Every data lake has the same problem underneath. You store Parquet files in cloud object storage. Cheap. Scalable. Open. That part works fine. Then a pipeline fails halfway through a write. You end up

Markdown supported · ⌘↵ to post
No responses yet.