Cloud‑Native Evolution of NetEase Yanxuan: Practices, Phases, and Lessons Learned

At the beginning of 2019 Yanxuan’s services approached a thousand, and the system grew rapidly. To support this growth, the Yanxuan team and NetEase Qingzhou launched a joint cloud‑native platform project, marking the start of Yanxuan’s cloud‑native migration.

Background : Cloud‑native aims to make applications more agile and efficient in complex, changing business scenarios. Yanxuan’s goal was to simplify the system in three ways: (1) evolve the service mesh to push middleware down to the infrastructure layer, (2) containerize applications to achieve immutable infrastructure, and (3) adopt cloud‑based DevOps to reduce the complexity of application lifecycle management.

Key Considerations :

Standardization : repay technical debt and lay the groundwork for immutable infrastructure.

Reduce source changes : lower migration cost and avoid bugs introduced by migration.

High‑availability migration : ensure reliable online services, using gray‑release, version routing, and hybrid‑cloud edge gateway solutions.

Performance tuning : address performance challenges from service mesh and multi‑region deployments, e.g., introduce SR‑IOV container networking and optimize Istio configurations.

Practice Phases :

Phase 1 – “Lay the foundation” : Build a DevOps pipeline, introduce CMDB, establish a unified artifact‑centric deployment model (Opera), set up a real‑time log platform, and adopt Istio as the service‑mesh implementation.

Phase 2 – “Build the framework” : Define engineering guidelines (base‑image specs, CI standards), deployment verification processes, and API‑gateway traffic control (Envoy‑based).

Phase 3 – “Build the pipeline” : Cloud‑ify backing services (DB, cache, MQ), deepen service‑mesh capabilities (hot upgrades, gray releases, config slimming, SR‑IOV nodes), and strengthen the SNest service portal.

Phase 4 – “Build the rooms” : Construct pure cloud environments for testing and a “regression” environment that simulates heterogeneous on‑premise and cloud resources, enabling large‑scale migration validation.

Key Takeaways :

IP‑address reliance required the introduction of IPRange mechanisms and a shift to service‑mesh‑based routing.

Observability became critical as infrastructure complexity grew; logs, alerts, and Kubernetes events were integrated into a unified monitoring system.

Team impact: developers gained faster scaling, infra engineers focused on non‑intrusive middleware, QA ensured CI/CD compliance, and ops improved fine‑grained resource control.

Future Plans : Continue expanding the regression environment, leverage service‑mesh features to create version‑based test environments, and support performance‑sensitive applications for upcoming high‑traffic events.

Recruitment Notice : Yanxuan’s core technology team is hiring senior Java engineers, senior network engineers, and senior test developers to further advance business‑mid‑platform, service‑mesh, and DevOps capabilities.