Tencent Advertising Real-Time Strategy Data Framework: Architecture, Performance, and High Availability

In May 2021, Lin Liwei, the technical lead of Tencent Advertising's real‑time strategy data framework, shared a live recap of the thirteenth China System Architect Conference (SACC2021), describing how the framework supports the entire ad decision pipeline—from recall to final ranking—by calculating various strategy factors.

The framework addresses three core challenges: usability, high performance, and high availability. To improve usability, a unified strategy computation platform was built, abstracting log aggregation, providing flexible strategy interfaces, and supporting multi‑format outputs for developers, testers, operations, and product teams.

Performance optimizations include partial parsing of massive log data, early filtering of irrelevant fields, sampling for non‑critical metrics, shared in‑memory caches, and efficient interaction with external systems such as HBase, TubeMQ, and distributed databases. Scheduling improvements introduce self‑adjusting real‑time load balancing, reducing long‑tail latency across parallel processing tasks.

High‑availability measures consist of master‑slave deployments with consistent results via shared HDFS storage, master‑only writes with seamless failover, checkpoint mechanisms for incremental state recovery, and automated fault‑handling pipelines that enable rapid data catch‑up and hot‑loaded configuration changes.

Overall, the unified framework enables daily strategy iteration, reduces development cycles from weeks to days, and supports Tencent Advertising's multi‑billion‑yuan daily revenue by ensuring fast, reliable, and scalable ad strategy computation.