I can't wait to throw out all the columnar databases and the ETLs I have to support them, all just for a few queries.
I can't wait to throw out all the columnar databases and the ETLs I have to support them, all just for a few queries.
As a general rule of programming, I never want syntax sugaring or a chosen expression pattern to impact runtime performance. I have found it's optimal to write expressive code first and performant code second - and if there's a trivial expression transformation between the two a compiler or interpreter better be equipped to do it for you.
If you know the query plan you want, you can usually force it using CTEs.
Really I'd be happier without a query planner at all.
I think being happy not having any query planner at all is a bit extreme, but there are times when queries need to be tuned to force early data culling or adherence to indexes, I find these times to ultimately be pretty rare and generally have to review all the instances of our manual optimization whenever we update postgres to see which ones the query planner can now do better. This change would still allow you to sidestep the query planner, but would also open up expressive CTEs to use in performance sensitive queries.
Note that even if we get the pluggable storage work into v12 - which I hope and think we can do, I'm certainly spending more time on it than I'd like - it'll not include a columnar storage on its own. There's others working on storage engines, but the furthest along intended for core aren't, to my knowledge, columnar. And even if somebody submits one for core, it might be a while till it's fast enough to satisfy your demands ;)