Michael Paquier <michael@paquier.xyz> writes:
> On Wed, Dec 26, 2018 at 08:35:22PM -0500, Tom Lane wrote:
>> Oh, I see. Yeah, if we're ignoring errors anyway, the stat calls
>> seem redundant.
> For this one, I think that we could simplify as attached (this causes
> open() to fail additionally because of the sync flags, but that's not
> really worth worrying). Thoughts?
Actually, now that I think a bit more, this isn't a good idea. We want
standby_signal_file_found (resp. recovery_signal_file_found) to get set
if the file exists, even if we're unable to fsync it for some reason.
A counterexample to the proposed patch is that a signal file that's
read-only to the server will get ignored, which it should not be.
regards, tom lane