ClickHouse CloudでExecutable UDFがパブリックベータとして利用可能になりました

Executable User-Defined Functions (UDFs) are now available in public beta on ClickHouse Cloud, allowing users to write functions in Python and upload them as a zip file to be invoked from SQL, similar to built-in functions. This approach streams row data to sandbox processes managed by ClickHouse, supporting real-time anomaly detection, among other use cases. While executable UDFs have existed in self-hosted ClickHouse, the new cloud-based implementation allows users to deploy models without managing their own servers. The deployment process is simplified to a single upload in the Cloud console, where models are executed within a managed sandbox. A demonstration involves using a PyTorch autoencoder to score anomalies in stock trading ticks, with the results embedded into a web app using Next.js. This innovation addresses challenges in co-locating data and machine learning models, reducing the need for separate scoring services or complex integration architectures. The system allows inline execution of models with SQL queries, supporting materialized views and enabling real-time data processing. Additionally, the introduction of network-accessible UDFs in private beta enables external data fetching for enhanced context around detected anomalies, showcasing ClickHouse Cloud's potential for integrating machine learning within SQL workflows seamlessly.