On Wed, 2010-12-22 at 10:37 +0100, Dejan Muhamedagic wrote: > On Wed, Dec 22, 2010 at 09:57:40AM +0100, Holger Teutsch wrote: > > On Tue, 2010-12-21 at 19:03 +0100, Dejan Muhamedagic wrote: > > > Hi, > > > > > > On Tue, Dec 21, 2010 at 05:30:52PM +0100, Holger Teutsch wrote: > > > > Hi, > > > > I would like to submit a libvirt based stonith plugin for review and > > > > possible inclusion to glue. > > > > The plugin uses the client of libvirtd (i.e. virsh) _in the virtual > > > > machines_ and connects remotely to libvirtd on the hypervisor. > > > > Therefore is works with whatever transport or hypervisor that libvirt > > > > supports or will support. > > > > > > Just a note that the reset command should try to boot the host in > > > case it was down too. No objections here to the rest of the code. > > > > As a data center guy I would not expect this. In particular when startup > > fencing comes into play. > > When I _power down_ a cluster member for good reasons and start only one > > node I would not like the other one to be powered on automatigically. > > The power switch is the ultimate thing we control all this stuff > > If you want to keep the node down why not use the poweroff action > for stonith? >
Unfortunately libvirt has no state "powered on / not running" or "persistent power off". I'm pretty sure that e.g HP's ilo/ipmi implementation of "reset" would not power on but would be ignored on a powered off machine. So that might not be an issue with "real" servers. With a previous version of the script on my KVM test cluster startup fencing of pacemaker powered on a stopped machine and I think that is not what you want. > > > Any chance to support more than one host? > > > > I reasoned about this as well but as we can not assume 'host name' == > > 'domain id' that means domain_id has to be a list as well (with defaults > > or partial defaults). > > IIRC, there was one stonith agent which does this kind of > mapping. Alternatively, perhaps drop domain_id and allow > appending it in the hostlist (as in external/xen0), i.e. > "node1[:domain_id] ...". > > > I will think again about feasability with not overcomplicated code. > > This should reduce the configuration, so I think it's worth the > effort. Will go with your proposal. - holger > > > > > > Cheers, > > > > > > Dejan _______________________________________________________ Linux-HA-Dev: [email protected] http://lists.linux-ha.org/mailman/listinfo/linux-ha-dev Home Page: http://linux-ha.org/
