Re: show "aggressive" or not in autovacuum logs - Mailing list pgsql-hackers

From Masahiko Sawada
Subject Re: show "aggressive" or not in autovacuum logs
Date
Msg-id CAD21AoBiw96efy+tynvMLFQWERfPnhO53B=XFW9yyzEJN-fNaA@mail.gmail.com
Whole thread Raw
In response to Re: show "aggressive" or not in autovacuum logs  (Kyotaro HORIGUCHI <horiguchi.kyotaro@lab.ntt.co.jp>)
List pgsql-hackers
On Tue, Apr 4, 2017 at 10:09 AM, Kyotaro HORIGUCHI
<horiguchi.kyotaro@lab.ntt.co.jp> wrote:
> Hello,
>
> At Fri, 31 Mar 2017 18:20:23 +0900, Masahiko Sawada <sawada.mshk@gmail.com> wrote in
<CAD21AoBN-Hsih7HEP8ey6NUwQgwYj0O7eMuaRtas-6+CyafOgA@mail.gmail.com>
>> On Wed, Mar 29, 2017 at 12:46 PM, Kyotaro HORIGUCHI
>> <horiguchi.kyotaro@lab.ntt.co.jp> wrote:
>> > Hello, it would be too late but I'd like to propose this because
>> > this cannot be back-patched.
>> >
>> >
>> > In autovacuum logs, "%u skipped frozen" shows the number of pages
>> > skipped by ALL_FROZEN only in aggressive vacuum.
>> >
>> > So users cannot tell whether '0 skipped-frozen' means a
>> > non-agressive vacuum or no frozen-pages in an agressive vacuum.
>> >
>> > I think it is nice to have an indication whether the scan was
>> > "agressive" or not in log output.
>>
>> Good idea. I also was thinking about this.
>
> Thanks. Currently we cannot use "skipped-frozen" to see the
> effect of ALL_FROZEN.
>
>> > Like this,
>> >
>> >> LOG:  automatic aggressive vacuum of table "template1.pg_catalog.pg_statistic": index scans: 0
>> >
>> > "0 skipped frozen" is uesless in non-aggressive vacuum but
>> > removing it would be too-much.  Inserting "aggressive" reduces
>> > machine-readability so it might be better in another place. The
>> > attached patch does the following.
>> >
>> >>  LOG:  automatic vacuum of table "postgres.public.pgbench_branches": mode: normal, index scans: 0
>> >>  LOG:  automatic vacuum of table "postgres.public.pgbench_branches": mode: aggressive, index scans: 0
>> >
>>
>> Should we add this even to the manual vacuum verbose message?
>
> I forgot that. The patch adds the mode indication in the first
> message of VACUUM VERBOSE.
>
> | =# vacuum freeze verbose it;
> | INFO:  vacuuming "public.it" in aggressive mode
> | INFO:  "it": found 0 removable, 0 nonremovable row versions in 0 out of 0 pages
> ...
> | Skipped 0 pages due to buffer pins, 0 frozen pages.
>
> I still feel a bit uneasy about the word "aggressive" here.

I think we can use the word "aggressive" here since we already use the
word "aggressive vacuum" in docs[1], but it might be easily
misunderstood.

[1] https://www.postgresql.org/docs/9.6/static/routine-vacuuming.html

>Is it better to be "freezing" or something?

An another idea can be something like "prevent wraparound". The
autovaucum process doing aggressive vacuum appears in pg_stat_activity
with the word ".... (to prevent wraparound)". This word might be more
user friendly IMO.

Regards,

--
Masahiko Sawada
NIPPON TELEGRAPH AND TELEPHONE CORPORATION
NTT Open Source Software Center



pgsql-hackers by date:

Previous
From: David Rowley
Date:
Subject: Re: Making clausesel.c Smarter
Next
From: Robert Haas
Date:
Subject: Re: [POC] A better way to expand hash indexes.