Date: Mon, 5 Aug 2019 09:53:37 -0700
From: Andres Freund <andres@anarazel.de>
To: Tom Lane <tgl@sss.pgh.pa.us>
Cc: Robert Haas <robertmhaas@gmail.com>, David Fetter <david@fetter.org>,
	Ian Barwick <ian.barwick@2ndquadrant.com>,
	Tomas Vondra <tomas.vondra@2ndquadrant.com>,
	Stephen Frost <sfrost@snowman.net>,
	Amit Kapila <amit.kapila16@gmail.com>,
	"pgsql-hackers@postgresql.org" <pgsql-hackers@postgresql.org>
Subject: Re: [PATCH] Stop ALTER SYSTEM from making bad assumptions
Message-ID: <20190805165337.akba75vusf26irku@alap3.anarazel.de>
References: <20190803004558.jfkbngi6g64eaqku@alap3.anarazel.de>
 <CAOuzzgpuH+uvT3Yvw5KC=keEqGifSc4eXQ6DB=yrowYyre2YvA@mail.gmail.com>
 <20190803010802.rs3cz2kjq4dl4olu@alap3.anarazel.de>
 <20190803124111.2aaddumd7url5wmq@development>
 <16320.1564851579@sss.pgh.pa.us>
 <d294b0d6-c00b-8a38-864f-4943a9471a6e@2ndquadrant.com>
 <20190805162203.GR31493@fetter.org>
 <4768.1565022305@sss.pgh.pa.us>
 <CA+TgmobXGnq_MqFnD4UnRVWYLrmjVadV_-OtxfxFkA_gpYDqzA@mail.gmail.com>
 <5739.1565023404@sss.pgh.pa.us>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <5739.1565023404@sss.pgh.pa.us>
Precedence: bulk

Hi,

On 2019-08-05 12:43:24 -0400, Tom Lane wrote:
> Yeah, good point.  There are a lot of other cases where you really
> don't want system startup to happen, too.

Agreed.


> On the other hand, people have also opined that they want full error
> checking on the inserted values, and that seems out of reach with less
> than a full running system (mumble extensions mumble).

I think the error checking ought to be about as complete as the one we
do during a normal postmaster startup. Afaict that requires loading
shared_preload_library extensions, but does *not* require booting up far
enough to run GUC checks in a context with database access.

The one possible "extension" to that that I can see is that arguably we
might want to error out if DefineCustom*Variable() doesn't think the
value is valid for a shared_preload_library, rather than just WARNING
(i.e. refuse to start). We can't really do that for other libraries, but
for shared_preload_libraries it might make sense.  Although I suspect
the better approach would be to just generally convert that to an error,
rather than having some startup specific logic.

Greetings,

Andres Freund