Re: post-freeze damage control

On Tue, Apr 16, 2024 at 7:20 PM Jeff Davis <pgsql@j-davis.com> wrote:
> We maintain the invariant:
>
>    XLogCtl->logFlushResult <= XLogCtl->logWriteResult
>
> and the non-shared version:
>
>    LogwrtResult.Flush <= LogwrtResult.Write
>
> and that the requests don't fall behind the results:

I had missed the fact that this commit added a bunch of read and write
barriers; and I think that's the main reason why I was concerned. The
existence of those barriers, and the invariants they are intended to
maintain, makes me feel better about it.

Thanks for writing back about this!

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