How Alibaba’s Midway‑FaaS Framework Redefines Serverless Architecture

Community Background

The Serverless community is rapidly growing and has become one of the four major directions of Alibaba’s front‑end committee, offering both opportunities and challenges. Various cloud providers (AWS, Azure, Google, Alibaba Cloud, Tencent Cloud) offer their own function services, each with its own standards and innovations such as Alibaba Cloud’s CustomRuntime and Tencent Cloud’s Service 2.0.

Vendor Lock‑In Prevention

Different cloud providers use different signatures for function entry points. For example, Alibaba Cloud supports

req, res, context

and

event, context, callback

, while Tencent Cloud and AWS have their own variations. The community’s Serverless Framework uses a

serverless.yml

file to define resources for each platform, but plugin support varies across providers.

To avoid vendor lock‑in, we designed a framework that standardizes

serverless.yml

and abstracts the input parameters into a unified

FaaSContext

. The callback is converted to a Promise‑based async function, and the original context is exposed as a property of

FaaSContext

. This allows a consistent function signature:

async function handler(FaaSContext, event?)

.

Flexibility

We aim for a flexible deployment model that lets functions scale horizontally and vertically. Traditional web routes become independent HTTP triggers, each deployed as a separate function. By aggregating functions into a single instance when appropriate, we reduce cold‑start overhead and lower costs.

Our configuration‑only approach lets developers adjust deployment density (high‑density deployment) without changing code, supporting both individual and aggregated function deployments.

Development Efficiency

With growing team sizes, standardization, extensibility, and maintainability become critical. By introducing TypeScript interfaces such as

FaaSContext

, we provide strong typing and IDE support. The Midway‑FaaS framework, built on TypeScript, offers IoC, decorators, and other modern features while remaining compatible with existing FaaS workflows.

Runtime Extension

We defined a lifecycle with four stages—

RuntimeStart

,

FunctionStart

,

Invoke

,

Close

—each offering before/after hooks. The lightweight

LightRuntime

implements this lifecycle across platforms, while a

runtime-engine

manages execution.

Additionally, we introduced a

Layer

concept (inspired by AWS Lambda) that can be attached to any runtime to provide cross‑scenario capabilities such as unified monitoring, metrics, and migration support.

After nearly a year of iteration, the framework has been used to migrate Alibaba’s Taobao guide business and other BU projects to a Serverless model, successfully passing large‑scale promotional events.

We plan to open the Midway‑FaaS capabilities to the community; the code is already on GitHub and is in public beta, with a v1.0 release expected next January.