From: Tom Lane <tgl@sss.pgh.pa.us>
To: Nathan Bossart <nathandbossart@gmail.com>
cc: pgsql-hackers@lists.postgresql.org, alex work <alexwork033@gmail.com>,
        Robert Haas <robertmhaas@gmail.com>
Subject: Re: Slow GRANT ROLE on PostgreSQL 16 with thousands of ROLEs
In-reply-to: <20240326184819.GA3559028@nathanxps13>
References: <20240322163952.GA2347986@nathanxps13>
 <939945.1711126395@sss.pgh.pa.us> <20240322165448.GA2350322@nathanxps13>
 <20240322193506.GA2352051@nathanxps13> <1084642.1711140109@sss.pgh.pa.us>
 <20240325144743.GA2618709@nathanxps13> <2490242.1711379319@sss.pgh.pa.us>
 <20240325151647.GA2662364@nathanxps13> <20240326165918.GA3350222@nathanxps13>
 <2999167.1711476963@sss.pgh.pa.us> <20240326184819.GA3559028@nathanxps13>
Comments: In-reply-to Nathan Bossart <nathandbossart@gmail.com>
	message dated "Tue, 26 Mar 2024 13:48:19 -0500"
MIME-Version: 1.0
Content-Type: text/plain; charset="us-ascii"
Content-ID: <3088175.1711480080.1@sss.pgh.pa.us>
Date: Tue, 26 Mar 2024 15:08:00 -0400
Message-ID: <3088176.1711480080@sss.pgh.pa.us>
Archived-At: 
 <https://www.postgresql.org/message-id/3088176.1711480080%40sss.pgh.pa.us>
Precedence: bulk

Nathan Bossart <nathandbossart@gmail.com> writes:
> On Tue, Mar 26, 2024 at 02:16:03PM -0400, Tom Lane wrote:
>> ... I'm not sold on your "ROLES_LIST_BLOOM_THRESHOLD * 10"
>> value.  Maybe it doesn't matter though.

> Yeah, I wasn't sure how much to worry about this.  I figured that we might
> as well set it to a reasonable estimate based on the description of the
> parameter.  This description claims that the filter should work well if
> this is off by a factor of 5 or more, and 50x the threshold sounded like it
> ought to be good enough for anyone, so that's how I landed on 10x.  But as
> you point out, this value will be disregarded altogether, and it will
> continue to be ignored unless the filter implementation changes, which
> seems unlikely.  If you have a different value in mind that you would
> rather use, I'm fine with changing it.

No, I have no better idea.  As you say, we should try to provide some
semi-reasonable number in case bloom_create ever starts paying
attention, and this one seems fine.

>> I do not like, even a little bit, your use of a static variable to
>> hold the bloom filter pointer.

> Ah, yes, that's no good.  I fixed this in the new version.

My one remaining suggestion is that this comment isn't very precise
about what's happening:

     * If there is a previously-created Bloom filter, use it to determine
     * whether the role is missing from the list.  Otherwise, do an ordinary
     * linear search through the existing role list.

Maybe more like

     * If there is a previously-created Bloom filter, use it to try to
     * determine whether the role is missing from the list.  If it
     * says yes, that's a hard fact and we can go ahead and add the
     * role.  If it says no, that's only probabilistic and we'd better
     * search the list.  Without a filter, we must always do an ordinary
     * linear search through the existing list.

LGTM other than that nit.

			regards, tom lane