Sander van Vugt wrote: > Hi Niels, > > I haven't followed the most recent developments for the Xen HA resource > (so I'm sorry if I missed something), but there is a conceptional issue > I have with Heartbeat orchestrating a live migration: what would be the > criterium to start such a migration? Typically Heartbeat reacts when > either a node or a resource is dead. If a Xen resource is dead it cannot > be live migrated anymore (since it is dead), so you would need to > predict a disaster from happening in the next couple of seconds/minutes > and I don't really see how to do that. > > Best Regards, > Sander > > > On Thu, 2008-01-03 at 13:17 +0100, Niels de Carpentier wrote: >> I'm trying to get live migration working in heartbeat, but somehow heartbeat >> will only do stop/start, and not try migrate_to/migrate_from. >> >> Both systems are running centos 5.1: >> >> heartbeat-pils-2.1.2-3.el5.centos >> heartbeat-stonith-2.1.2-3.el5.centos >> heartbeat-2.1.2-3.el5.centos >> >> The logs and the pe-input have been attached. Does anybody have any clue >> what >> I'm doing wrong, and how I can get live migration to work? >> >> The config is: >> >> <group id="group_xen_sysman001"> >> <primitive class="ocf" id="sysman001" provider="heartbeat" >> type="drbdlvmxen"> >> <operations> >> <op id="sysman001_mon" interval="10s" name="monitor" >> timeout="30s"/> >> </operations> >> <instance_attributes id="sysman001_iattr"> >> <attributes> >> <nvpair id="sysman001_iattr_0" name="drbd" value="drbd3"/> >> <nvpair id="sysman001_iattr_1" name="vg" >> value="raid5_logging"/> >> <nvpair id="sysman001_iattr_2" name="xmfile" >> value="/etc/xen/sysman001.cfg"/> >> </attributes> >> </instance_attributes> >> <meta_attributes id="sysman001_mattr"> >> <attributes> >> <nvpair id="sysman001_mattr_0" name="allow_migrate" >> value="true"/> >> </attributes> >> </meta_attributes> >> </primitive> >> </group> >> >> >> Anyother question I have is about the order of the migrate_to/migrate_from >> commands. Are they both issued at the same time, or is the migrate_to >> command >> issued first? >> >> Thanks, >> >> Niels de Carpentier >> _______________________________________________
you hit the nail on the head. if the xen vm has failed, its kind of hard to "live migrate" it. i believe live migration was intended to be used as a maintenance function, not an HA function. instead, you should monitor the status of the xen vm (weather its actually running properly or not) as well as the physical boxen and migrate the xen vm to another node that can run it in the event of a physical node failure of xen vm failure. currently i have it "kind of" working with drbd and xen coupled together. three xen vm's run on node a and three xen vm's on node b. when node a fails, the drbd resources and the xen vm's that were on node a get migrated to node b. the only issue that i see with this the boot delay of the xen vm on node b........maybe that delay is acceptable, maybe its not. for me it is ;) i have a pair of slow DL580's and its pretty quick to boot. i do not have any sort of monitoring (from a heartbeat RA perspective) of the xen vm's in place....just the physical box. iirc, the xen RA currently in stable is limited to an older version of xen which may explain why it doesn't work as expected. hope this gets you on the right track..... _Terry _______________________________________________ Linux-HA mailing list [email protected] http://lists.linux-ha.org/mailman/listinfo/linux-ha See also: http://linux-ha.org/ReportingProblems
