2019年8月1日(木) 1:41 Tom Lane <tgl@sss.pgh.pa.us>: > > Robert Haas <robertmhaas@gmail.com> writes: > > Yeah, but I have to admit that this whole design makes me kinda > > uncomfortable. Every time somebody comes up with a new figure of > > merit, it increases not only the number of paths retained but also the > > cost of comparing two paths to possibly reject one of them. A few > > years ago, you came up with the (good) idea of rejecting some join > > paths before actually creating the paths, and I wonder if we ought to > > try to go further with that somehow. Or maybe, as Peter Geoghegan, has > > been saying, we ought to think about planning top-down with > > memoization instead of bottom up (yeah, I know that's a huge change). > > It just feels like the whole idea of a list of paths ordered by cost > > breaks down when there are so many ways that a not-cheapest path can > > still be worth keeping. Not sure exactly what would be better, though. > > Yeah, I agree that add_path is starting to feel creaky. I don't > know what to do instead though. Changing to a top-down design > sounds like it would solve some problems while introducing others > (not to mention the amount of work and breakage involved). > Hmm... It looks the problem we ought to revise about path construction is much larger than my expectation, and uncertain for me how much works are needed.
Although it might be a workaround until fundamental reconstruction, how about to have a margin of estimated cost to reject paths? Current add_path() immediately rejects lesser paths if its cost is even a little more expensive than the compared one. One the other hands,
Hmm.. I don't think so. Currently add_path() uses fuzzy comparisons on costs of two paths, although the fuzz factor (1%) is hard coded and not user-controllable.
I understand it is not an essential re-design of path-construction logic, and may have limitation. However, amount of works are reasonable and no side- effect. (current behavior = 0% threshold). How about your opinions?
How's about Tom's suggestion on adding another dimension in add_path() to be considered, just like how it considers paths of better sort order or parallel-safe?