Re: Logical decoding on standby

On 2017-03-23 09:14:07 +0800, Craig Ringer wrote:
> On 23 March 2017 at 07:31, Andres Freund <andres(at)anarazel(dot)de> wrote:
> > On 2017-03-23 06:55:53 +0800, Craig Ringer wrote:
>
> >> I was thinking that by disallowing snapshot use and output plugin
> >> invocation we'd avoid the need to support cancellation on recovery
> >> conflicts, etc, simplifying things considerably.
> >
> > That seems like it'd end up being pretty hacky - the likelihood that
> > we'd run into snapbuild error cross-checks seems very high.
>
> TBH I'm not following this. But I haven't touched snapbuild much yet,
> Petr's done much more with snapbuild than I have.

We can't just assume that snapbuild is going to work correctly when it's
prerequisites - pinned xmin horizon - isn't working.

> We're not going to have robust logical replication that's suitable for
> HA and failover use on high load systems until 2020 or so, with Pg 12.
> We'll need concurrent decoding and apply, which nobody's even started
> on AFAIK, we'll need sequence replication, and more.

These seem largely unrelated to the topic at hand(nor do I agree on all
of them).

> So I'd really, really like to get some kind of HA picture other than
> "none" in for logical decoding based systems. If it's imperfect, it's
> still something.

I still think decoding-on-standby is simply not the right approach as
the basic/first HA approach for logical rep. It's a nice later-on
feature. But that's an irrelevant aside.

I don't understand why you're making a "fundamental" argument here - I'm
not arguing against the goals of the patch at all. I want as much stuff
committed as we can in a good shape.

> What do _you_ see as the minimum acceptable way to achieve the ability
> for a logical decoding client to follow failover of an upstream to a
> physical standby? In the end, you're one of the main people whose view
> carries weight in this area, and I don't want to develop yet another

I think your approach here wasn't that bad? There's a lot of cleaning
up/shoring up needed, and we probably need a smarter feedback system. I
don't think anybody here has objected to the fundamental approach?

Greetings,

Andres Freund