How Chanjet Scaled SaaS for 1.3M SMEs with Cloud‑Native Architecture

Background

Chanjet, a SaaS provider for Chinese micro‑enterprises, originally allocated a dedicated virtual machine to each tenant. This caused low resource utilization, high O&M cost, and limited elasticity.

Challenges of the Legacy Model

Isolated VMs for each tenant → poor utilization.

Multiple software versions required separate patch pipelines.

Planned maintenance caused brief service outages.

Unable to scale quickly during traffic spikes.

Cloud‑Native Design Goals

Adopt a cloud‑native architecture that shares compute and storage, uses containerization, and follows micro‑service principles to lower cost, enable rapid elastic scaling, and simplify product iteration.

Architecture Overview

Front‑end presentation layer : micro‑frontend framework built with qiankun for H5 applications.

Business middle‑platform : micro‑services on Alibaba Cloud EDAS, GraphQL data model, RocketMQ for decoupling.

Technical middle‑platform : container management, DevOps pipeline, service governance, tracing.

Operations middle‑platform : unified monitoring, alerting, self‑healing.

Infrastructure layer : PolarDB with tenant‑level sharding, read/write splitting, middleware abstraction.

Micro‑service Refactoring

Business domains were split into four layers—core, business, application, and interface—each implemented as an independent Spring Cloud micro‑service managed by EDAS. Service registration, discovery, and governance are handled by Spring Cloud; EDAS integrates with Alibaba Cloud Kubernetes (ACK) for elastic scaling based on CPU, response time, or custom metrics.

Data Consistency Strategy

Strong transactional consistency was replaced by eventual consistency across distributed services. High‑frequency operations use the TCC (Try‑Confirm‑Cancel) pattern; less time‑critical processes rely on asynchronous message queues to achieve eventual consistency.

Container Management & DevOps

All core applications run in Docker containers orchestrated by Kubernetes. A GitOps pipeline built on GitLab, Jenkins, Rundeck, and K8s automates build, test, and deployment, guaranteeing identical environments from development to production.

Key pipeline steps:

Source code stored in GitLab.

Jenkins builds Docker images.

Rundeck triggers deployments to ACK clusters.

Environment variables and a central configuration center ensure consistent settings across stages.

Database Migration and Scaling

Initial MySQL instances lacked read/write separation. After evaluating Sharding‑JDBC, Chanjet migrated to PolarDB, which provides native read/write splitting, automatic scaling, and parallel query capabilities. The migration introduced a one‑write‑multiple‑read topology and reduced query latency by 20‑40 %.

Micro‑frontend Strategy

A lightweight base application and shared reusable components were built. Each sub‑application follows a common protocol for loading, communication, and build, enabling independent deployment and reducing coupling.

Observability and Full‑Link Tracing

Integration with Alibaba Cloud ARMS and a custom business‑timeline model provides end‑to‑end tracing. Business operations are mapped to REST API request IDs; a WebFilter captures request/response details and correlates them with ARMS trace IDs, cutting fault‑diagnosis time by more than 50 %.

Gray‑Release (Canary) Mechanism

Gray releases are applied at the tenant level. Separate database and application instances are provisioned for the gray group with strict compatibility rules (add‑only schema changes, isolated enum values). Routing is performed via Nginx Lua scripts that inspect tenant IDs.

Business and Technical Benefits

Elastic scaling supports sudden traffic spikes (e.g., live streaming, flash sales).

Read/write splitting and PolarDB improve query performance by 20‑40 %.

DevOps automation reduces deployment time by 4× and personnel effort by 50 %.

Resource consolidation saves approximately 1,900 ECS/POD instances.

Observability delivers a 5‑nine SLA, enhancing user satisfaction.