Case Study: Weibo achieves real-time processing and online machine learning at scale with Ververica Platform

Real-time processing and online Machine Learning at Weibo with Apache Flink & Ververica Platform on Alibaba Cloud

Weibo, China’s largest microblogging platform, faced the challenge of running real-time data processing and online machine learning at massive scale while reducing maintenance overhead from multiple disparate processing engines. To address high throughput and strict latency needs across online and offline workflows, Weibo adopted Apache Flink together with Ververica Platform (deployed on Alibaba Cloud Real Time Compute) to modernize and consolidate its ML pipelines.

Using Ververica Platform and Apache Flink, Weibo unified sample generation, multi‑stream joins, stateful processing (RocksDB), and online model training into a single pipeline integrated with its WeiLearn and WeiPS services. The Ververica-enabled solution scaled to hundreds of instances and petabytes of data, supported ~100 billion model parameters and ~950k–1M QPS for model serving, reduced iteration cycles from weeks to about 10 minutes, and simplified operations and development across batch and stream workloads.

Weibo