Though a little sceptical of embedding DuckDB. It is easy and better to isolate Read/Write paths, and it has a lot of other benefits.
We actually separate Read/Write paths. BemiDB reads by levering DuckDB as a query engine. And it writes to Iceberg completely separately from DuckDB. I'm curious if that's what you imagined.
In the data lake, I would have a gateway provisioning DuckDB on demand for each user.
Though a little sceptical of embedding DuckDB. It is easy and better to isolate Read/Write paths, and it has a lot of other benefits.