[Linux-HA] Master / Slave resource

[Andrew Beekhof]

On 10/18/06, Keisuke MORI <kskmori at intellilink.co.jp> wrote:
> "Andrew Beekhof" <beekhof at gmail.com> writes:
>
> > On 10/18/06, Keisuke MORI <kskmori at intellilink.co.jp> wrote:
> >> Hi Andrew,
> >>
> >> "Andrew Beekhof" <beekhof at gmail.com> writes:
> >>
> >> > This may help you http://linux-ha.org/v2/Concepts/MultiState
> >> > particularly section 6
> >> >
> >> > (  yes i know the docs are generally hopeless and hard to navigate :-(  )
> >>
> >> Sorry but I've already read the docs and have been trying to make it work,
> >> and still have the questions in my previous mail.
> >>
> >> My understanding from the doc regarding to my question 1,
> >> using "notify" is mandatory
> >
> > where does it say that?
> > the example has notifications turned off so i don't know why you'd think this.
>
> Oh, I mean in order to make the drbd OCF RA work.
> The RA only executes crm_master in the post start event.
>
> >
> >> and "post start" should be notified
> >> to the RA, but as long as I tried the notification didn't come.
> >
> > Did you configure the CRM to send them?
>
> Yes, it's like this.
>        <master_slave id="drbd0-ms" notify="true">
>
> I also confirmed that other notifications (for
> demote/stop) are notified during my test.
>
>
> >
> >> I have no idea whether it is because of my config's fault or the
> >> RA's fault or a Heartbeat's bug or the design issue.
> >
> > Neither do I since your config was not included ;-)
> >
>
> Sorry, I sent it in the other mail.
> http://www.gossamer-threads.com/lists/linuxha/users/35573#35573
> I noticed that I forgot it after I went to the bed that day :-)

oh is that what that was

>
>
> >>
> >> Regarding to my Q.4 and 5, I could not find related descriptions
> >> on the docs. Please point me out if I'm missing.
> >
> > Q4: you cant.  master/slave resources can contain groups but not the
> > other way around
>
> OK. Do you have any plan to implement that feature?
> I think "failover along with the master" is very important for
> real usage.

you'll be able to colocate resources with a m/s resource soon - but by
using a rsc_colocation constraint not by putting them into a group
together.

> > Q5 : http://linux-ha.org/v2/Concepts/MultiState#head-e5d5a9318a3fdb0dffab64280c0fa0840a721b19
> >
> > you'd need update the CIB and change the preference for one of the
> > other instances
>
> Understand. I can do it by changing the attribute value as same
> as crm_master does, correct?

right.  crm_master is just a C-wrapper for cibadmin

> I think it would be convenient if
> crm_master can be used from the admin's command line.

hmmm. probably.

> >> > On 10/17/06, Keisuke MORI <kskmori at intellilink.co.jp> wrote:
> >> >> Hi,
> >> >>
> >> >> I'm trying to use master/slave feature with heartbeat 2.0.7.
> >> >>
> >> >> I know that the master/slave is not yet ready to use "out of the box"
> >> >> according to some articles from the ML archive, but I would like
> >> >> to know how is it designed to work, what the current status of
> >> >> the implementation and what the rest things to do.
> >> >>
> >> >> For my testing, I configured OCF drbd RA which comes with 2.0.7
> >> >> on two nodes. When I start heartbeat, slave resources start on
> >> >> each node but none of them get promoted.
> >> >>
> >> >> My questions are:
> >> >>
> >> >> 1. The drbd OCF RA apparently expects to get "post start"
> >> >>    notification to issue crm_master command, but I never see the
> >> >>    "pre/post start" notification in the log (see the attached
> >> >>    log for details).
> >> >>
> >> >> 2. If the RA is wrong, how should it be fixed?
> >> >>
> >> >> 3. In my real application, I expect to work as the following sequence.
> >> >>    Is it possible?
> >> >>    i)  the heatbeat starts the resources on all nodes as "slave" state.
> >> >>    ii) after all resources have launched, the heartbeat choose
> >> >>        which one should be the master (determined by a config. anyhow)
> >> >>        and issue promote to it.
> >> >>
> >> >> 4. How can I make a group of m/s resources?
> >> >>    What I want is to let some resources (typically Filesystem or IP)
> >> >>    failover along with the master resource.
> >> >>    The DTD says that <master_slave> can contain <group> but I'm
> >> >>    not sure if it's meant for it.
> >> >>
> >> >> 5. How can I manually move the master?
> >> >>    (i.e. crm_resource -M for master resource)
> >> >>
> >> >>
> >> >> Any help or advice is appreciated.
> >> >> Thank you.
> >> >> --
> >> >> Keisuke MORI
> >> >> NTT DATA Intellilink Corporation
> >> >>
> >> >> _______________________________________________
> >> >> 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
> >> >>
> >> > _______________________________________________
> >> > 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
> >>
> >> --
> >> Keisuke MORI
> >> NTT DATA Intellilink Corporation
> >>
> >> _______________________________________________
> >> 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
> >>
> > _______________________________________________
> > 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
>
> --
> Keisuke MORI
> NTT DATA Intellilink Corporation
>
> _______________________________________________
> 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
>