Am 23.06.25 um 10:11 schrieb DERUMIER, Alexandre: > I don't known if it's the best place, but 1 thing missing currently, > if ressource affinity, like for example, > > if a vm use a specific storage, it need to run on a node where the > storage is present. > Same for the number of cores of vm (the host numbers of cores need to > be >= than the vm cores). > > I don't known if it could be possible to reuse your rules framework, > and add some kind of implicit rules. (and maybe them dynamic if vm > config change) ? > > > With current HA for example, you setup local zfs mirroring between 2 > nodes, and if the vm is migrated to the wrong node, the vm is > completly stuck, and the only way to fix it to disable HA and move the > vm config file manually to the correct node.
These are definitively good points and wanted features, but I do not think that we need to have them in the first version applied, IMO the series is already big enough as is. So, it's probably easier to add them later on, as auto-generated constraints derived from the VM config and updated if that config changes; as we track the "config version" in pmxcfs already anyway it should be possible to detect if something might have changed quite cheaply. This would also benefit from re-evaluating if we can transform the HA stack into a general resource manager (which it basically already is prepared for) with HA+fencing for specific guests as opt-in extra feature, as that would better fit the balancing feature, which should take all guests into account, and also allow setting affinity constraints on guests that are not HA managed. But again, I would do that on top of this series, it should not really change anything for the implementation here. _______________________________________________ pve-devel mailing list pve-devel@lists.proxmox.com https://lists.proxmox.com/cgi-bin/mailman/listinfo/pve-devel
