On Tue, Jul 20, 2021 at 12:26 PM Tom Lane <tgl@sss.pgh.pa.us> wrote:
> > I can try that on the gcc farm in a bit.
Thanks!
> Hmm, it compiles cleanly, but something seems drastically wrong,
> because performance is just awful. On the other hand, I don't
> know what sort of storage is underlying this instance, so maybe
> that's to be expected?
Ouch. I assume this was without wal_method=minimal (or it'd have
reached the new code and failed completely, based on the pg_test_fsync
result).
> open_datasync n/a*
I'm waiting for access, but I see from man pages that closed source
ZFS doesn't accept DIRECTIO_ON, so it may not be possible to see it
work on an all-ZFS system that you can't mount a new FS on. Hmm.
Well, many OSes have file systems that can't do it (ext4 journal=data,
etc). One problem is that we don't treat all OSes the same when
selecting wal_sync_method, even though O_DIRECT is complicated on many
OSes. It would also be nice if the choice to use direct I/O were
independently controlled, and ... [trails off]. Alright, I'll leave
this on ice for now.
