Message-ID: <5B46CBB3.2090300@lab.ntt.co.jp>
Date: Thu, 12 Jul 2018 12:32:03 +0900
From: Etsuro Fujita <fujita.etsuro@lab.ntt.co.jp>
User-Agent: Mozilla/5.0 (Windows NT 6.1; WOW64;
 rv:11.0) Gecko/20120327 Thunderbird/11.0.1
MIME-Version: 1.0
Subject: Re: Expression errors with "FOR UPDATE" and postgres_fdw with
 partition
 wise join enabled.
References: 
 <CAKcux6ktu-8tefLWtQuuZBYFaZA83vUzuRd7c1YHC-yEWyYFpg@mail.gmail.com>
 <5B1E5071.1000704@lab.ntt.co.jp> <5B23A964.9020800@lab.ntt.co.jp>
 <5B28FB19.3090809@lab.ntt.co.jp>
 <CAFjFpRedQBySj5Vm-eq+gATP6sqJJfXzPrV+Ld+K6DnaOGkFAQ@mail.gmail.com>
 <5B362B38.2000103@lab.ntt.co.jp>
 <CAFjFpRf55MT-9NnaRwp0wkDUTuwb++UA_un9Twai=qNVePh-TQ@mail.gmail.com>
 <5B3CB833.4040005@lab.ntt.co.jp>
 <CAFjFpRevODMmZ8MF3oLsXr9mSH9-mWC_fxmPWTYCSqHVr_Zwqg@mail.gmail.com>
 <5B3F4BAA.3050601@lab.ntt.co.jp>
 <CAFjFpRf2vd0C6B8g1TmpJN8RkPECgZy8iGe9UYZTy9EWu1eL2Q@mail.gmail.com>
 <5B4340E9.3030707@lab.ntt.co.jp>
 <CAFjFpRcwrESJBDkmm1kvtV234nkw=dr9nJ=RWJ4pmsVQqF4avg@mail.gmail.com>
 <5B434886.9010107@lab.ntt.co.jp>
 <CAFjFpRf-t_dM=qirB5+95wT1hSCLs8mQqbfRSKP=xUuhnOhbOQ@mail.gmail.com>
 <5B442A40.2040903@lab.ntt.co.jp>
 <CAFjFpRfxsg3dXEDQPdUD+HYqE6v=jEYFiEtEgXuF+q+MVpA8eg@mail.gmail.com>
 <5B45B777.7010502@lab.ntt.co.jp>
 <CAFjFpRf-n75_mN=EFAvY-YY1xd_AXEDUdedW7sDX5poxpqs7fA@mail.gmail.com>
In-Reply-To: 
 <CAFjFpRf-n75_mN=EFAvY-YY1xd_AXEDUdedW7sDX5poxpqs7fA@mail.gmail.com>
Content-Type: text/plain; charset=UTF-8; format=flowed
Content-Transfer-Encoding: 7bit
To: Ashutosh Bapat <ashutosh.bapat@enterprisedb.com>
Cc: Rajkumar Raghuwanshi <rajkumar.raghuwanshi@enterprisedb.com>,
        pgsql-hackers <pgsql-hackers@postgresql.org>
Precedence: bulk

(2018/07/11 20:02), Ashutosh Bapat wrote:
> On Wed, Jul 11, 2018 at 1:23 PM, Etsuro Fujita
> <fujita.etsuro@lab.ntt.co.jp>  wrote:
>> Actually, even if we could create such an index on the child table and the
>> targetlist had the ConvertRowtypeExpr, the planner would still not be able
>> to use an index-only scan with that index; because check_index_only would
>> not consider that an index-only scan is possible for that index because that
>> index is an expression index and that function currently does not consider
>> that index expressions are able to be returned back in an index-only scan.
>> That behavior of the planner might be improved in future, though.

> Right and when we do so, not having ConvertRowtypeExpr in the
> targetlist will be a problem.

Yeah, but I don't think that that's unsolvable; because in that case the 
CRE as an index expression could be converted back to the whole-row 
Var's rowtype by adding another CRE to the index expression for that 
conversion, I suspect that that special handling could allow us to 
support an index-only scan even when having the whole-row Var instead of 
the CRE in the targetlist.  (Having said that, I'm not 100% sure we need 
to solve that problem when we improve the planner, because there doesn't 
seem to me to be enough use-case to justify making the code complicated 
for that.)  Anyway, I think that that would be a matter of future 
versions of PG.

>>> At places in planner we match equivalence members
>>> to the targetlist entries. This matching will fail unexpectedly when
>>> ConvertRowtypeExpr is removed from a child's targetlist. But again I
>>> couldn't reproduce a problem when such a mismatch arises.
>>
>>
>> IIUC, I don't think the planner assumes that for an equivalence member there
>> is an matching entry for that member in the targetlist; what I think the
>> planner assumes is: an equivalence member is able to be computed from
>> expressions in the targetlist.
>
> This is true. However,
>
>>   So, I think it is safe to have whole-row
>> Vars instead of ConvertRowtypeExprs in the targetlist.
>
> when it's looking for an expression, it finds a whole-row expression
> so it think it needs to add a ConvertRowtypeExpr on that. But when the
> plan is created, there is ConvertRowtypeExpr already, but there is no
> way to know that a new ConvertRowtypeExpr is not needed anymore. So,
> we may have two ConvertRowtypeExprs giving wrong results.

Maybe I'm missing something, but I don't think that we need to worry 
about that, because in the approach I proposed, we only add CREs above 
whole-row Vars in the targetlists for subplans of an Append/MergeAppend 
for a partitioned relation at plan creation time.

Best regards,
Etsuro Fujita