On Mon, Sep 2, 2013 at 7:10 PM, Jeff Davis <pgsql@j-davis.com> wrote:
> I'd like to submit a patch to just make it into a PGC_POSTMASTER and
> remove the code to support changing it.
Makes sense to me.
I wonder, is anyone really running in production with full_page_writes
off? I talked to someone a while ago who used Postgres on ZFS with
data journaling, and was perfectly well aware of the fact that
theoretically he could safely turn the setting off, and yet chose not
to. Now, I know that Greg Smith's book describes the conditions in
which it's acceptable and the precautions that should be taken and so
on, but in my (admittedly relatively limited) experience, no one
actually does it in production. My sample size for people who at least
considered doing it (that both believed that Postgres could write
pages atomically on their hardware/FS, and also knew that
full_page_writes exists and what it means) is only 1.
At least in people's minds, it might be that the knowledge that no one
runs with full_page_writes off is enough to put them off. It's like
building Postgres with a non-standard BLCKSZ -- even if you had solid
evidence that it would help performance in your case, would you really
want to do it?
--
Peter Geoghegan