This message was deleted.
# k3s
a
This message was deleted.
q
So a bit of progress. Looks like the system is using Mellanox infiniband adapters and I have to install some nvidia drivers.
c
could also be related to checksum offload. that frequently causes problems with vxlan. Assuming the ports are open between nodes.
q
Just rebooting the node but it takes a little while. I've not used infiniband before. I was confused as the pod can ping out to the internet but not locally.
No joy - I'll take a look at vxlan
q
Did not work off the bat. restarting the CNI ds
That didn't work, poking around more I notice that when I run
lsmod | grep xt_phy
on my working node I get 22 instances but on the node that doesn't pass traffic work there are zero. So I think there is an issue in there with iptables. Maybe flush it out and reboot? Any clues gratefully received.
Ah physdev's came up but still no traffic flowing
I found my issue. For some reason this node is defaulting to using a different network adapter from the other nodes. Kubernetes networking is not my strongpoint. To fix the issue I setup the agent by passing in the network interface that flannel should use:
Copy code
--flannel-iface ens1