From: Tom Lane <tgl@sss.pgh.pa.us>
To: Andres Freund <andres@anarazel.de>
cc: "Sisson, David" <David.Sisson@dell.com>,
        Tomas Vondra <tomas.vondra@enterprisedb.com>,
        "pgsql-bugs@lists.postgresql.org" <pgsql-bugs@lists.postgresql.org>
Subject: Re: BUG #17757: Not honoring huge_pages setting during initdb causes
 DB crash in Kubernetes
In-reply-to: <20230124003737.owezrb6ffen6dhb3@awork3.anarazel.de>
References: <17757-dbdfc1f1c954a6db@postgresql.org>
 <4ce40288-5b88-f9fb-ea11-fc932c61472b@enterprisedb.com>
 <1113115.1674343827@sss.pgh.pa.us>
 <127328f2-4e01-e9ec-c9b1-76aef967343f@enterprisedb.com>
 <20230122020826.x6geac6qjiinr66a@awork3.anarazel.de>
 <LV2PR19MB5765BF10D8D3FD015C6011468EC89@LV2PR19MB5765.namprd19.prod.outlook.com>
 <20230123195504.uvlcd3aytn3jg744@awork3.anarazel.de>
 <2139743.1674514306@sss.pgh.pa.us>
 <20230124003737.owezrb6ffen6dhb3@awork3.anarazel.de>
Comments: In-reply-to Andres Freund <andres@anarazel.de>
	message dated "Mon, 23 Jan 2023 16:37:37 -0800"
MIME-Version: 1.0
Content-Type: text/plain; charset="us-ascii"
Content-ID: <2151513.1674521119.1@sss.pgh.pa.us>
Content-Transfer-Encoding: quoted-printable
Date: Mon, 23 Jan 2023 19:45:19 -0500
Message-ID: <2151514.1674521119@sss.pgh.pa.us>
Archived-At: 
 <https://www.postgresql.org/message-id/2151514.1674521119%40sss.pgh.pa.us>
Precedence: bulk

Andres Freund <andres@anarazel.de> writes:
> On 2023-01-23 17:51:46 -0500, Tom Lane wrote:
>> Yeah, I think that has enough other potential applications to be worth
>> doing.  Here's a quick draft patch (sans user-facing docs as yet).
>> It injects any given values into postgresql.auto.conf, not
>> postgresql.conf proper.  I did that mainly because the latter looked
>> beyond the abilities of the primitive string-munging code we have in
>> there, but I think it can be argued to be a reasonable choice anyway.

> Oh, I had thought we'd just pass them on with -c to the processes that i=
nitdb
> starts. But perhaps just persisting them isn't a bad idea...

It certainly seems to me that that would be the mainstream use-case,
so why not fill in the file as the user probably wants?  They can
always change it.  Also, as I mentioned, the expectation is that
initdb will set up a known-working combination of settings; and
we don't really know that if we leave off whatever was injected by
"-c".  In the case at hand, if we don't propagate "huge_pages =3D off"
to the installed configuration, the server still won't work.

			regards, tom lane