Subject: Re: PATCH: two slab-like memory allocators
To: Robert Haas <robertmhaas@gmail.com>, Andres Freund <andres@anarazel.de>
References: <20170228034232.4h7ouyhxq4lbbao2@alap3.anarazel.de>
 <1240.1488254244@sss.pgh.pa.us>
 <20170228051630.3qcevakcaaa6b4ef@alap3.anarazel.de>
 <1671.1488259784@sss.pgh.pa.us>
 <20170228074420.aazv4iw6k562mnxg@alap3.anarazel.de>
 <20170228184122.extbigaer5ryrt3z@alap3.anarazel.de>
 <20170301041835.wli3viwnxr5wonfv@alap3.anarazel.de>
 <20170301042936.usv23f4iaumeyptf@alap3.anarazel.de>
 <20170301225545.eoevemua234g4fzq@alap3.anarazel.de>
 <CA+TgmoY6F5ry7hiabiWFvU40bVgVruLB9YGP0D2O4Tb_zNXOUw@mail.gmail.com>
 <20170306174406.pfs3as44gc7a5ps2@alap3.anarazel.de>
 <CA+TgmoaQ34gXj3G96xqU6chxTOHSHxip07+GySRxS-5FgHs+NQ@mail.gmail.com>
Cc: Tom Lane <tgl@sss.pgh.pa.us>, Petr Jelinek
 <petr.jelinek@2ndquadrant.com>, Petr Jelinek <petr@2ndquadrant.com>,
 John Gorman <johngorman2@gmail.com>,
 pgsql-hackers <pgsql-hackers@postgresql.org>
From: Tomas Vondra <tomas.vondra@2ndquadrant.com>
Message-ID: <b09976b6-b662-da59-3f65-a4396e8de5b3@2ndquadrant.com>
Date: Mon, 6 Mar 2017 19:49:56 +0100
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:45.0) Gecko/20100101
 Thunderbird/45.4.0
MIME-Version: 1.0
In-Reply-To: 
 <CA+TgmoaQ34gXj3G96xqU6chxTOHSHxip07+GySRxS-5FgHs+NQ@mail.gmail.com>
Content-Type: text/plain; charset=utf-8; format=flowed
Content-Transfer-Encoding: 7bit
Precedence: bulk
Sender: pgsql-hackers-owner@postgresql.org

On 03/06/2017 07:05 PM, Robert Haas wrote:
> On Mon, Mar 6, 2017 at 12:44 PM, Andres Freund <andres@anarazel.de> wrote:
>> On 2017-03-06 12:40:18 -0500, Robert Haas wrote:
>>> On Wed, Mar 1, 2017 at 5:55 PM, Andres Freund <andres@anarazel.de> wrote:
>>>> The issue was that on 32bit platforms the Datum returned by some
>>>> functions (int2int4_sum in this case) isn't actually a separately
>>>> allocated Datum, but rather just something embedded in a larger
>>>> struct.  That, combined with the following code:
>>>>         if (!peraggstate->resulttypeByVal && !*isnull &&
>>>>                 !MemoryContextContains(CurrentMemoryContext,
>>>>                                                            DatumGetPointer(*result)))
>>>> seems somewhat problematic to me.  MemoryContextContains() can give
>>>> false positives when used on memory that's not a distinctly allocated
>>>> chunk, and if so, we violate memory lifetime rules.  It's quite
>>>> unlikely, given the required bit patterns, but nonetheless it's making
>>>> me somewhat uncomfortable.
>>>>
>>>> Do others think this isn't an issue and we can just live with it?
>>>
>>> I think it's 100% broken to call MemoryContextContains() on something
>>> that's not guaranteed to be a palloc'd chunk.
>>
>> I agree, but to me it seems the only fix would be to just yank out the
>> whole optimization?
>
> Dunno, haven't looked into it.
>

I think it might be fixable by adding a flag into the chunk, with 'true' 
for regular allocations, and 'false' for the optimized ones. And then 
only use MemoryContextContains() for 'flag=true' chunks.

The question however is whether this won't make the optimization 
pointless. I also, wonder how much we save by this optimization and how 
widely it's used? Can someone point me to some numbers?

regards

-- 
Tomas Vondra                  http://www.2ndQuadrant.com
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services


-- 
Sent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-hackers