This message was deleted.
# harvester
a
This message was deleted.
r
The two types of VM networks I can create in Harvester UI have different options. Apparently, the UntaggedNetwork (PVID=1) simply inherits the configuration of the associated cluster network. The other type is L2VlanNetwork, and for that type I can optionally specify a particular DHCP server (or default to broadcast) or a subnet CIDR and gateway.
l
In my case, we will have to write up some custom integration either with DHCP or IPAM. If with IPAM, does it use whereabouts CNI IPAM? Does it auto detect IP conflicts with existing IPs inside the same VLAN? I assume there’s no way to pre-reserve an existing IP inside the VLAN via IPAM?
r
To me as a user, Harvester doesn't seem to manage the addresses in that way. It's not providing DHCP.
I'm looking at this diagram to understand how it works. https://docs.harvesterhci.io/v1.1/networking/deep-dive
l
I know it has to use an outside DHCP server. But for IPAM, it seems to have its internal IPAM.. probably something from kubevirt project.
yeah in the #example-2 in the diagram, it has:
Copy code
# Example 2
{
    "cniVersion": "0.3.1",
    "name": "untagged-network",
    "type": "bridge",
    "bridge": "oob-br",
    "promiscMode": true,
    "ipam": {}
}
note that the ipam is empty. how to populate that? or by default that’s already working on its own??
There’s no doc on this, which is why I also opened a question issue on harvester github..
b
I don't know anything about IPAM, but we run a DCHP server on the VM VLAN...
r
Is it essential for the metal switch to be vlan aware, i.e. trunk mode, for VM networks to have isolated layer 2?
b
Not sure. In the Harvester training they say that one can use the management network also tho create VMs on it... even if it is not advised to do so. Therefore I'd guess it's not essential.
r
It's true you can create VMs directly in Harvester on the mgmt network, but that's not the case when creating cluster nodes through Rancher with Harvester Cloud Provider. Only the VM networks defined in Harvester can be assigned to those nodes, so the "untagged" pre-defined network, or a created VM network with VLAN=1 will work with an unmanaged metal switch.