Blogging by Heath™

TL;DR - Recently set out to speed up a slow query behind a multi-column cursor paginated endpoint. A simple switch to represent the where filters as a tuple makes a huge difference in performance on the exact same data with the same indices. Eg - change where ( a > c or (a = c and b > d )) to where (a, b) > (c, d). In the “tuple” case postgres can use the index more efficiently and get more of the needed rows using the index condition, versus walking it, reading rows in and filtering.