On Fri, 2020-05-08 at 20:47 +0200, Kab Naj wrote:
> Hello,
>
> thank you very much for possible alternatives.
> I still prefer if group unordered resources were possible, but
> probably there was no demand for it.
>
> I understand that in all three provided options I wouldn't use
> resource groups at all.
Correct
> In option 2 - "all with A" trick, I would use A as element to move
> the whole "group".
> In option 3 - I would use a tag as reference the whole "group",
> right?
Correct
> In pcs implementation, can I move a "group" by naming a tag of
> ordering or colocation resource sets ?
Not with the "move" command, which looks for a particular resource to
stop then start. But all "move" does is create a location constraint
for the resource, and you can do that with a tag.
Unfortunately the pcs constraint command currently checks that the
given name exists as a resource, so you can't create the constraint
that way. But you could use pcs cluster edit or cibadmin to add the XML
directly. Instead of:
pcs resource move rsc1
or equivalently with the pacemaker tools
crm_resource --move -r rsc1
you can add this XML to the configuration:
> I will need to test the behavior of cluster while moving, clearing,
> cleanup,..
> All my co-workers are used to "resource/service groups" as reference
> points, so I will need to change the procedures and the way of
> thinking.
>
> Regards
>
> Jan
>
>
> On Thu, May 7, 2020 at 7:04 PM Ken Gaillot
> wrote:
> > Hi,
> >
> > You have a few alternatives to groups.
> >
> > 1 - You can configure independent colocation constraints for each
> > resource. E.g. "B with A", "C with B", etc. This has the advantage
> > that
> > if you just want all the resources on the same node, you could
> > colocate
> > all later resources with the first one ("B with A", "C with A",
> > etc.),
> > so that there's no dependency between later resources (only the
> > first
> > resource has to be active for any of the others to be active,
> > taking
> > into account any ordering constraints).
> >
> > 2 - You can use resource sets in colocation constraints. You can do
> > the
> > "all with A" trick with this method using two resource sets, one
> > with
> > just A and the other non-sequential with all the rest. See:
> >
> > https://clusterlabs.org/pacemaker/doc/en-US/Pacemaker/2.0/html-single/Pacemaker_Explained/index.html#s-resource-sets-colocation
> >
> > 3 - You can use tags, and use a tag in a colocation constraint
> > resource
> > set. The main advantage of this approach would be if you want to
> > use
> > the logical group in more than one place. See:
> >
> > https://clusterlabs.org/pacemaker/doc/en-US/Pacemaker/2.0/html-single/Pacemaker_Explained/index.html#_tagging_configuration_elements
> >
> >
> > On Thu, 2020-05-07 at 18:06 +0200, Kab Naj wrote:
> > > Hello,
> > >
> > > I was trying to set parallel execution of resources in resource
> > > group, but I was not successful.
> > > The goal was to have resources within one resource group in one
> > > location but order of resources would rely on Ordering
> > constraints,
> > > thus possibly resources could run in parallel if constraints
> > allowed
> > > it.
> > >
> > > By default it is not the case and resources run one by one in
> > their
> > > order of resource group.
> > > I found the option that is designed to be used in resource clones
> > -
> > > "ordered"
> > > ordered - Should the copies be started in series (instead of in
> > > parallel). Allowed values: false, true.
> > >
> > > I tried to use this option in my resource group by setting
> > > "ordered=false"
> > > Resources could be started in parallel then, but I encountered
> > > strange and unpredictable behavior when some resource start was
> > not
> > > successful.
> > >
> > > I understand that "ordered=false" is documented to be used only
> > in
> > > resource clones, not in resource groups.
> > >
> > > Do we have other option that resources within resource group
> > would
> > > start in parallel and rely on Ordering constraints, not their
> > > resource group order?
> > > We have many logical resource groups, so we don't want to have
> > > resources without being added to any resource group.
> > >
> > > Regards
> > >
> > > Jan
> > ___
> > Manage your subscription:
> > https://lists.clusterlabs.org/mailman/listinfo/users
> >
> > ClusterLabs home: https://www.clusterlabs.org/
--
Ken Gaillot
___
Manage your subscription:
https://lists.clusterlabs.org/mailman/listinfo/users
ClusterLabs home: https://www.clusterlabs.org/
