How DevOps Transforms Operations: Building High‑Efficiency Teams at Meitu

1. Meitu's DevOps Practice

We first look at the traditional product delivery model, where development, testing, and release involve multiple departments and security teams from the early stages, making the process complex when DevOps is introduced.

1.1 Characteristics of the Traditional Model

Non‑automation : Departments are siloed, standards differ, and automation across the chain is difficult.

Slow iteration : Without automation, delivery cycles are long.

Poor environment consistency : Teams use different deployment standards. Meitu built a container platform, standardizing delivery via container images, similar to the DCP platform introduced at Sina.

The traditional model suffers from fragmented goals, duplicated effort, and high resource costs, especially in large teams.

1.2 What DevOps Can Solve

DevOps addresses two dimensions: organizational collaboration and the entire software delivery lifecycle, enabling faster, more efficient product delivery.

Key influences include thought leaders such as Chen Zheng‑Wei and the GOPS community, which provide standards for DevOps adoption.

2. Building an Efficient Team

We improve team efficiency from three aspects: organization, agile teams, and engineer culture.

2.1 Organizational Structure

We examined various structures, from clear hierarchical models to BG+BU models like Tencent. The goal is to flatten the organization, align goals, and create virtual teams that report both to their functional group and to project owners.

Virtual teams are formed with PMO involvement, allowing cross‑functional members to work together on projects such as containerization, sharing outcomes and costs.

2.2 Agile Teams

Products follow an agile workflow managed with Kanban. Tools include Confluence for cross‑department collaboration and lightweight tools like Slack and Trello for project‑level tracking, which significantly improve delivery speed.

2.3 Engineer Culture

Beyond efficient operations, we foster an engineer culture that encourages sharing, internal salons, and recognition. We introduced a "fault culture" with detailed fault‑management procedures, quantitative evaluation metrics, and post‑mortem analysis to continuously improve reliability.

3. Continuous Integration Practice

Meitu's product portfolio spans mobile apps, hardware, e‑commerce, and PC software. To handle diverse workloads, we split services across physical machines, containers, and public clouds.

Physical‑machine pipelines use Jenkins, while container pipelines use GitLab CI. We define five environments (Dev, Dev‑Common, Pre, Beta, Release) and automate promotion across them.

Containerized workloads run on a private K8s cloud built by the Meitu Cloud division, exposing unified APIs for deployment and monitoring.

4. Monitoring System Construction

Given the variety of environments, we built an integrated monitoring platform with layered metric collection, storage, and analysis. The system supports unified alerting, fault detection, and visualization via Grafana.

Alert aggregation is handled by a unified alert platform (UAP) that routes alerts to appropriate teams through enterprise WeChat, supporting convergence rules, severity classification, and visual dashboards.

Note: This article is compiled from the GOPS Global Operations Conference 2018 – Shanghai session.