Em 19/09/2019 15:34, Igor Neyman escreveu:
> -----Original Message-----
> From: Luís Roberto Weck [mailto:luisroberto(at)siscobra(dot)com(dot)br]
> Sent: Thursday, September 19, 2019 2:30 PM
> To: Michael Lewis <mlewis(at)entrata(dot)com>
> Cc: pgsql-performance(at)lists(dot)postgresql(dot)org
> Subject: Re: Slow query on a one-tuple table
>
> WARNING: This email originated from outside of Perceptron! Please be mindful of PHISHING and MALWARE risks.
>
> Em 19/09/2019 14:21, Michael Lewis escreveu:
>> Is this result able to be repeated?
> Yes, I can consistently repeat it.
>
> Postgres version is 11.1.
>
> Other executions:
>
> Index Scan using assessoria_pkey on public.assessoria (cost=0.25..2.47
> rows=1 width=62) (actual time=1.591..4.035 rows=1 loops=1)
> Output: asscod, asscambol
> Index Cond: (assessoria.asscod = 1)
> Buffers: shared hit=1187
> Planning Time: 0.053 ms
> Execution Time: 4.055 ms
>
> Index Scan using assessoria_pkey on public.assessoria (cost=0.25..2.47
> rows=1 width=62) (actual time=1.369..3.838 rows=1 loops=1)
> Output: asscod, asscambol
> Index Cond: (assessoria.asscod = 1)
> Buffers: shared hit=1187
> Planning Time: 0.033 ms
> Execution Time: 3.851 ms
>
> ________________________________________________________________________________________________________________
>
> But can you repeat it with "LIMIT 1"?
> Notice huge difference in "buffers hit" while doing (the same) Index Scan in two plans.
>
> Regards,
> Igor Neyman
With LIMIT 1, I get 3 shared buffers hit, pretty much always.