Join GitHub today
GitHub is home to over 28 million developers working together to host and review code, manage projects, and build software together.Sign up
UDP backend flannel0 interface uses a network address #1049
When running in a kubernetes cluster with the UDP backend configured via kubeadm with a pod CIDR (say 10.244.0.0/16) and a matching flannel manifest with 10.244.0.0/16 as a network flannel creates a flannel0 interface on the master node using 10.244.0.0/16 as an IP address. This prevents traffic from other minion nodes' containers back to the master, as it seems that minion hosts (which use 10.244.X.0/16 as their flannel0 interface IP) see the 10.244.0.0/16 as a network address and drop the packets instead of sending them back.
Manually adding a secondary IP address on the flannel0 interface on the master node and trying a ping through that results in traffic getting back to the master node.
Connectivity between containers running on the minion nodes and the master's flannel0 interface should be possible.
Packets get dropped due to flannel0 using a network address as an ip address on the master node.
I guess that starting from the first available host should work.
Steps to Reproduce (for bugs)
We're running local k8s clusters with flannel for container networking