[pgaddict]

I don't know which "architectural peculiarities" you have in mind, but the issues we have are largely independent of the architecture. So I'm not sure what exactly you think we should be fixing architecture-wise ...

The way I see it the issues we have in query planning/optimization are largely independent of the overall architecture. It's simply a consequence of relying on statistics too much. But there's always going to be inaccuracies, because the whole point of stats is that it's a compact/lossy approximation of the data. No matter how much more statistics we might collect there will always be some details missing, until the statistics are way too large to be practical.

There are discussions about considering how "risky" a given plan is - some plan shapes may be more problematic, but I don't think anyone submitted any patch so far. But even with that would not be a perfect solution - no approach relying solely on a priori information can be.

IMHO the only way forward is to work on making the plans more robust to this kind of mistakes. But that's very hard too.