On Wed, Jun 27, 2018 at 12:52 PM, Andres Freund <and...@anarazel.de> wrote: > A related question is if it matters - without complicating the code I > don't see how we could release all the locks in that loop. Therefore no > interrupts can be accepted. I hope I'm missing something?
I agree. In general, page deletion is the most complicated part of nbtree concurrency, by far (if we just had the basic L&Y, the concurrency aspects would be far easier to grasp). Doing better in _bt_unlink_halfdead_page() seems extremely difficult, and very unlikely to be worthwhile. -- Peter Geoghegan