MIME-Version: 1.0
References: 
 <CA+3i_M8m1k2gFch+tU0JmAQh9FRV+pFrfTXDrJo+BqmwsTmOhg@mail.gmail.com>
In-Reply-To: 
 <CA+3i_M8m1k2gFch+tU0JmAQh9FRV+pFrfTXDrJo+BqmwsTmOhg@mail.gmail.com>
From: Amit Langote <amitlangote09@gmail.com>
Date: Thu, 16 Apr 2026 18:28:47 +0900
Message-ID: 
 <CA+HiwqH+0rgMNQDog0AT9dVt0CGjXza_Li80njDhynWqzUwvZw@mail.gmail.com>
Subject: Re: Reject invalid databases in pg_get_database_ddl()
To: Lakshmi N <lakshmin.jhs@gmail.com>
Cc: "pgsql-hackers@lists.postgresql.org" <pgsql-hackers@lists.postgresql.org>,
 andrew@dunslane.net
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable
Archived-At: 
 <https://www.postgresql.org/message-id/CA%2BHiwqH%2B0rgMNQDog0AT9dVt0CGjXza_Li80njDhynWqzUwvZw%40mail.gmail.com>
Precedence: bulk

Hi,

On Thu, Apr 16, 2026 at 5:20=E2=80=AFPM Lakshmi N <lakshmin.jhs@gmail.com> =
wrote:
> pg_get_database_ddl() is not checking for databases in an invalid state
> before producing ddl statements. This caused the function to emit
> CONNECTION_LIMIT =3D -2, which is invalid SQL that Postgres rejects.
> A database row can be in this inconsistent state longer, for example
> server crashed during a drop database.
>
> Attached patch to fix this issue by doing a database_is_invalid_form()
> check early in pg_get_database_ddl_internal().

Thanks for the report.

Hmm, I see that the function will happily emit datconnlimit =3D -2 and
your patch catches that at the top instead of down below near this
code:

/* CONNECTION LIMIT */
if (dbform->datconnlimit !=3D -1)
{
    resetStringInfo(&buf);
    appendStringInfo(&buf, "ALTER DATABASE %s CONNECTION LIMIT =3D %d;",
                     quote_identifier(dbname), dbform->datconnlimit);
    statements =3D lappend(statements, pstrdup(buf.data));
}

which, I guess, makes sense.

The comment is correct but could be more explicit:

    /*
     * Reject invalid databases: datconnlimit =3D -2 would be emitted as
     * CONNECTION LIMIT =3D -2, which fails on replay.
     */

--=20
Thanks, Amit Langote