Jean-Pierre Ribeauville
2015-Sep-03 09:18 UTC
Re: [libvirt-users] about guest live migration configuration changes persistence ..
Hi, Thanks for infos. My concern is the following : In order to establish datas exchanges between hosts and guest , I'll add a channel device to a running Guest ( as I learned from you that it's possible to do that without stopping the Guest). When the guest migrates to another host , this one must be able to retrieve the channel device parameters to be able to dialog with the migrated guest. I plan to use dumpxml to retrieve these infos. Will these infos be updated on the second host where the guest has migrated ? Thx. J.P. -----Message d'origine----- De : Michal Privoznik [mailto:mprivozn@redhat.com] Envoyé : mercredi 2 septembre 2015 14:12 À : Jean-Pierre Ribeauville; libvirt-users@redhat.com Objet : Re: [libvirt-users] about guest live migration configuration changes persistence .. On 31.08.2015 09:27, Jean-Pierre Ribeauville wrote:> Hi, > > > Is a change , by edit it via virsh edit , in the "local" .xml > configuration file (i.e. adding a channel device) persistent when live migrating the guest on another host ?Yes. Since the 0.9.7 release, the configuration XML is transferred to the other side too. However, by default, migrated domains are just transient (unless you pass VIR_MIGRATE_PERSIST_DEST flag, or virsh migrate --persistent ..) so the configuration XML doesn't have much effect in that case. Michal
Michal Privoznik
2015-Sep-03 09:33 UTC
Re: [libvirt-users] about guest live migration configuration changes persistence ..
On 03.09.2015 11:18, Jean-Pierre Ribeauville wrote:> Hi, > > Thanks for infos. > > My concern is the following : > > In order to establish datas exchanges between hosts and guest , I'll add a channel device to a running Guest ( as I learned from you that it's possible to do that without stopping the Guest). > When the guest migrates to another host , this one must be able to retrieve the channel device parameters to be able to dialog with the migrated guest. > I plan to use dumpxml to retrieve these infos. > Will these infos be updated on the second host where the guest has migrated ?Sure. Otherwise migration would be broken in general. I mean, if guest uses say tap0 on the source, the device may already have been taken on the destination. Libvirt must then find/create another free device for the migrated domain to use, e.g. tap1. It's the same with other devices. Even consoles. But if you're using channel device (assuming you're using a unix socket as backend (=part visible within the host)), the path to the socket should be preserved. Michal
Jean-Pierre Ribeauville
2015-Sep-03 09:42 UTC
Re: [libvirt-users] about guest live migration configuration changes persistence ..
Hi, Effectively, I use unix socket as backend. Thx for help. J.P. -----Message d'origine----- De : Michal Privoznik [mailto:mprivozn@redhat.com] Envoyé : jeudi 3 septembre 2015 11:33 À : Jean-Pierre Ribeauville; libvirt-users@redhat.com Objet : Re: [libvirt-users] about guest live migration configuration changes persistence .. On 03.09.2015 11:18, Jean-Pierre Ribeauville wrote:> Hi, > > Thanks for infos. > > My concern is the following : > > In order to establish datas exchanges between hosts and guest , I'll add a channel device to a running Guest ( as I learned from you that it's possible to do that without stopping the Guest). > When the guest migrates to another host , this one must be able to retrieve the channel device parameters to be able to dialog with the migrated guest. > I plan to use dumpxml to retrieve these infos. > Will these infos be updated on the second host where the guest has migrated ?Sure. Otherwise migration would be broken in general. I mean, if guest uses say tap0 on the source, the device may already have been taken on the destination. Libvirt must then find/create another free device for the migrated domain to use, e.g. tap1. It's the same with other devices. Even consoles. But if you're using channel device (assuming you're using a unix socket as backend (=part visible within the host)), the path to the socket should be preserved. Michal