[pgpool-general: 3574] Re: Out of sync with actual db state (replication / load balancing mode)

[Yugo Nagata]

Hi Gervais,

On Thu, 26 Mar 2015 11:39:40 -0300
Gervais de Montbrun <gervais at silverorange.com> wrote:

> I have two servers running with replication, load balancing, watchdog (with virtual IP) enabled.
> 
> My plan is to remove any single point of failure from my database. The web sites I am running would point to the virtual interface’s IP on whatever is running as a primary in the pgpool and there would be two database servers behind the pool. Here is my setup as it stands:
> 
> dbserver0:
> 	pgpool - get’s virtual IP if it starts first
> 	postgres
> 
> dbserver1:
> 	pgpool
> 	postgres 
> 
> I have everything working except that the there can be a situation where one of the pgpools starts up with a stale pgpool_staus file and it thinks a backend node is either up when it is down (or down when it is up). I am not sure how to sort this out as this file seems to only be written to when pgpool stops.

When pgpool starts up with -D option, an old pgpool_status is discarded
and pgpool recognizes the current status of backends.
If you want to modify backend status while pgpool is runing,
you can use pcp_attach_node or pcp_detach_node command.

> 
> Can someone suggest how to sort this out. I know that you will likely need to review any number of config files and/or scripts in my configuration. Can you let me know what you would want to see before I sanitize them all and send them on.
> 
> Also, if my plan is completely off, I am open to any alternative suggestion as to how to best achieve my goal as described above.
> 
> Cheers,
> Gervais
> ------------------------------------
> Gervais de Montbrun
> Systems Administrator
> GTalk: gervais at silverorange.com
> 


-- 
Yugo Nagata <nagata at sraoss.co.jp>