removed subshelling for link creation

[hellt]

Hi @karimra @steiler

I'd had a festive morning where I wanted to look into the #68 more closely. @steiler I decided to create a new PR, since I had to use different approach than you proposed in #123, I will explain why.

1 Link MTU

First, I decided to add an MTU config parameter to the Link object to allow for MTU configuration on veth links https://github.com/srl-wim/container-lab/compare/native-veth?expand=1#diff-24901324638614d6ed95797828a77247d9dad583c83606ec43cce4f7ed2e8269R129

The Link MTU defaults to 1500 if not set - https://github.com/srl-wim/container-lab/compare/native-veth?expand=1#diff-24901324638614d6ed95797828a77247d9dad583c83606ec43cce4f7ed2e8269R475-R478

2 NSPath is part of Node

I added NSPath field to Node struct. This string field contains a path to the netns file descriptor of the namespace this node(container) is in. https://github.com/srl-wim/container-lab/compare/native-veth?expand=1#diff-24901324638614d6ed95797828a77247d9dad583c83606ec43cce4f7ed2e8269R122

The NSPath is used in the veth creation code, that takes this path to put a veth endpoint into the relevant netns

3 Removed InitVirtualWiring entirely

I removed the InitVirtualWiring func, because all it was doing is cleaning up the interfaces in the host NS which might be left hanging after the previous deployments. It is not needed anymore, since the code that creates the veth interfaces now will clean them up if something goes wrong during the creation.

4 veth creation

The root of this change is in the CreateVirtualWiring func https://github.com/srl-wim/container-lab/compare/native-veth?expand=1#diff-1a8d111c070c6cf01d39d5f2935d710f1e09fe35101cd75796f8cec113133b32R24

It now has a single private function that will handle both vethToNS and vethToBridge creation methods.

It uses a new struct vEthEndpoint which defines the veth endpoint https://github.com/srl-wim/container-lab/compare/native-veth?expand=1#diff-1a8d111c070c6cf01d39d5f2935d710f1e09fe35101cd75796f8cec113133b32R24

@steiler the reason I needed to create a new PR, is because I read how CNI creates veth interfaces, and they use ns package (github.com/containernetworking/plugins/pkg/ns) that has a Do() function that wraps all the thread locking techniques. That enables us to not tackle that ourselves and rely on a proven techniques.

I did test the new code on a large scale deployment (20,4,2 scenario) and it worked faster than original subshelling method. But it would be good if you try it yourselves.

close #68, #123

[hellt]

Thank you for your thorough review @karimra!