How Kaola Achieved Rapid Cloud‑Native Migration: Strategies, Challenges, and Lessons

Background

Kaola’s cloud‑native migration started in October 2019 with a four‑month deadline. The goal was rapid migration, making cloud‑native the optimal path.

Cloud‑Native Definition

Cloud‑native is a technology stack and methodology that embraces containers, continuous delivery, orchestration, open‑source, and micro‑service principles. It requires changes in architecture, development, and operations to produce portable applications for public, private, or hybrid clouds.

Alibaba Cloud provides native middleware such as RocketMQ, Kafka, ARMS, MSE, AHAS, PTS, and Function Compute, which form the foundation for Kaola’s migration.

Migration Phases

Phase 1 (Oct 2019 – Mar 2020) : Integrated databases, Redis, and ASI with minimal code changes.

Phase 2 : Adopted MSE for one‑stop micro‑service governance, eliminating frequent crashes.

Phase 3 : Aligned with group platforms to support major sales events (e.g., Double 11).

Access Strategy

Permission Scheme

Initially a single RAM sub‑account was shared by all developers for simplicity, resulting in coarse‑grained permissions. To tighten security, Kaola wrapped RAM‑based STS tokens with additional policies per user and integrated the tokens with SchedulerX via RoleSessionName to enforce per‑application access.

Message Scheme

The existing Kafka and RabbitMQ ecosystem was migrated to RocketMQ. RocketMQ offered drop‑in compatibility, higher performance, message tracing, and query capabilities.

RPC Scheme

Kaola replaced its custom Dubbo branch (Dubbok) and Nvwa registry with the group’s HSF + ConfigServer stack, using Alibaba Cloud EDAS extensions for Dubbo compatibility. After a month of functional testing and performance tuning, the new stack handled Double 11 traffic without issues.

SchedulerX Scheme

To migrate >13 000 scheduled tasks from the legacy kschedule platform, Kaola built a synchronization tool and a cloud‑native control platform to transfer task definitions and permissions, automating the migration and reducing manual effort.

Environment Isolation

Kaola used a “main‑branch + project‑environment” routing strategy, later mirrored by Alibaba Cloud SCM routing. SCM plugins were ported to Dubbok, enabling seamless environment switching without code changes.

High‑Availability Components

AHAS replaced the legacy NFC component, providing annotation, API, filter integrations, and JavaAgent‑based injection without code changes. AHAS added system‑load protection, circuit breaking, real‑time monitoring, and DingTalk alerts.

Infrastructure as Code (IaC)

Key Practices

Build‑Deploy System : Integrated AppStack and IaC with GitOps; all build, deployment, and static configuration are stored in source repositories for version consistency.

Lightweight Containers : Refactored container images to match group standards, separating admin and application containers and adopting lightweight containers for better isolation.

CPU‑Share Mode : Switched from CPU‑set (fixed CPU binding) to CPU‑share, allowing containers to use any CPU within the same NUMA node, improving stability and utilization.

Image‑Config Separation : Decoupled container images from static and release configurations, enabling image reuse across environments and simplifying rollbacks.

Implementation Strategy

Automation : Used service.cue templates to generate configuration and environment definitions, reducing migration time to ~1 minute per application.

Support Model : Early phases employed a dedicated on‑site support team with daily training; later phases shifted to a self‑service model with API‑driven pipeline creation.

Results

No major incidents during peak events (618, Double 11) despite extensive migration.

Full alignment with group deployment standards, achieving seamless inter‑service communication.

Stateful container deployments saved ~100 seconds per batch; CPU‑share reduced CPU usage to <55 % during peak load.

Decommissioned 250 servers; capacity scaling for major events completed in <0.5 person‑day.

Enhanced cloud product capabilities and resolved security/accounting issues.

Future Direction

Kaola plans to explore Service Mesh to further abstract distributed system complexities, accepting the associated performance overhead to achieve language‑agnostic, infrastructure‑transparent services.