On Wed, Aug 6, 2025 at 9:35 AM Peter Geoghegan <p...@bowt.ie> wrote: > On Tue, Aug 5, 2025 at 4:56 PM Tomas Vondra <to...@vondra.me> wrote: > > True, the complex patch could prefetch the leaf pages.
There must be a similar opportunity for parallel index scans. It has that "seize the scan" concept where parallel workers do one-at-a-time locked linked list leapfrog. > What I meant was that the complex patch can make up for the fact that > direct I/O presumably won't ever have an equivalent to simple > read-ahead. Just by having a very flexible prefetching implementation > (and without any special sequential access heuristics ever being > required). I'm not so sure, there are certainly opportunities in different layers of the system. I'm going to dust off a couple of experimental patches (stuff I talked to Peter about back in Athens), and try to describe some other vague ideas Andres and I have bounced around over the past few years when chatting about what you lose when you turn on direct I/O. Basically, the stuff that we can't fix with "precise" I/O streaming as I like to call it, where it might still be interesting to think about opportunities to do fuzzier speculative lookahead. I'll start a new thread.
