Database Selection and Service Platform Construction for Large-Scale Multi-Storage Scenarios

This article shares vivo's practices in database selection and service platform construction, presented by Xiao Bo, Database Architect at vivo Internet.

The content is divided into three main parts:

1. The Revolution of Self-Cognition

The author discusses the traditional perception of DBAs as fire-fighters who solve problems, create process specifications, handle architecture design, and manage data governance. The key transformation is to view DBA and DBMS as a product providing database service capabilities. The service targets include developers, testers, architects, and managers. This approach emphasizes DBA's value, changes the relationship between DBA and DBMS from management to unification, and enables DBA to output capabilities through product operation rather than just saying "NO" to users.

2. Demand-Driven Database Selection

Business scenarios are the soul of requirements. The author explains the framework: "Who" in "what environment" does "what/encounters what problems" and "how to solve" to "create what value".

Vivo's business scenarios include: recommendation systems (app store, content feeds, music, video, novels), account systems, advertising, transactions (VIP, themes, cloud services, games), downloads, and push notifications.

Database selection uses a radar chart tool to evaluate databases across five dimensions: data scale, response latency, data model, data security, and scalability. Common database products evaluated include MySQL, Redis, ElasticSearch, MongoDB, TiDB, KV Storage, TSDB, and GDB.

3. Service-Driven Platform Construction

Users of the database service platform include: software developers, test engineers, DBAs, and product managers. The platform serves four environments: development, testing, staging, and production (including IDC and public cloud).

The database product lifecycle includes: service application, service deployment, maintenance, data operations, service optimization, data recovery, and decommissioning.

Product design principles: (1) Don't lead product design from DBA perspective - even beginners can use databases; (2) Focus on user experience, balancing reliability, usability, and efficiency; (3) Machine learning and chaos engineering are just implementation methods, not the goal.

Platform architecture includes: abstracting common services (work orders, permissions), using AGENT for metadata/monitoring/inspection tasks, building shared capabilities, and unifying user interfaces to mask differences between public cloud and IDC.

Multi-environment management addresses network isolation between development/test and production environments through unified permission and access portal, environment isolation, and unified metadata management.