[Linux-ha-dev] new stonith external plugin

Serge Dubrouski sergeyfd at gmail.com
Thu Feb 1 08:51:45 MST 2007


On 2/1/07, Andrew Beekhof <beekhof at gmail.com> wrote:
> On 1/29/07, Serge Dubrouski <sergeyfd at gmail.com> wrote:
> > On 1/29/07, Dave Blaschke <debltc at us.ibm.com> wrote:
> > > Andrew Beekhof wrote:
> > > > On 1/29/07, Serge Dubrouski <sergeyfd at gmail.com> wrote:
> > > >> On 1/29/07, Lars Marowsky-Bree <lmb at suse.de> wrote:
> > > >> > On 2007-01-29T10:10:44, Andrew Beekhof <beekhof at gmail.com> wrote:
> > > >> >
> > > >> > > >Attached is an update version. It supports on and off commands now.
> > > >> > > >User still have to configure hostlist in a form "node:xen0_host
> > > >> > > >node:xen0_host ...".
> > > >> > > thats going to be a pain when the VMs are also resources (and
> > > >> therefor
> > > >> > > moving around).
> > > >> > >
> > > >> > > or aren't we dealing with this case here?  i loose track
> > > >> sometimes :-)
> > > >> >
> > > >> > That's not being dealt with, AFAIK. In that case, the command wouldn't
> > > >> > be a xm shutdown/start, but a "crm_resource ..." invocation;
> > > >> potentially
> > > >> > one trying to stop a specific clone. I'm not sure _you_ want to deal
> > > >> > with that yet. ;-)
> > > >> >
> > > >> > However, I think it could be simplified.
> > > >> >
> > > >> > First, in many cases, all the virtual guests are going to be on a
> > > >> single
> > > >> > physical node. (ie, test "clusters".) Then only supplying one node to
> > > >> > ssh to should be sufficient.
> > > >>
> > > >> Let's nod discuss trivial cases.
> > > >>
> > > >> >
> > > >> > Second, by providing a simple list of all physical nodes, the system
> > > >> > should be able to automatically figure out which node it needs to
> > > >> ssh to
> > > >> > to shot the guest. It could try to autopopulate the hostlist from
> > > >> > /etc/xen/vm/* or some other directory, as well.
> > > >>
> > > >> Ok, Let's say user provided us with 3 Xen0 Nodes and ALL 3 of them
> > > >> have config file for node1 in /etc/xen/vm. Then where should I start
> > > >> node1 when "on" command is called?
> > > > wouldn't you just use the same node name that you used for "off"?
> > > >
> > > > or are "on" and "off" distinct operations sent to you from the stonithd?
> > > >
> > > > by that i mean that stonithd will never just tell you to turn a node
> > > > on, only to stop or reset it in some way.  so if you were able to do
> > > > the "off", then you know you can also do the "on".
> > > Sorry to interrupt the flow of this thread, but I have been off chasing
> > > down other non-HA problems...
> > >
> > > Yes, "on" and "off" are distinct operations sent from stonithd.
> > > However, in the event of a STONITH condition, I believe that only
> > > "reset" is sent by stonithd (hence why "on" and "off" are optional
> > > parameters, while "reset" is required).  Indeed, even some STONITH
> > > plugins do not recognize the "on" and "off" parameters.  The "on" and
> > > "off" parameters were added to the stonith command so that one could
> > > reset and power on/off nodes from the command line, so it is possible
> > > for a plugin to receive the "on" or "off" command but that would
> > > indicate that the request came from outside stonithd (command line,
> > > script).  I am not able to grep through the code right now (my pSeries
> > > machine is no longer accepting connections to HMCs so it is lost in
> > > space) to verify my statements, but I believe them to be true from my
> > > journeys through STONITH code...
> >
> > In this case if we all are agrre I can do following for this plugin:
> >
> > 1. Accept a simple list of possible Xen0 hosts as a required parameter.
>
> excellent
>
> > 3. Implement "reset" and "off" commands and ignore "on" command.
>
> great
>
> > 2. Accept xen_vm_config_dir as an optional parameter
>
> do we still need this if we're ignoring "on"?
> no objection, just interested...

As far as I know STONITH specification requires agent to guarantee
that "reset" really stopped a node. To implement that I used "xm
destroy" with pinging a node after that and following "xm start". "xm
start" requires config file. I'll play with "xm reboot" to check if I
can  replace "xm destroy"  -> "xm start" with one "xm reboot".

>
> > It is still not clear what additional commands have to be implemented.
> > I mean do I have to have OCF's "meta-" commands and which of stonith
> > "getinfo-" commands are really required.
>
> lmb?
> _______________________________________________________
> Linux-HA-Dev: Linux-HA-Dev at lists.linux-ha.org
> http://lists.linux-ha.org/mailman/listinfo/linux-ha-dev
> Home Page: http://linux-ha.org/
>


More information about the Linux-HA-Dev mailing list