Tencent is the largest Internet service provider in China. Typical services include WeChat, games, payment, cloud storage and computing.
Tencent serves billions of users and millions of enterprises, and some services like WeChat, are required to have the ability to handle more than 200,000 TPS requests at peak time.
To do this, Tencent has built an elastic and scalable database service system, namely TDSQL, which can efficiently support their ever-growing service requests.
TDSQL is deployed and runs on top of more than ten thousands of compute nodes.
In this talk, we present the main challenges that we have encountered, and give our practice of conceptual modeling on TDSQL.
First, failures of compute nodes often occur in an X86-based large-scale distributed system architecture. To address this issue, we introduce a fault tolerance model to guarantee the high availability of the services.
Second, Tencent serves a huge number of requests, while different types of requests require different storage and compute resources. To improve the resource utilization, we propose a resource scheduling model that enables TDSQL to serve the requests elastically.
Third, TDSQL provides a hybrid data modeling to support various data models, and develops DBaaS services to serve 10,000 + DB instances.
Finally, we present how to fast develop applications in terms of conceptual modeling on top of TDSQL.