This message was deleted.
# k3s
a
This message was deleted.
g
I added this to the initial install command :
Copy code
--flannel-iface ens10 --node-ip 10.1.1.1
b
Thanks for responding. Lots of questions based on your response 🙂 1. So you did not have to set --flannel-backend? It worked with the default vxlan option? 2. Shouldn't the interface be the VLAN interface? (which they suggest calling something like "ens10.4000" 3. What about MTU settings? The VLAN has to be 1400, but the interfaces seen by the pods is 1500.
g
You are on Hetzner cloud nodes, right ?
I am on CX41 nodes, and attaching an internal network showed up as interface ens10. It worked with the options above.
If using robot nodes, I added /etc/netplan/01-netcfg.yaml (setting mtu to 1400 on vlan 4001) and used
Copy code
--flannel-iface enp7s0.4001
👍 1
b
These are dedicated root servers using robot. Setting --flannel-iface to the vlan does make the pods get MTU 1400. How many of these options do you also have to specify for the agent?
My VLAN is working as it should (I think) at least I can do curl back from one server to another just fine (as well as ping/ssh), its just the CNI that is messed up I think.
g
Ah - I missed the "bare metal" part of your post. Please ignode the first part. I only had one robot server - no agents. The full list of install options was :
Copy code
INSTALL_K3S_EXEC="server --node-ip 10.0.1.2 --flannel-iface enp7s0.4001"
b
Ok thanks. So was this a single-node cluster, or did you connect it with hcloud CX instances as agents? And did you then pass --flannel-iface when setting up the agents?
g
This is from my working notes. But I am pretty sure all I had to do was to add --flannel-iface and --node-ip on both the servers and agents. Best of luck 🙂
b
Thanks for your help, I might just need that luck :)