Date: Thu, 8 Dec 2022 09:15:03 -0800
From: Nathan Bossart <nathandbossart@gmail.com>
To: Tom Lane <tgl@sss.pgh.pa.us>
Cc: Isaac Morland <isaac.morland@gmail.com>,
	Justin Pryzby <pryzby@telsasoft.com>,
	Pavel Luzanov <p.luzanov@postgrespro.ru>,
	pgsql-hackers@postgresql.org
Subject: Re: add \dpS to psql
Message-ID: <20221208171503.GA2113@nathanxps13>
References: <20221206193606.GB3078082@nathanxps13>
 <661148f4-c7f1-dec1-2bc8-29f3bd58e242@postgrespro.ru>
 <20221208023929.GB27893@telsasoft.com>
 <20221208041900.GB216874@nathanxps13>
 <39743.1670473532@sss.pgh.pa.us>
 <CAMsGm5drUvuH789D98Wo_bZggNhUn1LsBYuheJUbKaBsErtY9w@mail.gmail.com>
 <20221208050754.GA221222@nathanxps13>
 <45224.1670476523@sss.pgh.pa.us>
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Disposition: inline
Content-Transfer-Encoding: 8bit
In-Reply-To: <45224.1670476523@sss.pgh.pa.us>
Archived-At: 
 <https://www.postgresql.org/message-id/20221208171503.GA2113%40nathanxps13>
Precedence: bulk

On Thu, Dec 08, 2022 at 12:15:23AM -0500, Tom Lane wrote:
> Nathan Bossart <nathandbossart@gmail.com> writes:
>> On Wed, Dec 07, 2022 at 11:48:20PM -0500, Isaac Morland wrote:
>>> My previous analysis
>>> shows that there is no vast hidden demand for new privilege bits. If we
>>> implement MAINTAIN to control access to VACUUM, ANALYZE, REFRESH, CLUSTER,
>>> and REINDEX, we will cover everything that I can find that has seriously
>>> discussed on this list, and still leave 3 unused bits for future expansion.
> 
>> If we added CLUSTER, REFRESH MATERIALIZED VIEW, and REINDEX as individual
>> privilege bits, we'd still have 13 remaining for future use.
> 
> I think the appropriate question is not "have we still got bits left?".
> It should be more like "under what plausible scenario would it be useful
> to grant somebody CLUSTER but not VACUUM privileges on a table?".
> 
> I'm really thinking that MAINTAIN is the right level of granularity
> here.  Or maybe it's worth segregating exclusive-lock from
> not-exclusive-lock maintenance.  But I really fail to see how it's
> useful to distinguish CLUSTER from REINDEX, say.

The main idea behind this work is breaking out privileges into more
granular pieces.  If I want to create a role that only runs VACUUM on some
tables on the weekend, why ѕhould I have to also give it the ability to
ANALYZE, REFRESH, CLUSTER, and REINDEX?  IMHO we should really let the user
decide what set of privileges makes sense for their use-case.  I'm unsure
the grouping all these privileges together serves much purpose besides
preserving ACL bits.

The other reason I'm hesitant to group the privileges together is because I
suspect it will be difficult to reach agreement on how to do so, as
evidenced by past discussion [0].  That being said, I'm open to it if we
find a way that folks are happy with.  For example, separating
exclusive-lock and non-exclusive-lock maintenance actions seems like a
reasonable idea (which perhaps is an argument for moving VACUUM FULL out of
the VACUUM privilege).

[0] https://postgr.es/m/67a1d667e8ec228b5e07f232184c80348c5d93f4.camel%40j-davis.com

-- 
Nathan Bossart
Amazon Web Services: https://aws.amazon.com