gistfile1.txt

# push mode will look about like this:

if __name__ == '__main__':

    inventory = TomlInventory("inventory/inventory.toml")
    inventory.ssh([WebServerRole(), AnotherRole() ])

# if you want to run check mode, you can still access it, though it is not the default:

    inventory.ssh([WebServers(), AnotherRole() ], check=True)

# if you want --check or --apply to be CLI options, it looks like the lines below
# if you don't supply --check or --apply the script will *not* activate this way

if __name__ == '__main__':

    inventory = TomlInventory("inventory/inventory.toml", use_cli=True)
    inventory.ssh([ WebServers(), AnotherRole() ])

# "ssh" is the name of a connection plugin.  You will be able to add your own plugins.
# local mode is basically an inventory of just one host, with only one plugin

if __name__ == '__main__':

    local = Local(use_cli=True)
    local.run([ WebServers(), AnotherRole() ])

# the return structure from ".run" will be an object that contains meaningful data about the results of each
# host, and in Jupyter context should be some interesting widget, or have some helper functions to show it as
# a widget (TBD)

Controls about the degree of parallelism enabled should be key=value arguments to the inventory object, which all will inherit from a base inventory class.

for readability:

local.run([ WebServers(), AnotherRole() ])

should be equivalent to:

local.run(WebServers(), AnotherRole())

I considered role.push(inventory). This is a bit more clear, and honestly, what this allows is:

webservers = inventory.filter_groups('webservers')
webservers.ssh(role)

This is much more explicit, and eliminates the code in the role that says where it goes, and this is probably a good idea to ax.

In doing the above, this would also mean:

local.run(role)

.ssh would remain a method on an inventory, as ".filter_groups" returns a subset of inventory (same with filter_hosts).