Evolution of Architecture and High‑Availability Design: Insights from Tencent Cloud Experts

This article records a Cloud+ community salon where three senior engineers – Ma Wenshuang (Tencent Cloud Block Storage), Wang Chao (Beike Platform), and Wang Xiaobo (Tongcheng Yilong) – share their views on software architecture evolution, high‑availability practices, and career development for architects.

Architecture Evolution : Wang Xiaobo divides architecture history into three stages – monolithic (circa 2006), distributed/high‑concurrency (mid‑2010s), and data‑driven architectures powered by AI and machine learning. He emphasizes that each stage reflects changing business needs and user expectations.

Cloud Disk Case Study : Ma Wenshuang recounts the design of Tencent Cloud Disk (2013‑present). The first generation combined TFS, TSSD, and CKV but suffered from reliability issues because it was built on existing components without a user‑centric design. The second generation simplified the stack, unified sector size, and achieved higher availability. The third generation removed the access layer, introduced mixed SSD/HDD storage, and later adopted SPDK and RDMA to reduce latency.

High‑Availability Design : Ma explains that availability should be measured from the request perspective (successful requests / total requests) rather than raw downtime. He warns about “sub‑healthy” servers whose performance degrades silently and stresses the need for full‑link monitoring, rapid isolation of faulty nodes, and capacity planning (e.g., load balancers, QPS budgeting).

Project Architecture Control : Wang Chao advises a systematic approach when taking over a new project: map code modules, middleware, and third‑party services; identify risk points; design solutions before implementation; and avoid over‑engineering by matching architecture complexity to actual business requirements.

Improving Architecture Skills : The speakers recommend hands‑on practice—building personal labs, experimenting with Kafka, load balancers, and cloud services—combined with studying books, open‑source projects, and community resources. They also highlight the importance of proactive learning, setting concrete goals, and sharing work through open‑source contributions.

Q&A Highlights : Topics include incident response (damage‑limiting and rapid root‑cause analysis), load‑balancer high‑availability via BGP routing and cluster migration, data‑masking for privacy, and the future of architecture (continual evolution but core principles remain stable). The session ends with recruitment invitations for engineers, architects, and product managers.