Обсуждение: CTE optimization fence

Hello,
I am running into performance issues with large CTE "WITH" queries (just 
for selecting, not updating).  I was surprised to find that the queries 
run much faster if I convert the CTEs to subqueries. From googling, I 
see that this is due to CTE acting as an optimization fence in PG. 
Unfortunately due to the application I'm dealing with, converting all 
CTE queries to subquery model is not feasible. Plus, the readability of 
CTE is a big bonus.

I see there was some discussion last year about removing the CTE 
optimization fence (e.g. 
http://www.postgresql-archive.org/CTE-inlining-td5958992.html) but can't 
find anything more recent. Does anyone know if this is still under 
consideration? For what it's worth, I would love some way to make CTEs 
inlined/optimized.

Thank you very much to the developers for a truly amazing database system.

Thanks
Guy

Guy Burgess <guy@burgess.co.nz> writes:
> I see there was some discussion last year about removing the CTE 
> optimization fence (e.g. 
> http://www.postgresql-archive.org/CTE-inlining-td5958992.html) but can't 
> find anything more recent. Does anyone know if this is still under 
> consideration?

Nothing's actually happened since then ... it's still under consideration,
but we have to settle on a way of controlling it.

            regards, tom lane

Tom Lane schrieb am 27.06.2018 um 05:48:
>> I see there was some discussion last year about removing the CTE 
>> optimization fence (e.g. 
>> http://www.postgresql-archive.org/CTE-inlining-td5958992.html) but can't 
>> find anything more recent. Does anyone know if this is still under 
>> consideration?
> 
> but we have to settle on a way of controlling it.

+1 from me. 

I am running more and more into situations where people consider this a bug rather than a feature.

FWIW, I think a GUC that switches between the current (mostly unwanted, at least surprising) 
way and one where the CTE is optimized together with the main query would suit "most" people.

For sake of compatibility this could default to the current behaviour

I don’t think the fence analogy works. Try wall (a la Berlin).
I count myself amongst those who thought “Ooh this little CTE will garner just what the rest of the query needs”. Only
tofind the planner didn’t groc that optimization.  

Is it a matter of deciding to trust the SQLer and runtime analyzing the CTE product before continuing? As an SQLer I
havedoubts about that precept but without _some_ change in attack CTEs approach fluff.  

> On Jun 26, 2018, at 11:45 PM, Thomas Kellerer <spam_eater@gmx.net> wrote:
>
> Tom Lane schrieb am 27.06.2018 um 05:48:
>>> I see there was some discussion last year about removing the CTE
>>> optimization fence (e.g.
>>> http://www.postgresql-archive.org/CTE-inlining-td5958992.html) but can't
>>> find anything more recent. Does anyone know if this is still under
>>> consideration?
>>
>> but we have to settle on a way of controlling it.
>
> +1 from me.
>
> I am running more and more into situations where people consider this a bug rather than a feature.
>
> FWIW, I think a GUC that switches between the current (mostly unwanted, at least surprising)
> way and one where the CTE is optimized together with the main query would suit "most" people.
>
> For sake of compatibility this could default to the current behaviour
>

Tom Lane schrieb am 27.06.2018 um 05:48:
>> I see there was some discussion last year about removing the CTE
>> optimization fence (e.g.
>> http://www.postgresql-archive.org/CTE-inlining-td5958992.html) but can't
>> find anything more recent. Does anyone know if this is still under
>> consideration?
>
> but we have to settle on a way of controlling it.

+1 from me.

I am running more and more into situations where people consider this a bug rather than a feature.

FWIW, I think a GUC that switches between the current (mostly unwanted, at least surprising)
way and one where the CTE is optimized together with the main query would suit "most" people.

For sake of compatibility this could default to the current behaviour

--

Andreas Joseph Krogh

CTO / Partner - Visena AS

Mobile: +47 909 56 963

andreas@visena.com

www.visena.com

On 06/27/2018 09:58 AM, Andreas Joseph Krogh wrote:
>      >
>      > but we have to settle on a way of controlling it.
> 
>     +1 from me.
> 
>     I am running more and more into situations where people consider
>     this a bug rather than a feature.
> 
>     FWIW, I think a GUC that switches between the current (mostly
>     unwanted, at least surprising)
>     way and one where the CTE is optimized together with the main query
>     would suit "most" people.
> 
>     For sake of compatibility this could default to the current behaviour
> 
> +1 from me. The default should be "no fence" for sake of least surprise 
> I think. Documenting the change would be sufficient.
> I hope this will be picked up in the comming V12-cycle.


FYI this subject has been discussed in this thread : 
https://www.postgresql.org/message-id/5351711493487900%40web53g.yandex.ru

Regards,

On 06/27/2018 09:58 AM, Andreas Joseph Krogh wrote:
>      >
>      > but we have to settle on a way of controlling it.
>
>     +1 from me.
>
>     I am running more and more into situations where people consider
>     this a bug rather than a feature.
>
>     FWIW, I think a GUC that switches between the current (mostly
>     unwanted, at least surprising)
>     way and one where the CTE is optimized together with the main query
>     would suit "most" people.
>
>     For sake of compatibility this could default to the current behaviour
>
> +1 from me. The default should be "no fence" for sake of least surprise
> I think. Documenting the change would be sufficient.
> I hope this will be picked up in the comming V12-cycle.


FYI this subject has been discussed in this thread :
https://www.postgresql.org/message-id/5351711493487900%40web53g.yandex.ru

Regards,
 

--

Andreas Joseph Krogh

CTO / Partner - Visena AS

Mobile: +47 909 56 963

andreas@visena.com

www.visena.com

On 2018-Jun-27, Andreas Joseph Krogh wrote:

> I know. I hate the INLINE proposal and hope default-behaviour will be like in 
> other DBs, inline like sub-query as default. GUC for preserving fence is what I 
> hope will happen.

I am probably biased ;-) but I like my proposal #1 in this email
https://www.postgresql.org/message-id/20170503173305.fetj4tz7kd56tjlr%40alvherre.pgsql
where you have to mark queries were the current behavior is desired
(WITH MATERIALIZED), and leave all other queries alone.  I don't
remember any dissent on that.

-- 
Álvaro Herrera                https://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services