I tried "select * from items limit 10" and it is slowly iterating through the shards without returning. I got up to 60 shards before I stopped. Selecting just one shard makes that query return instantly. As mentioned elsewhere I think duckdb can work faster by only reading the part of a parquet file it needs over http.

I was getting an error that the users and user_domains tables aren't available, but you just need to change the shard filter to the user stats shard.