Date: Thu, 16 Jan 2020 21:00:38 +0100
From: Tomas Vondra <tomas.vondra@2ndquadrant.com>
To: Tom Lane <tgl@sss.pgh.pa.us>
Cc: Andres Freund <andres@anarazel.de>, Tomas Vondra <tv@fuzzy.cz>,
	pgsql-hackers@postgresql.org
Subject: Re: SlabCheck leaks memory into TopMemoryContext
Message-ID: <20200116200038.rfqfnqsrjcdv5kwk@development>
References: <20200116044119.g45f7pmgz4jmodxj@alap3.anarazel.de>
 <23952.1579151393@sss.pgh.pa.us>
 <20200116061732.vvpyexenaortpu5y@alap3.anarazel.de>
 <27213.1579155900@sss.pgh.pa.us>
 <20200116064143.opcu2vhcktddpzbh@alap3.anarazel.de>
 <19743.1579188421@sss.pgh.pa.us>
 <20200116162500.pcgxmykirm677mt6@development>
 <20200116164849.75rzho6zn4qppnf3@alap3.anarazel.de>
 <20200116170153.qp36chsyhikvgvxt@development>
 <2646.1579195983@sss.pgh.pa.us>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii; format=flowed
Content-Disposition: inline
In-Reply-To: <2646.1579195983@sss.pgh.pa.us>
Precedence: bulk

On Thu, Jan 16, 2020 at 12:33:03PM -0500, Tom Lane wrote:
>Tomas Vondra <tomas.vondra@2ndquadrant.com> writes:
>> On Thu, Jan 16, 2020 at 08:48:49AM -0800, Andres Freund wrote:
>>> I don't get why it's advantageous to allocate this once for each slab,
>>> rather than having it as a global once for all slabs. But anyway, still
>>> clearly better than the current situation.
>
>> It's largely a matter of personal preference - I agree there are cases
>> when global variables are the best solution, but I kinda dislike them.
>> It seems cleaner to just allocate it as part of the slab, not having to
>> deal with different number of chunks per block between slabs.
>> Plus we don't have all that many slabs (like 2), and it's only really
>> used in debug builds anyway. So I'm not all that woried about this
>> wasting a couple extra kB of memory.
>
>A positive argument for doing it like this is that the overhead goes away
>when the SlabContexts are all deallocated, while a global variable would
>presumably stick around indefinitely.  But I concur that in current usage,
>there's hardly any point in worrying about the relative benefits.  We
>should just keep it simple, and this seems marginally simpler than the
>other way.
>

I think the one possible argument against this approach might be that it
adds a field to the struct, so if you have an extension using a Slab
context, it'll break if you don't rebuild it. But that only matters if
we want to backpatch it (which I think is not the plan) and with memory
context checking enabled (which does not apply to regular packages).


regards

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