Re: Remove Deprecated Exclusive Backup Mode

Greetings,

* Robert Haas (robertmhaas@gmail.com) wrote:
> On Mon, Feb 25, 2019 at 11:33 AM Stephen Frost <sfrost@snowman.net> wrote:
> > > Removing an option that people are currently using, and which they
> > > find better than other available options for reasons with which I
> > > understand that you disagree, will make users more sad.  Happy is
> > > better.
> >
> > I don't want to see more users stumbling over the issues with the
> > exclusive backup interface.  A better interface exists, and has existed
> > since 9.6.  The exclusive backup method was deprecated in 2016.  One of
> > the really bad things about the exclusive backup method is that it
> > *looks* like it works well and that there aren't any issues with it, but
> > when users hit the issues, they get very sad.  We might have 1000s of
> > happy users who never run into those issues and therefore don't want to
> > see us change anything, but what about the 10s or 100s of users who do
> > hit those issues, what do we tell them?  Seems like we're saying "well,
> > sorry, we knew those issues existed and it's unfortunate you hit them,
> > but there's this better thing that you *should* have been using and then
> > you wouldn't have hit those issues."  That doesn't seem likely to make
> > people happy with us.
>
> Sure, nobody wants users to keep hitting the same problems over and
> over again, but the above is still reductionist.  If we take the view
> that anything that can cause data corruption in any scenario, no
> matter how remote, is more than everything else, then we should just
> stop shipping minor releases and halt all other development work until
> we deal with https://commitfest.postgresql.org/22/528/ that has been
> open for 14 CommitFests and until we've fully dealt with every aspect
> of fsync-gate.  I don't see you treating either of those issues with
> the same sense of urgency with which you're treating this one, so
> evidently you feel entitled to decide which
> potentially-data-corrupting issues you want to attack first.

I'm not advocating for removing it today or tomorrow, or in some point
release.  We're already to the stage where we're talking about
something that wouldn't hit until late 2020.  Also, frankly, while there
might be something I could do to help, the whole madness with fsync gate
seems to be getting dealt with by some very smart people who know a
whole lot more about the internals of Linux and FreeBSD and I'd rather
not get in their way.  On the other hand, dealing with backups and
issues around backups has been something that I at least hope I've
gotten decent at.

> And I agree that you should have exactly that right.  Along similar
> lines, I'd like our users to have the right to decide when they want
> to upgrade from using exclusive backup mode to non-exclusive backup
> mode, instead of being forced into making a change at a time decided
> by our fiat.

Our users *have* that option, and have for over two years now, and
it'll have been 4 years by the time v13 comes out.  I agree that the
non-exclusive backup mode isn't as simple as the exclusive backup mode-
to use your example, what we *used* to be doing with fsync() was sure a
lot simpler than what we're going to be doing in the future, but
sometimes people have to make changes, even to more complicated APIs,
to make sure that they're doing things the right way to ensure that
their data doesn't get eaten.

Thanks!

Stephen