Re: Revisiting pg_stat_statements and IN() (Was: Re: pg_stat_statements fingerprinting logic and ArrayExpr)

On Wed, Nov 25, 2015 at 9:13 AM, Lukas Fittl <lukas(at)fittl(dot)com> wrote:

> On Mon, Nov 23, 2015 at 11:53 PM, Peter Geoghegan <pg(at)heroku(dot)com> wrote:
>
>> One specific justification he gave for not using pg_stat_statements was:
>>
>> "Doesn’t merge bind vars in IN()" (See slide #11)
>>
>> I wonder:
>>
>> * How do other people feel about this? Personally, I've seen enough
>> problems of this kind in the field that "slippery slope" arguments
>> against this don't seem very compelling.
>>
>
> As someone who runs a little monitoring service thats solely based on
> pg_stat_statements data, ignoring IN list length would certainly be a good
> change.
>
> We currently do this in post-processing, together with other data cleanup
> (e.g. ignoring the length of a VALUES list in an INSERT statement).
>
> Given the fact that pgss data is normalized & you don't know which plan
> was chosen, I don't think distinguishing based on the length of the list
> helps anyone really.
>
> I see pg_stat_statements as a high-level overview of which queries have
> run, and which ones you might want to look into closer using e.g.
> auto_explain.
>

I still have the plans to try to marry pg_stat_statements and auto_explain
for the next iteration of "online query plans" extension I was proposing a
few months ago, and the first thing I was going to look into is rectifying
this problem with IN() jumbling. So, have a +1 from me.

--
Alex