How Tencent CloudBase Powers Mini‑Program Development with a High‑Performance Distributed NoSQL Database

Overview

Tencent CloudBase (TCB) provides a serverless, high‑performance NoSQL cloud database for Mini‑programs. It aims to satisfy five core requirements—security, ease of use, low cost, high performance, and schema‑less flexibility—while keeping operational overhead minimal.

Background

Mini‑programs combine lightweight front‑ends with serverless back‑ends, removing the need for developers to provision or manage servers, networks, or traditional databases. This model accelerates development cycles and reduces operational costs.

Core Requirements

Security : Data must be protected against loss and unauthorized access.

Ease of Use : Plug‑and‑play access via SDKs without operational overhead.

Low Cost : Pay‑as‑you‑go pricing with fine‑grained billing per operation ( query, insert/update, delete).

High Performance : NoSQL engine supporting high‑concurrency reads and writes.

Flexibility : Schema‑less design ( no‑schema) with elastic scaling.

Architecture

The system consists of three logical layers:

Mini‑program client.

Access layer (keeper + agent) handling authentication, connection pooling, QPS throttling, and request routing.

Storage layer (distributed NoSQL instances) providing the actual data store.

Supporting modules include control, alarm, backup, and metadata services.

Access Control

• Permission control : Each user can only access their own database. Default role is read‑write; read‑only accounts can be created. Authentication leverages WeChat’s “免鉴权” to hide credential management.

• Connection control : Two‑level limits—client‑side limits based on instance type (free/paid) and backend pool limits via connection pooling to master/replica nodes.

• QPS control : Requests pass through the access layer where rate limits are enforced and excess traffic is queued or rejected.

Data Security

• Replication : Three‑copy distributed storage across different machines and availability zones, using a Raft‑like consensus protocol for strong consistency.

• High availability : Automatic failover swaps a replica to master when a node fails.

• Backup & point‑in‑time recovery : Transparent full and incremental snapshots triggered by write intensity. Users can restore any point within the last 7 days, optionally per table.

• Multi‑AZ deployment : Default deployment spans three AZs; cross‑region deployment is also supported.

Elastic Scaling

Real‑time load monitoring drives dynamic resource adjustment. A sliding‑window algorithm with spike‑filtering smooths scaling actions, handling periodic traffic patterns (e.g., food‑delivery peaks, gaming evenings).

Hot Migration

When an instance changes state (e.g., free → paid) the system migrates data without downtime. The migration follows three phases:

Data Sync – Full data copy followed by continuous operation‑log replication.

Cutover – Source becomes read‑only, the access layer buffers requests, the target catches up, then switches to active.

Status Change – Target cluster is activated and blocked requests are released.

Migration state and timeout are persisted in etcd. Benchmarks show 3 k QPS with 100 % writes and 100 % success rate during migration, confirming user‑transparent hot migration.

Intelligent DBA

An automated operations platform monitors node health, performs automatic failover, and issues alerts. Index optimization runs automatically: when frequent full‑table scans are detected, the system creates appropriate indexes, reducing average query latency by >50 %.

Automatic Index Management

• Asynchronous index creation to avoid blocking user requests.

• Risk control : Limits on index‑creation frequency and total index count prevent CPU overload and write amplification.

• Dynamic index lifecycle : Indexes are removed or updated when query patterns change, respecting the left‑most matching principle.

After enabling automatic indexing, overall query latency dropped by more than half across the platform.

Monitoring & Alerting

The platform collects >69 metrics (node health, QPS, latency, storage usage, etc.) with second‑level granularity, enabling proactive detection of anomalies and automated remediation.

Future Work

Finer‑grained monitoring and richer dashboards for internal use.

More advanced intelligent DBA features (e.g., predictive scaling, workload‑aware index recommendations).

Kubernetes‑based elastic scaling for containerized database instances.

Specialized APIs such as Chinese‑text indexing for quiz‑type Mini‑programs.