We add a new experimental feature to allow Mininet to run across
a cluster of machines. This is currently implemented via a set
mix-in classes that provide remote nodes that are implemented
via a connection to a remote shell, and remote links which are
tunnels across servers. In this preliminary implementation,
both control and data connections are made via ssh, but this
could change in the future.
A MininetCluster class is provided which allows existing code
to be used with minimal modification - all that is required is
to provide a list of servers to use. A customizable placement
algorithm may also be specified. An experimental CLI subclass
is also provided to make it easier to examine node placement;
status and links commands can also check whether nodes and
tunnels are still running.
Although this is an experimental feature, it does include a
--cluster option to make it convenient to start up a Mininet
simulation over a cluster, and a script to assist with setting
up the prerequisite authentication via ssh key pairs.
The cluster feature is preliminary and missing some obvious
important features, such as parallel startup and multiple tunnel
types, which we hope to add in the future.
It appears that under certain conditions, such as when a
namespace exits, both ends of a veth pair may get dumped
into the root namespace. We therefore now remove an interface
both from its home namespace and from the root namespace.
There are a bunch of these remaining, but I don't think the right course is
to 'fix' all of them to make pep8 happy, but instead to either change
the test in pep8 to consider that a continuation line may itself
be continued halfway, OR, to change the code in these lines to be more
readable by removing the need for all those nested continuations.
Personally, I find multiply-broken lines (aka nested continuations) really
hard to read.
Bob Lantz
backb1
cody burkard
Darshan Thaker
Brian O'Connor
Andrew Ferguson
Brandon Heller
Angad Singh
Nikhil Handigol