Storage Governance and Optimization Practices for Meeting Control Systems

This article details the storage governance and optimization practices implemented for a meeting control (会控) system to address challenges posed by high concurrent user traffic, disaster recovery requirements, and storage limitations.

The system initially faced significant storage pressure due to special period PCU surges, with all backend storage relying on Redis. Key challenges included shared storage instances for personal and enterprise meetings causing interference, lack of storage layer isolation despite logical layer multi-SET isolation, Redis instance reaching product limits, high availability requirements, and difficulties meeting 100k QPS entry demands.

The storage governance initiative established five core objectives: disaster recovery/multi-active deployment, data isolation, business transformation, parallel expansion, and storage consolidation. Implementation involved creating a storage proxy layer to consolidate all storage access, implementing policy sinking for unified control, applying dynamic-static separation to reduce hot key pressure, introducing multi-level caching with fastcache, adding overload protection mechanisms, establishing comprehensive monitoring, and optimizing performance through techniques like object pooling and lock-free queues.

For disaster recovery, the solution deployed cross-region active-active architecture between Guangzhou and Shanghai, with manual traffic switching mechanisms and detailed failover procedures. Personal and enterprise data isolation was achieved through identity-based storage separation, while meeting ID routing encoding enabled efficient request routing by embedding region and identity information directly in the meeting ID uint64.

The sharding (分库) approach adopted a dual-write plus data migration strategy to ensure consistency during transition, with real-time and delayed reconciliation processes to verify data consistency between old and new instances. The article concludes with lessons learned, risk analysis, and future platformization aspirations for the storage sharding framework.