Am 19.02.2014 13:24, schrieb Alexandre DERUMIER: >>> but what about hot plugging cpus? I'm already working on hot plugging >>> memory. >>> >>> So not allowing to modify will fail for hot plug memory and hot plug >>> cpu. These two will become very important the next years or even month. > > >>> B. Do not allow to modify the configuration while the VM is running (and >>> hotplug disabled) > > I think that dietmar want to say, no allowing to change the config file > (through api) when vm is online, > only for values which cannot be change online or not hotplug. (like maxcpu, > cpumodel,change disk if hotplug is disabled... for example) > > (I think this could also avoid some messages in the forum like : "I have > change a value, do a reset of the vm, and I don't see the new value" > If we force to change them offline (non hotplug values), it'll work when the > vm will start) > >>> A. store the configuration at VM start (and use that configuration to start >>> after migration) > I think it's also more complex, because you need to manage 2 configuration, > if you hotplug something, you need to update vmstart config and current > config.
Sure. I already understood it that way. But i'll give you some examples where i see problems. 1.) In the past we had changes to the CPU features for example x2apic. If we just save the config we have the problem that a that after a pve software upgrade this feature is suddently enabled after migration or after restoring from a snapshot. Just because the current version of pve renders a command line with x2apic and the old did not. While this explicit example did not break migration there might be other changes to cpu type, pci handling or whatever which might break migration or restoring from snapshot if we don't save the generated command line. 2.) An other example will be mem hotplug support. If you start a VM with 16GB it will simply have 16GB. If you than hot add 16GB, the machine has 32GB while having an additional memdev and dimmdev. If you migrate you need to add the memdev and dimmdev to the migration command line. If you stop the VM and start the VM you don't have these dimmev and memdev devices as you have 32GB directly incl. in the machine. Stefan > ----- Mail original ----- > > De: "Stefan Priebe - Profihost AG" <[email protected]> > À: "Dietmar Maurer" <[email protected]>, [email protected] > Envoyé: Mercredi 19 Février 2014 12:51:37 > Objet: Re: [pve-devel] saving kvm command line for migration > > > Am 19.02.2014 12:24, schrieb Dietmar Maurer: >>> B.) is to unflexible. >> >> I think this is good enough. > > but what about hot plugging cpus? I'm already working on hot plugging > memory. > > So not allowing to modify will fail for hot plug memory and hot plug > cpu. These two will become very important the next years or even month. > >>> Even it might be easier to implement. >> >> Note: much, much, much easier to implement ;-) > > mhm i don't see a real problem but may be i'm missing something. > >>> We have the same problem for snapshots. >> >> Why? We save the configuration for snapshots, so this is OK. >> >>> We just save the config and HOPE that the newly >>> generated command line matches the old one. >> >> No, we do not hope that. Instead, it is perfectly valid that the command >> lines differ! > > > Stefan > _______________________________________________ > pve-devel mailing list > [email protected] > http://pve.proxmox.com/cgi-bin/mailman/listinfo/pve-devel > _______________________________________________ pve-devel mailing list [email protected] http://pve.proxmox.com/cgi-bin/mailman/listinfo/pve-devel
