Subject: Re: backend crash on DELETE, reproducible locally
To: Andres Freund <andres@anarazel.de>, Tom Lane <tgl@sss.pgh.pa.us>
Cc: =?UTF-8?Q?Ond=c5=99ej_Bouda?= <obouda@email.cz>,
 Alvaro Herrera <alvherre@2ndquadrant.com>,
 pgsql-hackers@lists.postgresql.org, pgsql-general@lists.postgresql.org
References: <20181106194553.dynhd2xuivo56ymd@alvherre.pgsql>
 <33d144b9-5208-8d32-7e95-94ed4d2931b8@email.cz>
 <450.1541535425@sss.pgh.pa.us>
 <134c09b5-3eb0-bafd-2aeb-f383e1e21b8a@email.cz>
 <2432.1541538215@sss.pgh.pa.us> <2877.1541538838@sss.pgh.pa.us>
 <fc1fde8f-2e1d-3f8d-a583-82ba383c7fd2@email.cz>
 <4375.1541540840@sss.pgh.pa.us>
 <20181106215457.h36ccb5wywatcqhz@alap3.anarazel.de>
From: Tomas Vondra <tomas.vondra@2ndquadrant.com>
Message-ID: <b331a23e-8e34-b2f2-3a2b-f1e1c770d3b6@2ndquadrant.com>
Date: Tue, 6 Nov 2018 23:11:29 +0100
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:60.0) Gecko/20100101
 Thunderbird/60.2.1
MIME-Version: 1.0
In-Reply-To: <20181106215457.h36ccb5wywatcqhz@alap3.anarazel.de>
Content-Type: text/plain; charset=utf-8
Content-Language: en-US
Content-Transfer-Encoding: 8bit
Precedence: bulk

On 11/6/18 10:54 PM, Andres Freund wrote:
> On 2018-11-06 16:47:20 -0500, Tom Lane wrote:
>> =?UTF-8?Q?Ond=c5=99ej_Bouda?= <obouda@email.cz> writes:
>>>> Ondřej, as a short-term workaround you could prevent the crash
>>>> by setting that index's recheck_on_update property to false.
>>
>>> Thanks for the tip. I am unsuccessful using it, though:
>>> # ALTER INDEX public.schedulecard_overlap_idx SET (recheck_on_update = 
>>> FALSE);
>>> ERROR:  unrecognized parameter "recheck_on_update"
>>
>> Oh, for crying out loud.  That's yet a different bug.
>> I'm not sure that it's the fault of the recheck_on_update
>> feature proper though; it might be a pre-existing bug in
>> the reloptions code.  Looks like somebody forgot to list
>> RELOPT_KIND_GIST in RELOPT_KIND_INDEX, but is that the
>> fault of commit c203d6cf8 or was it busted before?
> 
> Looks new:
> +   RELOPT_KIND_INDEX = RELOPT_KIND_BTREE|RELOPT_KIND_HASH|RELOPT_KIND_GIN|RELOPT_KIND_SPGIST,
> 
> there aren't any other "for all indexes" type options, so the whole
> category didn't exist before.
> 
> It also strikes me as a really bad idea, even if RELOPT_KIND_GIST
> wouldn't have been omitted: It breaks index am extensibility.
> 

Does it? The RELOPT_KIND_* stuff is hard-coded in reloptions.h anyway,
so I'm not sure how this particular thing makes it less extensible?

That being said, we also have RELOPT_KIND_BRIN, and that seems to be
missing from RELOPT_KIND_INDEX too (and AFAICS the optimization works
for all index types).

regards

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