Re: LogwrtResult contended spinlock

On Fri, Feb 23, 2024 at 1:18 AM Jeff Davis <pgsql(at)j-davis(dot)com> wrote:
> I don't see the global non-shared variable as a huge problem, so if it
> serves a purpose then I'm fine keeping it. Perhaps we could make it a
> bit safer by using some wrapper functions.

I actually really hate these kinds of variables. I think they likely
mask various bugs (where the value might not be up to date where we
think it is) and anti-bugs (where we actually rely on the value being
out of date but we don't know that we do because the code is so
opaque). My vintage 2021 adventures in getting rid of the global
variable ThisTimeLineID and a few other global variables found some
actual but minor bugs, also found a bunch of confused code that didn't
really do what it thought it did, and took up a huge amount of my time
trying to analyze what was happening. I'm not prepared to recommend
what we should do in this particular case. I would like to believe
that the local copies can be eliminated somehow, but I haven't studied
the code or done benchmarking so I don't really know enough to guess
what sort of code changes would or would not be good enough.

--
Robert Haas
EDB: http://www.enterprisedb.com