Hi,
Here's an updated patch using GetFakeLSNForUnloggedRel() instead of the
atomic counter. I think this looks much nicer and less invasive, as it
simply uses XLogCtl shared memory (instead of having to pass a new
pointer everywhere).
We still need to pass the is_parallel flag, though. I wonder if we could
get rid of that too, and just use GetFakeLSNForUnloggedRel() for both
parallel and non-parallel builds? Why wouldn't that work?
I've spent quite a bit of time testing this, but mostly for correctness.
I haven't redone the benchmarks, that's on my TODO.
regards
--
Tomas Vondra
