It's actually a substantial re-write. They got rid of multiple hierarchies, in other words a process can now belong to only one cgroup. Threads cannot belong to cgroups different from their parent anymore.
v1 was okay given that slicing the system into partitions was rarely needed outside of VPS service provider use cases. Then container goldrush happened and user applications started using cgroups in earnest exposing some design shortcomings, or rather excessive flexibility.
As someone who maintains a container runtime (runC) I don't really agree with the statement that v1 had too much flexibility. Sometimes you want to have different resource pools for different resources. That's a useful feature. And while I understand the reasoning of the no internal process constraints, there's just some issues there that just make life annoying (you have to create two subtrees if you actually just wanted to make one). The reason some of these constraints and requirements came about is (IMO) because of bugs in cgroup controllers and the maintainers of the controllers didn't want to fix the bugs (for example, the CPU controller would weight processes in an internal node differently to processes in leaf nodes).
However, the fact that you can only have one hierarchy is a good thing (it makes container setup much nicer). Too bad that systemd will probably assume that it should own the root hierarchy (which will probably result in countless conflicts with container runtimes that don't buy into nspawn).
