Building a Real-Time Data Computing Platform for Tencent Games: Practices and Architecture

The presentation begins with the background of Tencent Games' need for real‑time data processing to improve user experience and revenue, highlighting the limitations of offline data pipelines such as high latency and poor feedback.

It then introduces OneData , a unified real‑time big‑data development platform that standardizes data ingestion, provides SQL‑based and graphical configuration for metrics, offers an online WebIDE for UDF development, and integrates Flink for event‑time processing, asynchronous I/O, and stateful computation.

Next, the OneFun service layer is described, which exposes game data through API‑style function services built on Kubernetes, supporting multiple runtimes (Privileged V8, V8, Go VM, Lua VM) and enabling rapid, secure deployment of data‑service interfaces.

The architecture further evolves into a micro‑service and ServiceMesh‑managed ecosystem, allowing independent development, deployment, and scaling of functional modules, while providing traffic governance, security, and observability across east‑west and north‑south flows.

Operational practices such as standardized development processes, automation of resource allocation and deployment, and a one‑stop configuration system are emphasized to reduce cost, improve efficiency, and achieve over 60% savings in compute and storage for more than 13,000 metrics.

Finally, several real‑world use cases are showcased, including real‑time player reward systems, social recommendation, and large‑scale metric aggregation, demonstrating how the platform supports both online and offline scenarios across hundreds of games.