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

On Tue, Sep 3, 2013 at 07:47:22PM +0200, Magnus Hagander wrote:
>
> On Sep 3, 2013 6:23 PM, "Josh Berkus" <josh(at)agliodbs(dot)com> wrote:
> >
> > Peter,
> >
> >
> > > I wonder, is anyone really running in production with full_page_writes
> > > off?
> >
> > We're doing so with clients on Joyent (ZFS). And also for "ephemeral"
> > postgres instances -- ones where we've also turned off fsync.
> >
> > It's also quite reasonable to assume that future filesystems will also
> > be able to make better torn page guarantees.
> >
> > So it should remain a setting. I see no reason why it should be a
> > SIGHUP setting, though, if there's any code maintenance required to
> > support it -- frankly, I only change this setting at first system
> > deployment.
> >
>
> This matches my experience as well - people certainly use it, but don't change
> it during runtime.
>
> Not having looked at the code, but doesn't pg_start_backup use at least parts
> of the same code path? That one will definitely still be able to modify it at
> runtime....

Yes, this was already mentioned in the thread, but didn't get noticed by
a few folks. In summary, no changes needed here.

--
Bruce Momjian <bruce(at)momjian(dot)us> http://momjian.us
EnterpriseDB http://enterprisedb.com

+ It's impossible for everything to be true. +