Re: main vs primary

Hi,

Thank you for the update.

I have lool into <https://github.com/pgpool/pgpool2/issues/118>.
I think this issue might be caused by the setting "failover_when_quorum_exists = off".
I will verify the behavior and get back to you with feedback.

________________________________________
差出人: Luca Ferrari <fluca1978(at)gmail(dot)com>
送信: 2025 年 7 月 31 日 (木曜日) 17:58
宛先: Bo Peng <pengbo(at)sraoss(dot)co(dot)jp>
Cc: pgpool-general(at)lists(dot)postgresql(dot)org <pgpool-general(at)lists(dot)postgresql(dot)org>
件名: Re: main vs primary

On Wed, Jul 30, 2025 at 12:03 PM Bo Peng <pengbo(at)sraoss(dot)co(dot)jp> wrote:
>
> Hi,
>
> The "primary node" refers to the PostgreSQL primary server in a streaming replication setup.
> The "main node" indicates the main backend node, usually the first live backend (i.e., the one with the smallest node ID).

Thanks, much more clear now.

>
> > LOG:  failover: 2 follow backends have been degenerated
> > LOG:  failover: set new primary node: 1
> > LOG:  failover: set new main node: 2
> >
> > in my scenario, the third host (so number 2) has been promoted. So
> > what is the "primary" node 1 doing?
>
> If node2 is promoted to the primary, the primary node should be shown as "node: 2".
> Could you share more logs related to the failover, as well as your configuration file?
>

This is something I'm fighting from days. I've an open issue, updated
almost daily, I don't think repeating all the infor here is
appropriate but I could open a new thread if you believe it is worth.
See <https://github.com/pgpool/pgpool2/issues/118>.
In short: I've three nodes, I shutdown the first and the second gets
promoted. Failover and follow primary are fine, watchdog is happy.
Then I stop the second node, expecting the third to be promoted, and
this happens in fact. But pgpool stops working, and pcp reports the
third node as "standby primary", not "primary primary".
In particular, see behavior in
<https://github.com/pgpool/pgpool2/issues/118#issuecomment-3138977151>.

Any help is appreciated.

Thanks,
Luca