[Linux-HA] Best effort HA

[Andrew Beekhof]

On 5/14/07, Kai Bjørnstad <kai at scali.com> wrote:
> Hi again Yan
>
> comments within
>
> On Friday 11 May 2007 00:57:36 Yan Fitterer wrote:
> > Kay,
> >
> > Sorry I couldn't answer this earlier...
> >
> > When I run the pe input file on my system here (2.0.7-1.2), the ptest
> > utility moves all resources to the second node, as expected. None are left
> > on the first, AFAICS. See output attached. So either your version of HB
> > comes to a different conclusion -> a bug, or something else is going on...
> >
>
> Did 2.0.7 have the turn off ordering on groups features? I thought that came
> in 2.0.8.
> Anyway we have compiled it ourselves, but it is a straight build/package from
> the standard 2.0.8 linux-ha tar-ball.
>
> > To me, the issue where you have a group where "collocation=true" and
> > resources don't move together looks like a bug. If you can reproduce this,
> > I think you should open a bugzilla for it, and see where it goes.
> >
> I'll see if a can reproduce it on a rpm built by suse or fedora.
> I plan to see if we can build a 2.0.9 or something to see if it's fixed.
> I'll open a bugzilla on it of not....
>
> > Something else: I'm not sure how well things will work with a group where
> > "ordered=false", but with ordering constraints attached to the resources
> > within. I'm not sure this will work, as it could be valid for the pengine
> > to give precedence to the group's "ordered=false" setting, and ignore
> > completely the manual ordering constraints.
> >
> I see your point. I am actually wondering if the semantics around groups are a
> bit strange? Intuitively you would think that a resource group is just like
> any other resource. But as soon as you turn of colocation and ordering you
> suddenly encounter a problem. If n of m resources within a group fail, what
> is then the state of the group (started or failed)?

failed

>
> > Maybe somebody with better understanding can answer this one?
> >
> > Where is the 2.0.8 version you have coming from? What platform? I'm
> > starting to wonder if the 2.0.8 you're using is "sane"...
> >
> tar ball downloaded from linux-ha, just made our own rpm-spec file
>
> > To avoid the issue above where you create a group without ordering, but
> > need some ordering nevertheless, maybe you should try a config getting rid
> > of the group, and doing the co-location manually. Bit of a pain to write,
> > but IMHO worth a try.
>
> I'll see if I have time to write basically the same configuration with
> explicit rules (no groups) and see what happens. Unfortunately I have already
> blown all estimates on my tasks have others I need to follow up and a
> deadline approaching.... you know the drill :-)
>
> But I am still wondering, what rule is causing a full resource shutdown when
> using a normal resource group (ordering=true, colocation=true) if any
> resource (including the one listed at the end) fail after a failover?
>
> If the reason is the colocation rule, I will never get my best-effort config
> to work.

it can be either depending on the scenario.

if ordering=true and the first resource has to stop - then all
resources after it also have to stop

if colocated=true and the first resource cant run anywhere - then
there is nowhere for the other resources to run either

> I'll do some more tests and pop a new mail...soon...
>
> cheers
> Kai
>
> >
> > Yan
> >
> > >>> On Thu, May 10, 2007 at 11:09 AM, in message
> >
> > <200705101209.25641.kai at scali.com>, Kai Bjørnstad <kai at scali.com> wrote:
> > > Hum, maybe you wanted the pe-input file instead
> > > (at time of lsb_conserver failure)
> > >
> > > Kai
> > > --
> > > Kai R. Bjrnstad
> > > Senior Software Engineer
> > > dir. +47 22 62 89 43
> > > mob. +47 99 57 79 11
> > > tel. +47 22 62 89 50
> > > fax. +47 22 62 89 51
> > > kai.b at scali.com
> > >
> > > Olaf Helsets vei 6
> > > N0621 Oslo, Norway
> > >
> > > Scali - www.scali.com
> > > Scaling the Linux Datacenter
> >
> > _______________________________________________
> > Linux-HA mailing list
> > Linux-HA at lists.linux-ha.org
> > http://lists.linux-ha.org/mailman/listinfo/linux-ha
> > See also: http://linux-ha.org/ReportingProblems
>
>
>
> --
> Kai R. Bjørnstad
> Senior Software Engineer
> dir. +47 22 62 89 43
> mob. +47 99 57 79 11
> tel. +47 22 62 89 50
> fax. +47 22 62 89 51
> kai.b at scali.com
>
> Olaf Helsets vei 6
> N0621 Oslo, Norway
>
> Scali - www.scali.com
> Scaling the Linux Datacenter
> _______________________________________________
> Linux-HA mailing list
> Linux-HA at lists.linux-ha.org
> http://lists.linux-ha.org/mailman/listinfo/linux-ha
> See also: http://linux-ha.org/ReportingProblems
>