search

nabla-containers / runnc

OCI-interfacing Container runtime for Nabla Containers
https://nabla-containers.github.io/
ISC License
257 stars 22 forks source link

Fix Network Issue #86

Open lestich opened 4 years ago

lestich commented 4 years ago

Got Nabla-Containers running on Kubernetes with RuntimeClass and containerd 1.3.4. I faced the same issue like #83 with every nabla-image. Removing the deletion of the MasterIP (eth0) solved the problem for me. Another Problem was the already existing network-bridge so i built in a little check for br0. This fix worked for me, i tested it with docker and k8s/containerd/calico

Closes #83

frzifus commented 4 years ago

check: https://travis-ci.com/github/nabla-containers/runnc/jobs/365877633#L346

lestich commented 4 years ago

Build System seems to be broken. Dep-related errors

lestich commented 4 years ago

Althoug it works with this fix it's a quite hacky solution. Eth0 keeps its ip address with a placeholder MAC and the Unikernel Interface (ukvmif0) gets the same IP with the MAC address from ETH0. So now we have the same IP address twice with different MAC Adresses. Maybe it's just working because Calico is smart ;).

We need to find some way to save the IP Address of the veth-interface inside the nabla network namespace so we can delete it as proposed in the original source code. Maybe we could create a dummy network interface that keeps the IP and MAC ?

The problem described in #83 occurred to me when the runtime crashed or could not start at all. K8s tries to restart the Container in the same network namespace but now the IP Adress from (veth) eth0 is missing (because it was deleted in the run before) and it keeps crashing ("master should have an ip").