From: Tom Lane <tgl@sss.pgh.pa.us>
To: Andres Freund <andres@anarazel.de>
cc: pgsql-hackers@lists.postgresql.org, Hadi Moshayedi <hadi@moshayedi.net>
Subject: Re: tableam scan-API patch broke foreign key validation
In-reply-to: <20190406182258.pvk5zkqdc6dobfg7@alap3.anarazel.de>
References: <19030.1554574075@sss.pgh.pa.us>
 <FD203188-892D-4FD6-99F4-7E85027B60A7@anarazel.de>
 <19246.1554574409@sss.pgh.pa.us>
 <20190406182258.pvk5zkqdc6dobfg7@alap3.anarazel.de>
Comments: In-reply-to Andres Freund <andres@anarazel.de>
	message dated "Sat, 06 Apr 2019 11:22:58 -0700"
MIME-Version: 1.0
Content-Type: text/plain; charset="us-ascii"
Content-ID: <19905.1554575674.1@sss.pgh.pa.us>
Content-Transfer-Encoding: quoted-printable
Date: Sat, 06 Apr 2019 14:34:34 -0400
Message-ID: <19906.1554575674@sss.pgh.pa.us>
Precedence: bulk

Andres Freund <andres@anarazel.de> writes:
> The relevant thread is:
> https://www.postgresql.org/message-id/20190325180405.jytoehuzkeozggxx%40=
alap3.anarazel.de

Yeah, I just found that --- would have seen it sooner if David had
not elected to make it a new thread.

> Wonder if you have an opinion on:

>> I've also noticed that we should free the tuple - that doesn't matter
>> for heap, but it sure does for other callers.

Why should this code need to free anything?  That'd be the responsibility
of the slot code, no?

>> But uh, is it actually ok
>> to validate an entire table's worth of foreign keys without a memory
>> context reset? I.e. shouldn't we have a memory context that we reset
>> after each iteration?
>> Also, why's there no CHECK_FOR_INTERRUPTS()? heap has some internally o=
n
>> a page level, but that doesn't seem all that granular?

These are good questions.  Just eyeing RI_FKey_check(), I think
that it might not have any significant leaks because most of the work
is done in an SPI context, but obviously that's pretty fragile.

The memory-context stuff in your WIP patch seems wrong, btw;
the second or later iteration of the loop would trash oldcxt.

But clearly we need a test case here.  I'll adjust Hadi's example
so that there's more than one tuple to check, and push it.

			regards, tom lane