Re: [HACKERS] MERGE SQL Statement for PG11

On 31 October 2017 at 12:56, Stephen Frost <sfrost@snowman.net> wrote:
> Simon,
>
> * Simon Riggs (simon@2ndquadrant.com) wrote:
>> On 30 October 2017 at 19:55, Stephen Frost <sfrost@snowman.net> wrote:
>> > I don't think MERGE should be radically different from other database
>> > systems and just syntax sugar over a capability we have.
>>
>> I've proposed a SQL Standard compliant implementation that would do
>> much more than be new syntax over what we already have.
>>
>> So these two claims aren't accurate: "radical difference" and "syntax
>> sugar over a capability we have".
>
> Based on the discussion so far, those are the conclusions I've come to.
> Saying they're isn't accurate without providing anything further isn't
> likely to be helpful.

I'm trying to be clear, accurate and non-confrontational.

I appreciate those are your conclusions, which is why I need to be
clear that the proposal has been misunderstood on those stated points.

What else would you like me to add to be helpful? I will try...

I've spent weeks looking at other implementations and combing the
SQLStandard with a fine tooth comb to see what is possible and what is
not. My proposal shows a new way forwards and yet follows the standard
in horrible detail. It has taken me nearly 2 months to write the doc
page submitted here. It is not simply new syntax stretched over
existing capability. The full syntax of MERGE is quite powerful and
will take some months to make work, even with my proposal.

I'm not sure what else to say, but I am happy to answer specific
technical questions that have full context to allow a rational
discussion.

>> > Time changes
>> > many things, but I don't think anything's changed in this from the prior
>> > discussions about it.
>>
>> My proposal is new, that is what has changed.
>
> I'm happy to admit that I've missed something in the discussion, but
> from what I read the difference appeared to be primairly that you're
> proposing it, and doing so a couple years later.
>
>> At this stage, general opinions can be misleading.
>
> I'd certainly love to see a MERGE capability that meets the standard and
> works in much the same way from a user's perspective as the other RDBMS'
> which already implement it.

The standard explains how it should work, with the proviso that the
standard allows us to define concurrent behavior. Serge has explained
that he sees that my proposal covers the main use cases. I don't yet
see how to cover all, but I'm looking to do the most important use
cases first and other things later.

> From prior discussions with Peter on
> exactly that subject, I'm also convinced that having that would be a
> largely independent piece of work from the INSERT .. ON CONFLICT
> capability that we have (just as MERGE is distinct from similar UPSERT
> capabilities in other RDBMSs).
>
> The goal here is really just to avoid time wasted developing MERGE based
> on top of the INSERT .. ON CONFLICT system only to have it be rejected
> later because multiple other committers and major contributors have said
> that they don't agree with that approach.  If, given all of this
> discussion, you don't feel that's a high risk with your approach then by
> all means continue on with what you're thinking and we can review the
> patch once it's posted.

It is certainly a risk and I don't want to waste time either. I am
happy to consider any complete proposal for how to proceed, including
details and/or code, but I will be prioritising things to ensure that
we have a candidate feature for PG11 from at least one author, rather
than a research project for PG12+.

The key point here is concurrency. I have said that there is an
intermediate step that is useful and achievable, but does not cover
every case. Serge has also stated this - who it should be noted is the
only person in this discussion that has already written the MERGE
statement in SQL, for DB2, so I give more weight to his technical
opinion than I do to others, at this time.

-- 
Simon Riggs                http://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services


-- 
Sent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-hackers