On 2015-12-09 19:36:11 +0900, Michael Paquier wrote:
> On Wed, Dec 9, 2015 at 4:41 PM, Andres Freund <andres@anarazel.de> wrote:
> > On 2015-12-09 16:30:47 +0900, Michael Paquier wrote:
> >> > I'm kinda wondering if it wouldn't have been better to go through shared
> >> > buffers in ResetUnloggedRelationsInDbspaceDir() instead of using
> >> > copy_file().
> >>
> >> For deployment with large shared_buffers settings, wouldn't that be
> >> actually more costly than the current way of doing? We would need to
> >> go through all the buffers at least once and look for the INIT_FORKNUM
> >> present to flush them.
> >
> > We could just check the file sizes on disk, and the check for the
> > contents of all the pages for each file.
>
> By doing it at replay, the flushes are spread across time. And by
> doing it at the end of recovery, all the flushes would be grouped. Do
> you think that's fine?
The point is that we'd no flushes, because the data would come directly
from shared buffers...