[pgpool-hackers: 3604] Re: PCP command to interact with Pgpool-II cluster

[Muhammad Usama]

On Thu, Apr 23, 2020 at 2:06 PM Muhammad Usama <m.usama at gmail.com> wrote:
>
> On Tue, Apr 21, 2020 at 12:33 PM Tatsuo Ishii <ishii at sraoss.co.jp> wrote:
> >
> > Usama,
> >
> > Thank you for the propsal.
> >
> > > Hi Hackers,
> > >
> > > When PCP was first added to Pgpool-II it was meant to manage/control
> > > the single Pgpool-II instance.
> > > And after the addition of watchdog more often than not multiple
> > > Pgpool-II nodes are installed the system
> > > and there is not much work/thoughts have gone into the enhancements of
> > > the PCP utilities to make
> > > them work on the Pgpool-II cluster.  So lately I was thinking to
> > > enhance/add the PCP utilities to make them
> > > more cluster-mode friendly.
> > > And for that purpose, I think we would require to add/enhance the
> > > following PCP commands.
> > >
> > > 1- pcp_reload_config
> > > ==================
> > > The utility to make the whole Pgpool-II cluster to reload the
> > > configuration files at once.
> > > I think this is required because currently if some installation has
> > > let say 3 node Pgpool-II cluster
> > > and we want to change one specific configuration parameter ( for
> > > example: adding a new backend node)
> > > we would require to go on and edit Pgpool configuration files on each
> > > node separately and then
> > > issue pgpool reload on each node one by one.
> > >
> > > In my opinion, we need multiple enhancement in this area.
> > > We need some mechanisms to let pgpool-II cluster use the centralized
> > > configuration
> > > ( except watchdog config) or at least a mechanism to push the
> > > configuration settings using some
> > > new utility like "pcp_load_config" that could take a pgpool.conf file
> > > and propagate it to all nodes
> > > using watchdog.
> >
> > Yeah, we definitely need it.
> >
> > > The centralized configuration part of the problem is a big task and I
> > > believe we could take it on after the
> > > "simplifying watchdog config" feature that Peng is working on.
> > > Meanwhile, I think we can work on pcp_reload_config utility in
> > > parallel to make reloading part easier.
> >
> > +1.
> >
> > > 2- pcp_restart_pgpool [cluster]
> > > ==========================
> > > Sometimes, a simple restart solves more problems than any other remedy
> > > or debugging.
> > > So I think we should have a PCP utility to restart a Pgpool-II node or
> > > all node on Pgpool-II cluster.
> > > something similar in functionality like "pg_ctl restart" in postgres
> >
> > +1.
> >
> > > 3- pcp_watchdog_re-elect
> > > ======================
> > > What if we want to change the master/coordinator node of the watchdog
cluster.
> > > There exists no good way for that currently. So I purpose we should
> > > have a command to make
> > > watchdog re-do the elections. I already had a discussion on this one
> > > with Ishii-San sometimes
> > > back so I will send a separate proposal for that as well
> >
> > Looking forward to seeing it.
> >
> > > 4- pcp_stop_pgpool enhancements
> > > ===================
> > > pcp_stop_pgpool should be able to stop all the Pgpool-II nodes in
> > > Pgpool-II cluster.
> > > And for that, I have also cooked up a POC patch which allows the
> > > [--all (-a)] option to
> > > pcp_stop_pgpool utility and when this --all is specified then the
> > > utility stops all the Pgpool-II
> > > nodes part of the watchdog cluster.
> > > So can you please try the patch and also suggest about the command
> > > line option we should
> > > use for making the PCP commands interact with cluster rather than the
> > > single node.
> > > Because I think --all would not be the best choice since it is used by
> > > other utilities for a different purpose,
> > > and we should come up with new standard command-line switch for all
> > > utilities if they want to operate on
> > > the cluster rather than on a single node.
> >
> > +1.
> >
> > What about:
> >
> > --cluster, -c   /* affect to whole watchdog cluster */
> > --local, -l     /* affect only to local watchdog cluster */
>
> Yes seems like a good option since both -c and -l are currently not
> used by any PCP utility.
>
>
> >
> > We actually need --cluster only but for completeness I feel like we'd
> > want to have --local as well (it exactly behaves as PCP commands do
> > today).
> >
> > > Or another way would be to leave the pcp_stop_pgpool utility untouched
> > > and add a new utility pcp_stop_cluster ?
> > >
> > > So what are your suggestions and thoughts on this and do you have
> > > other functions especially related to
> > > cluster mode of pgpool-II that needed controlling/management using
PCP system?
> >
> > I have tested your patch and it worked as expected. Great!
>
> Thanks, I will make the argument changes (-c and -l) and push it.


I have been thinking about this further, Instead of introducing -c and -l
how about adding new argument scope which can be assigned either local or
cluster

i.e.

--scope=[local | cluster]
and the short form would be -s=[l|c]

Regards
Muhammad Usama

>
>
> Best regards
> Muhammad Usama
>
> >
> > 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-hackers/attachments/20200428/2bec645f/attachment.html>