How Tencent’s TDSQL‑C Achieves PB‑Scale Storage and Serverless Elasticity

Traditional Database Architecture Bottlenecks

Before developing TDSQL‑C, Tencent built several cloud database products and discovered limitations such as local storage capacity that cannot keep up with business growth, single‑master multi‑replica backup latency, and difficulty achieving horizontal scalability.

Design Goals for a Cloud‑Native Database

The team required automatic capacity expansion, zero master‑replica delay, and responsive horizontal scaling, which led to the design of the cloud‑native database TDSQL‑C.

Four Distinctive Features of TDSQL‑C

PB‑level storage : Directly connects to network storage, supporting petabyte‑scale data.

Second‑level elastic scaling : Decoupled compute and storage enable automatic expansion and contraction with minimal latency.

Second‑level snapshot backup and point‑in‑time recovery : Continuous snapshots of data and redo logs provide rapid restore to any point in time.

Serverless, pay‑as‑you‑go model : Charges only for actual compute and storage usage, with second‑level monitoring.

Key Kernel Optimizations

Ultra‑fast backup and restore : Parallel backup and per‑cell redo‑log apply reduce backup time to one‑tenth of competitors.

Accelerated index creation : Parallel scanning, parallel merge‑sort, and batch B+‑tree construction shorten index build time.

Rapid instance start/stop : Parallelized buffer‑pool and transaction system initialization cut startup/shutdown latency by an order of magnitude.

Extreme elasticity : On‑demand log‑driven expansion and compute node scaling achieve second‑level scaling.

Instant column modification : Supports “instant modify column” for second‑level column changes without blocking reads or writes.

Secondary cache : Introduces a cache independent of the buffer pool, leveraging non‑volatile memory to mitigate storage I/O bottlenecks.

Future Directions

TDSQL‑C will focus on ultra‑high write performance through a faster log store and on cross‑region read services that provide higher availability, financial‑grade reliability, and disaster‑recovery capabilities.