Is there an easy way to generate a graph of Ansible role dependencies?

Question

Since version 1.3, Ansible has supported role dependencies to encourage reuse of role definitions. To audit and maintain larger orchestrations, it would be nice to have some way to easily generate a dependency graph of which roles depend on which other roles.

An example of dependency definitions might be roles/app_node/meta/main.yml:

---
dependencies:
  - { role: common, some_parameter: 3 }
  - { role: apache, port: 80 }
  - { role: postgres_client, 
      dbname: blarg, 
      other_parameter: 12 }

where roles/postgres_client/meta/main.yml might include something like

---
dependencies:
  - { role: postgres_common }
  - { role: stunnel, 
      client: yes,
      local_port: 5432
      remote_host: db_host
      remote_port: 15432 
    }

Such nested dependencies can get messy to maintain when the number of roles in an orchestration grows. I therefore wonder if anyone has found an easy way to generate a graph of such dependencies, either graphically (dot or neato?) or just as an indented text graph? Such a tool could help reduce the maintenance complexity.

Sorbose

Answer 1

The following python script worked for me:

Needs graphviz ( gv below ): pip install graphviz

#!/usr/bin/env python

import sys
import gv
from glob import glob
import yaml

g = gv.digraph('roles')

role_nodes = {}

def add_role(role):
    if role not in role_nodes:
        role_nodes[role] = gv.node(g, role)

def link_roles(dependent, depended):
    gv.edge(
        role_nodes[dependent_role],
        role_nodes[depended_role]
    )

for path in glob('roles/*/meta/main.yml'):
    dependent_role = path.split('/')[1]

    add_role(dependent_role)

    with open(path, 'r') as f:
        for dependency in yaml.load(f.read())['dependencies']:
            depended_role = dependency['role']

            add_role(depended_role)
            link_roles(dependent_role, depended_role)

gv.layout(g, 'dot')
gv.render(g, 'png', 'doc/ansible-roles.png')

Vested

This answer could be improved by explaining how it addresses the problem stated in the question. – Devilment

Running: "ImportError: No module named gv". pip install gv: "No distributions at all found for gv" – Psychosurgery

@TravisBear The gv module is part of Graphviz, it is not distributed on PyPI. You can install it using the package manager of your linux distribution (apt, yum...). Package name may be something like graphviz-python or python-graphviz. – Vested

I've made an alternative version using graphviz python package – Lettered

on SUSE : sudo zypper install python3-pygraphviz And then in .py files import graphviz as gv – Enchorial

Answer 2

You can use Graphviz to do this like the ginas project has.

Alessi

Answer 3

# install gv on ubuntu
sudo apt-get install python3-gv

Ambler

This answer could be improved by showing how python-gv answers OP's problem. – Merralee

I think this should actually just be added to sebn's answer. It seems to be the missing piece for most of those who don't get how to install gv. – Sorbose