traditional Java applications combined with Alibaba middleware, which caused slow iteration due to heavy product review, API design, front‑back integration, and release cycles. Rapidly changing content demands and diverse data sources made the classic server‑side architecture inefficient, leading to high development cost and low flexibility.

With the rise of Serverless, the team explored Function‑as‑a‑Service (FaaS) to encapsulate foundational capabilities and let FaaS handle business logic, giving birth to the Youku FaaS platform.

Platform Design and Technical Challenges

Design Goals

The platform aims to provide a generic function computing environment where developers can write, run, and manage one or more functions that expose services via micro‑services, HTTP, or event triggers. Functions should be deployable within seconds without restarting host applications, enabling rapid rollback and reducing non‑coding time.

Key characteristics of functions include short lifecycle, non‑daemon execution, statelessness, reusability of existing services or third‑party resources, and millisecond‑level execution.

Overall Architecture

The core architecture consists of a gateway layer, runtime containers, a one‑stop operation and publishing platform, and underlying basic services.

The gateway receives function calls and routes them to the appropriate cluster based on function identifiers. The function container layer hosts the execution engine, managing instance lifecycle, on‑demand loading, pre‑warming, and unloading.

The one‑stop publishing platform (FaaS Platform) offers developers function creation, version submission, rollback, and monitoring, integrating with Alibaba's centralized monitoring system for real‑time metrics.

The platform builds on Alibaba middleware and Youku's content distribution services, providing isolated execution environments for functions.

Frontend is built with React and Dva, using Ant Financial's Bigfish framework and Odin scaffolding to enable component reuse and rapid development.

The front‑end follows an MVC‑like layered design, with State, View, and Action components, and supports both centralized and decentralized function invocation models.

Main Features

Low Development Cost

The platform offers a one‑stop cloud development environment, abstracting underlying services and middleware so developers focus on business logic. It supports rapid development, middleware and data source integration, and direct invocation of encapsulated capabilities.

Runtime Isolation

Isolation is achieved at two levels: within the function container (each function instance runs in an isolated JVM environment) and across virtual application clusters (different functions are bound to separate clusters, ensuring physical isolation).

Transparent Operations Monitoring

Function execution logs are collected and reported to Alibaba's monitoring services in real time, allowing developers to monitor function health directly from the platform.

Technical Challenges

Function Execution Engine Design

The engine manages loading, pre‑warming, scheduling, execution, and unloading of function instances. Functions are written in Groovy, leveraging the JVM. Each function has its own versioned code repository; the engine loads the latest version, compiles it, and pools instances for reuse.

Lazy loading and automatic eviction prevent memory bloat when many functions reside in the same JVM.

Third‑Party JAR Loading Capability

The platform provides a custom classloader that allows functions to load third‑party JARs without restarting the container, ensuring isolation between functions and their dependencies.

FaaS Platform in Practice

In Youku's content distribution, FaaS is used for data source encapsulation and API template field mapping. Developers can add new data sources or modify field logic by updating Groovy functions, achieving second‑level deployment without touching the underlying Java services.

Summary and Outlook

The platform has improved iteration speed for content distribution, but remains in early stages. Future work includes supporting more programming languages, enhancing cloud IDE experience, optimizing resource scheduling and auto‑scaling, and exploring additional business scenarios to further increase flexibility and efficiency.