Date: Fri, 1 Dec 2017 19:10:27 -0300
From: Alvaro Herrera <alvherre@alvh.no-ip.org>
To: David Rowley <david.rowley@2ndquadrant.com>
Cc: Simon Riggs <simon@2ndquadrant.com>,
	Robert Haas <robertmhaas@gmail.com>,
	Pg Hackers <pgsql-hackers@postgresql.org>,
	Maksim Milyutin <milyutinma@gmail.com>
Subject: Re: [HACKERS] Proposal: Local indexes for partitioned table
Message-ID: <20171201221027.edqapmu4gcisvqbz@alvherre.pgsql>
MIME-Version: 1.0
Content-Type: text/plain; charset=iso-8859-1
Content-Disposition: inline
Content-Transfer-Encoding: 8bit
In-Reply-To: 
 <CAKJS1f9G6hnahJpoLaHavrKt0uPyYZncEi2rq__kLQcrGE_FMQ@mail.gmail.com>
User-Agent: NeoMutt/20170306-137-4415bd-dirty (1.8.0)
Precedence: bulk

David Rowley wrote:

> So, then this patch is only really intended as a syntax shortcut for
> mass adding of indexes to each partition?

This patch is intended to serve as a basis on which to construct further
features, just like every other patch we apply.

> I feel like we could do better here with little extra effort. The
> DETACH index feature does not really seem required for this patch. I
> think just ensuring a matching index exists on each leaf partition and
> creating any which don't exist before creating the index on the target
> partitioned table seems like the correct solution. That way we can
> make that index indisvalid flag have a valid meaning all the time.
> Some later patch can invent some way to replace a bloated index.

What you're saying is that I've written code for A+B, and you're
"interested in C (which is incompatible with B), so can we have A+C and
drop B".  But in reality, there exists (unwritten) D that solves the
incompatiblity between B and C.  I'm just saying it's essentially the
same to postpone C+D than to postpone B+D, and I already have B written;
plus that way we don't have to come up with some novel way to handle
pg_dump support.  So can we get A+B committed and discuss C+D later?

A = partitioned indexes
B = pg_dump support based on ATTACH
C = your proposed planner stuff
D = correct indisvalid setting for partitioned indexes (set to false
    when a partition does not contain the index)

The patch in this thread is A+B.

> Perhaps later we can invent some generic way to replace a physical
> leaf index for a given partitioned index perhaps with the same patch
> that might allow us to replace an index which is used by a constraint,
> which to me seems like a feature we should have had years ago.

This is a hypothetical feature E which would be nice (for partitioned
indexes and for ordinary indexes too) but is not strictly necessary.

-- 
Álvaro Herrera                https://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services