Re: Wrong estimation of rows for hash join - Mailing list pgsql-general

From Tom Lane
Subject Re: Wrong estimation of rows for hash join
Date
Msg-id 6204.1255704148@sss.pgh.pa.us
Whole thread Raw
In response to Re: Wrong estimation of rows for hash join  (Alban Hertroys <dalroi@solfertje.student.utwente.nl>)
List pgsql-general
Alban Hertroys <dalroi@solfertje.student.utwente.nl> writes:
> I'm also somewhat surprised to see an array of what appear to be
> integers be cast to bpchar[]. Did you define those coordinates(?) as
> character types? Numerical comparisons tend to be faster than string
> comparisons, which should make some difference on sequential scans.

Or, if the column can't be changed to an integer, at least consider
making it varchar not char.  The funny rules about trailing blanks
make char comparison a bit slower than varchar, IIRC.  They aren't
very conducive to fast hashing either ...

            regards, tom lane

pgsql-general by date:

Previous
From: Tom Lane
Date:
Subject: Re: savepoint name vs prepared transaction name
Next
From: Grzegorz Jaśkiewicz
Date:
Subject: Re: savepoint name vs prepared transaction name