Baidu Zhidao Cloud Migration Practice: From Legacy OXP to Cloud-Native Architecture

This article details Baidu Zhidao's comprehensive cloud migration journey, addressing the challenges of migrating an 18-year-old Q&A platform to cloud-native infrastructure. The migration was driven by the deprecation of Baidu's internal PaaS platform (ORP) and the need for improved resource elasticity, cost optimization, and enhanced stability.

Key Challenges:

1. Legacy system complexity: Multiple business scenarios, outdated architecture, inconsistent code styles, and high refactoring costs

2. Rapid business iteration: 780+ business requirements annually while maintaining zero-downtime migration

3. High traffic and stability requirements: Over 100 million daily PV with 99.99% SLA targets

4. Architecture evolution: Achieving multi-region disaster recovery capabilities

Migration Solution:

The team selected Pandora (Baidu's container platform) as the underlying infrastructure and Zhiyun platform for deployment and operations. This choice was based on Pandora's ability to support large-scale module deployments (up to 2K modules) without requiring significant code restructuring, making it suitable for the existing ODP monolithic architecture.

Traffic Switching Implementation:

The migration utilized Lua scripts in the access layer for traffic splitting, implementing various strategies from 1% to 100% traffic migration. The solution included business layer modifications to capture traffic markers and enable A/B testing between cloud and legacy environments.

Architecture Evolution Results:

By March 2023, 100% of Zhidao's traffic was successfully migrated to cloud. The platform achieved three consecutive quarters of 99.99% SLA with zero migration-related issues. Core pages now operate across three regions with four data centers (North China:Central China:South China = 4:3:3), providing N+1 cross-region disaster recovery capabilities. End-to-end latency for mini-program core interfaces decreased by 12%.