Re: pg bouncer issue what does sv_used column means

On Fri, Jun 12, 2015 at 09:37:36PM +0000, Sheena, Prabhjot wrote:
> Here is some more information
>
> pool_mode | transaction
>
> We have transactional pooling and our application is set up in such a way that we have one query per transaction. We have set default pool size to 100.
>
> This is output . As you guys can see active connection are 100 and 224 are waiting. We are planning to move default pool size to 250. Please suggest if you guys think otherwise
>
> pgbouncer=# show pools;
> database | user | cl_active | cl_waiting | sv_active | sv_idle | sv_used | sv_tested | sv_login | maxwait
> -----------+-----------+-----------+------------+-----------+---------+---------+-----------+----------+---------
> pgbouncer | pgbouncer | 2 | 0 | 0 | 0 | 0 | 0 | 0 | 0
> site | feature | 418 | 0 | 20 | 17 | 0 | 0 | 0 | 0
> site | service | 621 | 224 | 100 | 0 | 0 | 0 | 0 | 0
> site | zabbix | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0
>
> Prabhjot Singh
> Database Administrator
>
> CLASSMATES
> 1501 4th Ave., Suite 400
> Seattle, WA 98101
> 206.301.4937 o
> 206.301.5701 f
>
> From: Sheena, Prabhjot
> Sent: Friday, June 12, 2015 10:57 AM
> To: 'pgsql-general(at)postgresql(dot)org'; 'pgsql-performance(at)postgresql(dot)org'
> Subject: pg bouncer issue what does sv_used column means
>
> Guys we see spike in pg bouncer during the peak hours and that was slowing down the application. We did bump up the connection limit and it is helpful but now we again notice little spike in connection. And one thing that I notice that is different is jump in sv_used value when I run command show pools during problem times
>
>
> Can anyone please explain what value of sv_used means when i run show pools;
>
>
>
> Regards
> Prabhjot
>

Hi Parbhjot,

The spike in pgbouncer during peak hours just indicates that you are busier then. How
many sv_active do you have in non-peak hours? What kind of system is this on? I suspect
that your hardware cannot actually handle 100 simultaneous processes at once and if you
increase that to 250 processes there is a good likelyhood that your system response
will get even worse. Number of CPU to 2x number of CPU is typical for peak performance
throughput. Are you using a 50-core system? What do the I/O stats look like? You may be
I/O limited.

Regards,
Ken