Real-time Update of Column Store on Disk

This lecture discusses the challenges of supporting real-time updates on very large column stores on disk, focusing on the implementation of snapshot isolation and reasonable update throughput without compromising query performance. The instructor presents a solution involving in-place updates, isolation levels, and storage structures. The lecture also covers examples of updating tuples, instant updates, and performance evaluations. Furthermore, it explores the implementation of snapshot isolation, including high and low watermarks, and the cleaning process for 'delete intents'. The lecture concludes with the implementation details in Parquet C++ codebase and its integration with query engines.