Re: 9.6 -> 10.0

On Tue, Mar 22, 2016 at 7:03 PM, Josh berkus <josh(at)agliodbs(dot)com> wrote:

> On 03/22/2016 10:52 AM, Peter Geoghegan wrote:
> > On Tue, Mar 22, 2016 at 10:41 AM, Josh berkus <josh(at)agliodbs(dot)com> wrote:
> >> It's important to remember that PR strategy and engineering truth have
> >> only a passing acquaintance. While we don't want to promote vaporware,
> >> we do sometimes soft-pedal our own features to our project's detriment.
> >> In the current atomosphere of VC-funded hype, we'd do a bit better to
> >> trumpet our accomplishements early and often.
> >
> > I see what you mean.
> >
> > The question must be asked: What feature *would* meet that "major
> > version bump" standard? If it's not extensive parallelism, then I
> > don't know what else it could be.
>
> Well, if we had pglogical AND parallel, I would be pushing hard for
> 10.0. As it is, I was going to wait to see what else gets in.
>

If we had that, I think it would be a clear-cut case.

> As it is, we have parallel and we have all of the BDR dependancies
> merged in, no? That still seems like a new era for PostgreSQL; I think
> we can expect the next few releases to be all about (a) parallelizing
> more things and (b) building out clustering stuff.
>

Not entirely sure if we have all the requirements. The one on the queue is
sequence access methods, but I believe that's only really needed for BDR,
not pglogical.

Having chatted some more with Robert about this, there's another point.

Some people mention partitioning. I don't think that's enough *alone* to
bump to 10.0. In fact, I definitely think parallel is bigger than that, by
far.

Personally, I think the zero downtime upgrade is enough, that would mean
pglogical. Parallel is borderline. Others will say the other way around -
parallel is the clear cut, pglogical is the borderline.

There is also the vacuum freeze updates, which are *huge* for a very small
subset of users. That subset is not going to get smaller. I don't think
anybody argues that alone is enough to make it 10.0 or ever will be. And
there will always be features like that.

But in the end, this reasoning might lead us to never getting to 10.0, even
if we have one huge feature in each release. I think we can basically say
that there is no one feature that will make it good enough. And are we
really going to get all of them at once?

With a couple more of the "big but not 10.0-on-their-own" that are
currently in the CF I think it should be a 10.0. To answer Roberts question
before, specifically if we get things like multivariate statistics, casual
reads, multiple sync standbys, snapshot too old, relation extend
scalability and maybe unique joins, I definitely say we have that. And not
all of them, pick two or three of those and I think we have a 10.0. (oh,
and of course the updated backup APIs :P must have those! :P)

Bottom line is, there's a lot still in the queue. Let's not make a decision
until we know which of that gets in. If *none* of it gets in, we're still
on the fence. But let's hope that's not going to be the case.

--
Magnus Hagander
Me: http://www.hagander.net/
Work: http://www.redpill-linpro.com/