Inside Tencent’s DevOps Pipeline: How Continuous Delivery Powers Scalable Operations

Continuous delivery is the core engineering practice of DevOps; the delivery pipeline drives DevOps implementation. At the Cloud+ Summit, Tencent unveiled its DevOps pipeline platform and demonstrated how ZhiYun achieves high‑standard continuous deployment.

Introduction

The national “Internet+” strategy has created a new business model that blends enterprise services with the Internet, prompting many companies to deliver their products online. Every Internet product goes through planning, development, testing, and operations, forming a full IT value chain that ultimately creates commercial value.

Tencent's DevOps Practice

In DevOps, faster flow of the IT value chain means stronger product delivery capability and a competitive edge for the enterprise.

Tencent, a leading social network with massive user and device scale, structures its DevOps practice around four platform systems.

These four systems together form the DevOps pipeline, enabling Tencent’s massive services to smoothly progress from requirement design, code management, development testing, to release and operations.

TAPD supports agile project management, linking product requirements with development branches.

TGit provides code management and triggers continuous integration via webhooks.

CIS automates compilation, testing and other tasks, producing artifacts such as software packages or Docker images.

ZhiYun consumes CIS artifacts and automates business release and change operations.

Operability of Application Architecture

For Internet products, release is only the beginning; ongoing operation must ensure stable, reliable service. Tencent’s practice defines operability through eight continuous‑delivery principles that require development and operations teams to cooperate on functional and non‑functional requirements.

The operability process can be divided into four stages: unified architecture, operation standards, standard operation, and operation automation.

Internet business architecture is abstracted into three layers: access layer, logic layer, and data layer.

When planning technical choices, four principles are followed: framework‑based, component‑based, stateless, and distributed.

Framework reduces development effort; a framework can provide network communication for socket‑based C/S architectures, while business logic is loaded as dynamic libraries. Non‑functional operability requirements such as data reporting, unified logging, and management tools can also be integrated into the framework.

Components unify common services; for example, Tencent’s internal routing service provides load balancing, naming, fault tolerance, overload protection, and traffic scheduling, simplifying daily operation management.

Through unified planning and standardized continuous improvement, Tencent’s architecture evolves from highly customized to a common, scalable model, embedding non‑functional specifications via frameworks and components.

Tencent ZhiYun Continuous Deployment Practice

Long‑term growth requires more than a collection of tools; it demands a systematic, global approach to standardization, configuration, automation, and intelligence. ZhiYun’s functional diagram shows how Tencent manages massive social‑network services.

Effective configuration management is essential. In ZhiYun’s CMDB, each micro‑service cluster is defined as a module with two configuration categories: basic configuration and application configuration.

Basic configuration includes asset configuration (for accounting and budgeting), hardware configuration (for virtualization and device planning), and distribution information (recording upstream switches and IDC locations), supporting intelligent analysis of data center traversal and network failures.

Application configuration covers resource configuration (linking to image or artifact repositories), process configuration (custom orchestration of tools or interfaces), and change records (providing audit and monitoring data).

Daily operational steps are illustrated as a resource transmission and execution flow.

From unified planning, standardization, configuration, automation to linked monitoring, the continuous‑deployment pipeline creates a systematic operation capability model that enables global planning of deployment tools and processes.

Tool orchestration allows custom operation, approval, and service‑request workflows, integrating with CMDB data (business, owner, status) to solve coordination challenges and upgrade from offline ITIL processes to online automated flows.

Using ZhiYun’s automated scaling process, atomic operation tools or system interfaces are organized into optimal workflows, ensuring each step is strictly executed regardless of individual experience or documentation version, thus addressing the “documentation expires, staff leaves” problem.

With a unified operation system, efficiency improves and service quality is better guaranteed. The following self‑healing scenario shows how, by leveraging CMDB business attributes, an automated workflow registers configurations, enabling process monitoring to automatically recover from failures.

Conclusion

From Tencent’s many years of massive‑scale operation experience, DevOps permeates the entire software lifecycle. Release is not the end; a holistic, unified planning approach builds a standardized, orderly business architecture and provides a complete, systematic operation solution for healthy business growth.