Yuri's questions

The most common practice to connect VMs together is to use tap interfaces with the bridge between them. On a linux, bridge driver filters some classes of link-local multicast traffic, such as LACP frames. There is no easy mechanism to permit forwarding of LACP frames across the bridge interface and it requires manual patching of the bridge module code.

I would like to use LACP in my test network between the VMs without manual patching of linux bridge code on the host machine, there is no requirement to have connectivity between host and the VM.

I identified two potential solutions

1. socket p2p option - but it is TCP based and I have concerns about TCP over TCP performance;
2. hub - this option seems to go away very soon

What is a simplest possible solution for my need?

In all Chef examples that I've seen autoscaling works pretty easy - you provision something like DB host, then web servers. You can create as many webservers as you like - all of them are going to use the same DB host (no changes required).

But what if my recipe/node provisioning cookbook requires configuration changes on some other backend nodes? How I can do it with Chef? For example, I have existing BGP route reflector (bird), I provision new bird edge node and, as a dependency, I have to generate new iBGP peer config and reread config file on the BGP route reflector node.

What is the best practice? Should I use chef-push-jobs to push changes to BGP route-reflector nodes?