Re: How to tell ANALYZE to collect statistics from the whole table?

From: Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us>
To: AlexK987 <alex(dot)cue(dot)987(at)gmail(dot)com>
Cc: pgsql-performance(at)postgresql(dot)org
Subject: Re: How to tell ANALYZE to collect statistics from the whole table?
Date: 2015-01-25 05:14:41
Message-ID: 18230.1422162881@sss.pgh.pa.us
Views: Raw Message | Whole Thread | Download mbox | Resend email
Thread:
Lists: pgsql-performance

AlexK987 <alex(dot)cue(dot)987(at)gmail(dot)com> writes:
> The documentation states that "The extent of analysis can be controlled by
> adjusting the default_statistics_target configuration variable". It looks
> like I can tell Postgres to create more histograms with more bins, and more
> distinct values. This implicitly means that Postgres will use a larger
> random subset to calculate statistics.

> However, this is not what I want. My data may be quite skewed, and I want
> full control over the size of the sample. I want to explicitly tell Postgres
> to analyze the whole table. How can I accomplish that?

You can't, and you wouldn't want to if you could, because that would
result in slurping the entire table into backend local memory. All
the rows constituting the "random sample" are held in memory while
doing the statistical calculations.

In practice, the only stat that would be materially improved by taking
enormously large samples would be the number-of-distinct-values estimate.
There's already a way you can override ANALYZE's estimate of that number
if you need to.

regards, tom lane

In response to

Responses

Browse pgsql-performance by date

  From Date Subject
Next Message Joe Van Dyk 2015-01-25 05:41:13 Query performance
Previous Message Tomas Vondra 2015-01-25 01:29:27 Re: Initial insert