Date: Sun, 12 Jan 2020 05:10:08 +0100
From: Tomas Vondra <tomas.vondra@2ndquadrant.com>
To: Tom Lane <tgl@sss.pgh.pa.us>
Cc: Amit Kapila <amit.kapila16@gmail.com>, Noah Misch <noah@leadboat.com>,
	Amit Khandekar <amitdkhan.pg@gmail.com>,
	Alvaro Herrera from 2ndQuadrant <alvherre@alvh.no-ip.org>,
	Andres Freund <andres@anarazel.de>,
	Juan =?utf-8?B?Sm9zw6kgU2FudGFtYXLDrWE=?= Flecha
 <juanjo.santamaria@gmail.com>,
	PostgreSQL Hackers <pgsql-hackers@lists.postgresql.org>,
	Robert Haas <robertmhaas@gmail.com>,
	Thomas Munro <thomas.munro@gmail.com>
Subject: Re: logical decoding : exceeded maxAllocatedDescs for .spill files
Message-ID: <20200112041008.gddmpqunowlklc24@development>
References: 
 <CAJ3gD9dFYJnOxUsR3MeJ=3hJKZk5XqAx-6fYUNXt6jJOK3QFMQ@mail.gmail.com>
 <20200104185148.GA2270238@rfd.leadboat.com>
 <CAJ3gD9diTB8rqyUWo=jZrSQxvW7o-gGaO6uoF5racneH3+FYOg@mail.gmail.com>
 <20200109053704.GA2502006@rfd.leadboat.com>
 <22498.1578548741@sss.pgh.pa.us>
 <CAA4eK1JitJgbOv-ie3VCXW5A74j-v0b8S55YMunwmkFDZ_4NCg@mail.gmail.com>
 <14739.1578613901@sss.pgh.pa.us>
 <18117.1578616812@sss.pgh.pa.us>
 <20200112033930.6cy3n6ybft6pp7gw@development>
 <29690.1578801237@sss.pgh.pa.us>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii; format=flowed
Content-Disposition: inline
In-Reply-To: <29690.1578801237@sss.pgh.pa.us>
Precedence: bulk

On Sat, Jan 11, 2020 at 10:53:57PM -0500, Tom Lane wrote:
>Tomas Vondra <tomas.vondra@2ndquadrant.com> writes:
>> On Thu, Jan 09, 2020 at 07:40:12PM -0500, Tom Lane wrote:
>>> It seems reasonably likely to me that this result is telling us about
>>> an actual bug, ie, faulty back-patching of one or more of those fixes
>>> into v10 and perhaps earlier branches.
>
>> Well, one thing we did in 11 is introduction of the Generation context.
>> In 10 we're still stashing all tuple data into the main AllocSet. I
>> wonder if backporting a4ccc1cef5a04cc054af83bc4582a045d5232cb3 and a
>> couple of follow-up fixes would make the issue go away.
>
>Hm.  I'm loath to back-port Generation contexts.

Yeah, I agree. My suggestion was to try backpatching it and see if it
resolves the issue.

>But looking at
>a4ccc1cef5a04cc054af83bc4582a045d5232cb3, I see that (a) the
>commit message mentions space savings, but (b) the replaced code
>in reorderbuffer.c doesn't look like it really would move the needle
>much in that regard.  The old code had a one-off slab allocator
>that we got rid of, but I don't see any actual leak there ...
>remind me where the win came from, exactly?
>

Well, the problem is that in 10 we allocate tuple data in the main
memory ReorderBuffer context, and when the transaction gets decoded we
pfree() it. But in AllocSet that only moves the data to the freelists,
it does not release it entirely. So with the right allocation pattern
(sufficiently diverse chunk sizes) this can easily result in allocation
of large amount of memory that is never released.

I don't know if this is what's happening in this particular test, but I
wouldn't be surprised by it.


regards

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