[pgpool-general: 7163] Re: Long wait time when stopping Pgpool

[Nikhil Shetty]

Hi Team,

Thank you for the response. It was a mistake from our end, we had set jdbc
url as (pgpool_ip,master_ip,slavei_p) so if pgpool and master goes down, it
will start sending data to slave_ip.
This leads to problems in load balancing.



On Mon, Jul 20, 2020 at 5:50 AM Tatsuo Ishii <ishii at sraoss.co.jp> wrote:

> >  - I tried this option to stop and start pgpool but this affects load
> > balancing. We I use systemctl to stop and start pgpool, then write
> > statements are sent to primary only but when I use stopping and starting
> > with "pgpool stop -mf" and "pgpool -f pgpool.conf -D" respectively,
> pgpool
> > doesn't recognize write and sends to standby, this is resolved only after
> > restarting the application
>
> I am confused. How did you do when you say "I tried this option to
> stop and start pgpool"? Pgpool's stop option is nothing related to
> load balance.
>
> Anyway I suspect there's something wrong with the systemd setting for
> pgpool. Can you share /etc/sysconfig/pgpool?
>
> > Because systemd kill -9 pgpool to not give a chance to remove the socket
> > files.
> > - This should not be the case right, because in production there will be
> > connections all the time and after using systemctl these files should be
> > removed
>
> You can configure systemctl's parmeter for pgpool to shut down pgpool
> properly even if there are connections from clients. That's why I am
> asking /etc/sysconfig/pgpool.
>
> > Thanks and Regards,
> > Nikhil
> >
> >
> > On Fri, Jul 17, 2020 at 3:49 AM Tatsuo Ishii <ishii at sraoss.co.jp> wrote:
> >
> >> First of all, you are asking questions about propriety
> >> packages. Please ask questions regarding propriety products to the
> >> vendor.
> >>
> >> > Hi Team,
> >> >
> >> > I am testing pgpool for my workload.We have three nodes of pgpool used
> >> as a
> >> > load balancer and connection pooler across three nodes of database
> >> servers.
> >> >
> >> > All application connections to the database are through the pgpool.
> >> Without
> >> > stopping the application, if we want to stop the pgpool using
> "systemctl
> >> > stop edb-pgpool-4.0" then it takes a long time because connections are
> >> > still coming to the pgpool. Pgpool is stopped only after no new
> >> connections
> >> > appear + existing connections are terminated.
> >> >
> >> > Even after Pgpool is stopped, I can see that files(.s.PGSQL.9898 and
> >> > .s.PGSQL.9999) under socket_dir which in our case is /tmp are not
> >> removed.
> >> > This does not allow the pgpool to start using  "systemctl start
> >> > edb-pgpool-4.0". We have to manually remove the files under /tmp and
> then
> >> > start pgpool.
> >> >
> >> > I have two questions:
> >> > 1. Before stopping pgpool, how can we prevent new connections from
> coming
> >> > into pgpool without stopping the application and terminate the
> >> > existing connections so that the stop command does not wait.
> >>
> >> As other people already pointed out, there's an option to stop pgpool
> >> without waiting for clients disconnected to pgpool. Probably you can
> >> set the option to tweak systemectl related files. The way to do it
> >> depends how the package is created, which we don't know. If you are
> >> not sure, ask the vendor.
> >>
> >> > 2. After stopping pgpool, why are files under /tmp not removed?
> >>
> >> Because systemd kill -9 pgpool to not give a chance to remove the
> >> socket files.
> >>
> >> > PS:Files are removed from /tmp if there are no connections via
> >> pgpool(this
> >> > will never be the case in production environment)
> >>
> >> Because if there's no connection from clients, pgpool can remove the
> >> socket files before systemd kills pgpool.
> >>
> >> Best regards,
> >> --
> >> Tatsuo Ishii
> >> SRA OSS, Inc. Japan
> >> English: http://www.sraoss.co.jp/index_en.php
> >> Japanese:http://www.sraoss.co.jp
> >>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.sraoss.jp/pipermail/pgpool-general/attachments/20200727/517f9c7d/attachment.html>