From: Tom Lane <tgl@sss.pgh.pa.us>
To: Robert Haas <robertmhaas@gmail.com>
cc: Kyotaro HORIGUCHI <horiguchi.kyotaro@lab.ntt.co.jp>,
        Andres Freund <andres@anarazel.de>,
        Michael Paquier <michael@paquier.xyz>,
        Justin Pryzby <pryzby@telsasoft.com>,
        "pgsql-hackers@postgresql.org" <pgsql-hackers@postgresql.org>
Subject: Re: ENOSPC FailedAssertion("!(RefCountErrors == 0)"
In-reply-to: 
 <CA+TgmobXLXNeyod0wrmFp1xeusJJtTCzZPizqwHsfr=hVKS2DA@mail.gmail.com>
References: <20436.1531751966@sss.pgh.pa.us> <4993.1531773411@sss.pgh.pa.us>
 <14892.1531872065@sss.pgh.pa.us>
 <20180718.124220.08087886.horiguchi.kyotaro@lab.ntt.co.jp>
 <10690.1531924381@sss.pgh.pa.us>
 <CA+TgmobXLXNeyod0wrmFp1xeusJJtTCzZPizqwHsfr=hVKS2DA@mail.gmail.com>
Comments: In-reply-to Robert Haas <robertmhaas@gmail.com>
	message dated "Wed, 18 Jul 2018 13:03:20 -0400"
MIME-Version: 1.0
Content-Type: text/plain; charset="us-ascii"
Content-ID: <7341.1531937117.1@sss.pgh.pa.us>
Date: Wed, 18 Jul 2018 14:05:17 -0400
Message-ID: <7342.1531937117@sss.pgh.pa.us>
Precedence: bulk

Robert Haas <robertmhaas@gmail.com> writes:
> On Wed, Jul 18, 2018 at 10:33 AM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
>>> So couldn't we use TopTransactionResourceOwner instead of
>>> AuxProcessResrouceOwner? I feel a bit uneasy that bootstrap and
>>> standalone-backend have *AuxProcess*ResourceOwner.

>> Since the aux processes aren't running transactions, I didn't think
>> that TopTransactionResourceOwner was appropriate.  There's also
>> a problem for bootstrap and standalone backend cases: those do run
>> transactions and therefore create/destroy TopTransactionResourceOwner,
>> leaving nothing behind for ShutdownXLOG to use if it tries to use
>> that.  We need an extra resowner somewhere.

> FallbackResourceOwner?  DefaultResourceOwner? SessionResourceOwner?

Those names all suggest (to me anyway) that this resowner exists in
all, or at least most, processes.  That's not the situation as of this
patch, although I could imagine an alternate universe where it's true;
for example, if we decided there were a reason for normal backends to
have a session-lifespan resowner.  But even then, it might be better
to distinguish that from aux processes' use of resowners.

(I'm not really convinced that it'd be a good idea for normal backends
to have a session-lifespan resowner; that could mask bugs involving
trying to acquire resources outside a transaction.)

			regards, tom lane