MIME-Version: 1.0
Content-Type: text/plain; charset=US-ASCII;
 format=flowed
Content-Transfer-Encoding: 7bit
Date: Thu, 22 Oct 2020 10:34:28 +0900
From: Masahiro Ikeda <ikedamsh@oss.nttdata.com>
To: lchch1990@sina.cn
Cc: Amit Kapila <amit.kapila16@gmail.com>, pgsql-hackers
 <pgsql-hackers@postgresql.org>
Subject: Re: Add statistics to pg_stat_wal view for wal related parameter
 tuning
In-Reply-To: <2020102114544806663477@sina.cn>
References: <35ef960128b90bfae3b3fdf60a3a860f@oss.nttdata.com>,
 <CAA4eK1JbbP-QmZWiXwK+H-ZauJCGR6vbAVoGdyx_TTxRPuXzhQ@mail.gmail.com>,
 <06bc2d405b086a72fb14657de9befae8@oss.nttdata.com>
 <2020102114544806663477@sina.cn>
Message-ID: <ac8b1f5e589932254efe373e252e5dcb@oss.nttdata.com>
User-Agent: Roundcube Webmail/1.1.9
Precedence: bulk

On 2020-10-21 15:54, lchch1990@sina.cn wrote:
> I think it's really a more convenient way to collect wal usage
> information,
> with it we can query when I want. Several points on my side:

Thanks for your comments.


> 1. It will be nice If you provide a chance to reset the information in
> WalStats,
> so that we can reset it without restart the database.

I agree.

Now, pg_stat_wal supports reset all informantion in WalStats
using pg_stat_reset_shared('wal') function.

Isn't it enough?


> 2. I think 'wal_write_backend' is mean wal write times happen in
> backends. The describe in document is not so clear, suggest rewrite
> it.

OK, I'll rewrite to "Total number of times backends write WAL data to 
the disk".


> 3. I do not think it's a correct describe in document for
> 'wal_buffers_full'.

Where should I rewrite the description? If my understanding is not 
correct, please let me know.


> 4. Quite strange to collect twice in XLogInsertRecord() for
> xl_tot_len,
> m_wal_records, m_wal_fpi.

Yes, Amit-san pointed me that too.
I'll remove them from pg_stat_wal since pg_stat_statements and vacuum 
log
already shows the related statistics and there is a comment it's enough.

Anyway, if you need to the accumulated above statistics in pg_stat_wal,
please let me know.


> 5. I notice some code in issue_xlog_fsync() function to collect sync
> info,
> a standby may call the issue_xlog_fsync() too, which do not want to
> to collect this info. I think this need some change avoid run by
> standby
> side.

Thanks, I will fix it.

Regards,
-- 
Masahiro Ikeda
NTT DATA CORPORATION