Re: pg_stat_wal_write statistics view

On Tue, Aug 15, 2017 at 7:39 AM, Peter Eisentraut <
peter(dot)eisentraut(at)2ndquadrant(dot)com> wrote:

> On 3/29/17 22:10, Haribabu Kommi wrote:
> > Updated patch to use shared counter instead of adding pg_stat_ calls to
> send
> > the statistics from each background process/worker.
>
> Your patch needs to be rebased and the OID assignments updated.
>

Rebased patch is attached. And also it has taken care of the comments
provided
by Andres and Fuji Masao in the upthread.

I separated the walwriter statistics from the other background processes to
check
how much load does the other processes are really contributing to wal
writing.

Following is the test results during my performance test.

writes | walwriter_writes | backend_writes | dirty_writes |
walwriter_dirty_writes | backend_dirty_writes | write_blocks |
walwriter_write_blocks | backend_write_blocks | write_time |
walwriter_write_time | backend_write_time | sync_time | walwriter_sync_time
| backend_sync_time | stats_reset
--------+------------------+----------------+--------------+------------------------+----------------------+--------------+------------------------+----------------------+------------+----------------------+--------------------+-----------+---------------------+-------------------+-------------------------------
0 | 17748394 | 1383789657 | 0 |
0 | 0 | 0 | 21153194 |
3039806652 | 0 | 0 | 0
| 0 | 259250230 | 17262560725 | 2017-09-05
18:22:41.087955+10
(1 row)

I didn't find any other background processes contribution to the WAL
writing activity.
May be we can combine them with backend stats?

I ran the performance test on this patch with pgbench to find out the
impact of these
changes. Because of my low end machine, the performance results are varying
too
much, I will try to get the performance figures from an high end machine by
that time.

Attached the latest patch and performance report.

Regards,
Hari Babu
Fujitsu Australia