Re: Use extended statistics to estimate (Var op Var) clauses - Mailing list pgsql-hackers

From Tomas Vondra
Subject Re: Use extended statistics to estimate (Var op Var) clauses
Date
Msg-id 5b045793-f412-abca-45c5-4cffadbf54f3@enterprisedb.com
Whole thread Raw
In response to Re: Use extended statistics to estimate (Var op Var) clauses  (Tomas Vondra <tomas.vondra@enterprisedb.com>)
List pgsql-hackers
On 8/18/21 12:43 PM, Tomas Vondra wrote:
> Hi Mark,
> 
> This thread inspired me to do something fairly similar - a generator 
> that generates queries of varying complexity, executes them on table 
> with and without extended statistics. I've been thinking about that 
> before, but this finally pushed me to do that, and some of the results 
> are fairly interesting ...
> 
> I've pushed everything (generator and results) to this github repo:
> 
>    https://github.com/tvondra/stats-test
> 
> with a summary of all results here:
> 
>    https://github.com/tvondra/stats-test/blob/master/results.md
> 

FWIW I've pushed slightly reworked scripts and results - there are 
results from two machines - xeon and i5. Xeon is mostly the same as 
before, with some minor fixes, while i5 is does not allow clauses 
referencing the same column twice (per discussion in this thread).

I think there was a bug in the original plot script, combining incorrect 
data series in some cases, causing (at least) some of the strange 
patterns mentioned.

I've also made the charts easier to read by splitting the cases into 
separate plots and using transparency. I've also added png version back, 
because plotting the .svg is quite slow.


regards

-- 
Tomas Vondra
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company



pgsql-hackers by date:

Previous
From: Robert Haas
Date:
Subject: Re: The Free Space Map: Problems and Opportunities
Next
From: Justin Pryzby
Date:
Subject: Re: pgstat_send_connstats() introduces unnecessary timestamp and UDP overhead