[Linux-HA] stickiness for master in master/slave resource?

[Andrew Beekhof]

On 7/12/07, Andreas Kurz <andreas.kurz at gmail.com> wrote:
>
> > > > > > > I am using hearbeat 2.1.0 on CentOS 4.5 and drbd 8.0.3. I know
> drbd8
> > > > > > > is currently not supported but I give it a try ;-).
> > > > > > >
> > > > > > > I adopted the drbd ocf RA and configured a master/slave
> resource with
> > > > > > > resource_stickiness INFINITY and a location constraint to
> define the
> > > > > > > prefered master. I killed heartbeat on the drbd master and the
> second
> > > > > > > node does the failover as expected .... but .... when the
> stonithed
> > > > > > > node comes up again the drbd master resource stays on the
> second node
> > > > > > > as expected, but in a first attempt to negotiate who is the
> master for
> > > > > > > the drbd resource heartbeat demotes the drbd resource,
> including a
> > > > > > > stop of the drbd resource. Then the second nodes gets promoted
> again
> > > > > > > and the drbd starts up again on the second node.
> > > > > > >
> > > > > > > Is this a known behaviour, or is there something wrong with my
> > > > > > > configuration or is this a drbd8 problem?
> > > > > >
> > > > > > can you please attach the files
> > > > > >    /var/lib/heartbeat/pengine/pe-warn-72.bz2
> > > > > >    /var/lib/heartbeat/pengine/pe-warn-73.bz2
> > > > > > from smsdb06?
> > > > >
> > > > > of course, thanks for your help!
> > > >
> > > > ouch - we're not behaving properly here at all.
> > > > i'll get back to when its fixed.
> > >
> > > Not "fixed", but I can offer a simple work-around
> > >
> > > Change:
> > >            <nvpair id="remove-after-stop" name="remove-after-stop"
> > > value="true"/>
> > > to
> > >            <nvpair id="remove-after-stop" name="remove-after-stop"
> > > value="false"/>
> > >
> > > That particular option is not well tested, I'd not advise turning it
> on.
> > > Was there any particular reason you had it on?
> >
> > When reading the description of this option I found it a good idea to
> > use it, but I think I can live without it. I will test this
> > work-around.
>
> This workaround has no noteable effect on my config ....


doh! you're right.
it appears some changes I've made since that version are also required.

it didn't occur to me since I normally work with the very latest

you have 3 main options at this point:
* grab the latest from http://hg.linux-ha.org/dev/archive/tip.tar.bz2
* wait for alan to put out 2.1.1 (which has the required changes)
* wait (a week or so) for me to refresh the packages at
      http://software.opensuse.org/download/server:/ha-clustering/

additionally
> the resource_stickiness for the master resource does not work as
> expected and I have no idea how the values for the crm_master command
> are calculated.


Its all done in the RA, but I don't know what lmb based it on.

Once the drbd master stays on the second node even
> after the first node with the higher location-constraint score comes
> up again, after the next reset the first node gets drbd master
> although there is the resource_stickiness=INFINITY in the master-slave
> resource .....
>
> Is there an easy way to predict the master-slave behaviour .... or to
> change my config to get the expected drbd-master stickiness???
>

It will be much more intuitive with the changes i recently made.