Re: [9.4] Make full_page_writes only settable on server start?

On Tue, 2013-09-03 at 15:42 -0400, Robert Haas wrote:
> Although this is true, the administrator's estimate of whether that
> guarantee is or is not provided might not be as consistent as the
> hardware behavior itself. I am generally of the feeling that having
> to restart the server to change setting sucks, and while it surely
> sucks less for this setting than some, mostly because few people
> change this setting in the first place, I'm still not convinced that
> this is moving in the right direction.

I think code complexity matters quite a lot. If we can eliminate some
complex code in a complex area, and all we give up is a feature with
essentially no use case, that sounds like we're moving in the right
direction to me.

I suppose some might be using it as a hack when they really just want to
temporarily disable WAL during a load or something. Seems like a blunt
tool though, and I haven't heard of anyone doing that or suggesting it.
And it doesn't store the page hole anyway, so the FPI during a load is
ordinarily quite small.

> > Then, I intend to write another patch to make the full-page writes for
> > checksums honor the full_page_writes setting. That will be easier to
> > write once it's a PGC_POSTMASTER.
>
> I don't think I know exactly what this means.

XLogSaveBufferForHint() calls XLogCheckBuffer() but doesn't also look at
the full page writes setting (like in XLogInsert()). That means, if
checksums are enabled and full_page_writes is off, we'll still write
some full page images for checksums. I'd like to remedy that.

Regards,
Jeff Davis