Re: Guarantees/Semantics of pg_stats

On Sat, 2024-03-02 at 07:41 +0000, Baziotis, Stefanos wrote:
> I'm interested in learning more about the guarantees/semantics of pg_stats.
> For example, is there a guarantee that the n_distinct and most_common_vals
> fields will take into account any values appearing more than M times or
> maybe with frequence more than f? In what cases will n_distinct and
> most_common_vals will miss some values?

Table Statistics are not exact.  They are collected from a random sample of
the data, so they are never guaranteed to be exact.

Their purpose is to estimate the result row count and cost of execution plan
steps.  You can never use them as proof.

Yours,
Laurenz Albe