[ClusterLabs] Resource Agents: pgsql vs. pgsqlms

[Jehan-Guillaume de Rorthais]

On Thu, 11 Jul 2024 11:52:45 +0200
Oyvind Albrigtsen <oalbrigt at redhat.com> wrote:

> On 11/07/24 11:33 GMT, Jehan-Guillaume de Rorthais wrote:
> >On Thu, 11 Jul 2024 10:17:24 +0200
> >Oyvind Albrigtsen <oalbrigt at redhat.com> wrote:
> >[…]  
> >> >Since postgres ha is rather new to us and me having been lucky not
> >> >having had to deal with perl so far, just reading the agents
> >> >themselves does not really shed that much light on this issue.  
> >> The pgsql has seen more testing and usage, so I would use that unless
> >> there's a specific reason to use PAF instead.
> >>
> >> It's also written in shell, so if you debug by setting trace_ra=1 or
> >> by running "psc resource debug-<action> --full <rsc>" it will also be
> >> easier to debug, as it will list all the commands/checks it's running.  
> >
> >PAF will output debug messages if debug is enabled on Pacemaker side (in
> >/etc/sysconfig/pacemaker or /etc/default/pacemaker).
> >
> >It is true "debug-<action>" commands are not supported with PAF because they
> >are focused on non-promotable agents[1]. PAF rely on cluster contexts and
> >notify actions to make various checks, recovery and node election, which are
> >not set or triggered when using "debug-<action>" commands.  
> There's also debug-promote and debug-demote, so it should cover most
> use-cases.

Yes, but they are not triggered in a cluster transition context, with
surrounding notify actions. And, *if I remember correctly*, the
"OCF_RESKEY_CRM_meta_notify_*" environment vars were not set according to the
cluster real status. The pgsqlms code rely on these features to work correctly.