MIME-Version: 1.0
References: <CAA5RZ0uVOGBYmPEeGF2d1B_67tgNjKx_bKDuL+oUftuoz+=Y1g@mail.gmail.com>
 <CAGjGUAJEhvsjni3q5y0iGAptb6s5UpxKJiLj3A_1CUwGdrujdA@mail.gmail.com>
 <594fd111-8016-4f44-bc8c-e1a0d66f26f6@postgrespro.ru> <CAA5RZ0tBsAa-CY=-z_VPjFvMRWnKLzkt0QAUQxUx+CO1-YiS8g@mail.gmail.com>
In-Reply-To: <CAA5RZ0tBsAa-CY=-z_VPjFvMRWnKLzkt0QAUQxUx+CO1-YiS8g@mail.gmail.com>
From: Sami Imseih <samimseih@gmail.com>
Date: Fri, 10 Jan 2025 10:21:56 -0600
Message-ID: <CAA5RZ0uVgHa2JLSEsq8jDMRdUmyfhz9gg0UHhx0g2zoB7h_faA@mail.gmail.com>
Subject: Re: POC: track vacuum/analyze cumulative time per relation
To: Alena Rybakina <a.rybakina@postgrespro.ru>
Cc: wenhui qiu <qiuwenhuifx@gmail.com>, pgsql-hackers <pgsql-hackers@postgresql.org>
Content-Type: text/plain; charset="UTF-8"
Archived-At: <https://www.postgresql.org/message-id/CAA5RZ0uVgHa2JLSEsq8jDMRdUmyfhz9gg0UHhx0g2zoB7h_faA%40mail.gmail.com>
Precedence: bulk

> > this information gives a complete picture of vacuum efficiency, because
> > analyzing only total_time statistics does not give a complete picture of
> > what vacuum did: was it cleaning almost huge index, cleaning tables or
> > just sleeping.
>
> The purpose of total_time is to be able to calculate the average
> time of a vacuum per relation from pg_stat_all_tables (by
> total_time/vacuum_count) This is useful on its own to be able to
> trend slow downs of vacuum ( or improvements after vacuum tuning )
> for specific relations.
>
> The work going into [1] is more detailed and will be useful
> for deep dive investigations.In short, I think both proposals
> have merit.

Missed this point in my last reply. This patch also tracks
(auto)analyze per relation cumulative timing which can
be useful along with (auto)analyze_count to derive
the average.

Regards,

Sami