Date: Thu, 20 Apr 2023 15:02:42 -0700
From: Nathan Bossart <nathandbossart@gmail.com>
To: Kyotaro Horiguchi <horikyota.ntt@gmail.com>
Cc: zhjwpku@gmail.com, xu.xw2008@163.com, pgsql-bugs@lists.postgresql.org
Subject: Re: BUG #17903: There is a bug in the KeepLogSeg()
Message-ID: <20230420220242.GB1435672@nathanxps13>
References: <17903-4288d439dee856c6@postgresql.org>
 <20230420.120417.1609083651022565895.horikyota.ntt@gmail.com>
 <CAEG8a3L2BB5k9USO4fC_wQjULGX-1rqaX2JSHn1_0vco8KB28Q@mail.gmail.com>
 <20230420.175814.1157027125103997227.horikyota.ntt@gmail.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20230420.175814.1157027125103997227.horikyota.ntt@gmail.com>
Archived-At: 
 <https://www.postgresql.org/message-id/20230420220242.GB1435672%40nathanxps13>
Precedence: bulk

On Thu, Apr 20, 2023 at 05:58:14PM +0900, Kyotaro Horiguchi wrote:
> +		/*
> +		 * Slot minimum LSN could be greater than recptr. In such cases, no
> +		 * segments are protected by slots but we still need to keep segno in a
> +		 * reasonable range for subsequent calculations to avoid underflow.
> +		 */
> +		if (segno > currSegNo)
> +			segno = currSegNo;
> +

I wonder if it'd be better to instead change

	if (currSegNo - segno > slot_keep_segs)
to
	if (currSegNo > segno + slot_keep_segs)

and

	if (currSegNo - segno < keep_segs)
to
	if (currSegNo < segNo + keep_segs)

If segno > currSegNo, the first conditional would be false as expected, and
the second would be true as expected.  We could also use
pg_sub_u64_overflow() to detect underflow, but that might be excessive in
this case.

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