Fix multi-host provisioning

[apatard]

This patchset is fixing multiplatform support. Until now, when multiple instances are declared, the create.yml playbook is looping over the instance list and calls the vagrant module. This leads to one Vagrantfile per instance which are overwritten at each step of the loop.
This make debugging harder and - in some cases - may confuse vagrant.
This patchset is allowing to directly give molecule.yml instances list to the vagrant module and then the module is producing a proper Vagrantfile.

[apatard]

recheck

[sio]

Vagrant Cloud is broken at the moment, these CI failures are not because of your changes: hashicorp/vagrant#12390

[apatard]

@sio oh... Thanks for the info. I will refrain me from sending new PR or recheck then.

[sio]

I didn't mean to dissuade you from submitting new PRs, I was just letting you know that CI will not be functional for some time due to upstream service provider outage :)

It took me quite some time this morning to understand that the problem is not on my end and I wanted to save you the same trouble.

[apatard]

recheck

[apatard]

hm. weird. Probably not related to this PR but ...

2021-05-26T07:59:49.8041450Z TASK [Bootstrap python for Ansible] ********************************************
2021-05-26T07:59:49.8151860Z Wednesday 26 May 2021  07:59:49 +0000 (0:00:00.065)       0:00:00.065 *********
2021-05-26T08:00:06.3782880Z �[32mok: [instance]�[0m

So the prepare.yml did install python but

2021-05-26T08:00:08.6030260Z TASK [sample task] *************************************************************
2021-05-26T08:00:08.6045050Z Wednesday 26 May 2021  08:00:08 +0000 (0:00:00.039)       0:00:00.039 *********
2021-05-26T08:00:09.4077040Z �[1;35m[WARNING]: No python interpreters found for host instance (tried�[0m
2021-05-26T08:00:09.4098840Z �[1;35m['/usr/bin/python', 'python3.9', 'python3.8', 'python3.7', 'python3.6',�[0m
2021-05-26T08:00:09.4157470Z �[1;35m'python3.5', 'python2.7', 'python2.6', '/usr/libexec/platform-python',�[0m
2021-05-26T08:00:09.4160370Z �[1;35m'/usr/bin/python3', 'python'])�[0m
2021-05-26T08:00:09.4163550Z �[31mfatal: [instance]: FAILED! => {"ansible_facts": {"discovered_interpreter_python": "/usr/bin/python"}, "changed": false, "module_stderr": "Shared connection to 127.0.0.1 closed.\r\n", "module_stdout": "/bin/sh: /usr/bin/python: not found\r\n", "msg": "The module failed to execute correctly, you probably need to set the interpreter.\nSee stdout/stderr for the exact error", "rc": 127}�[0m
2021-05-26T08:00:09.4165610Z

Given that the scenario is the default configuration, aka:

---
dependency:
  name: galaxy
driver:
  name: vagrant
  provider:
    name: libvirt
platforms:
  - name: instance
provisioner:
  name: ansible

I'm not convinced that setting the box to use the test one is a good idea....

[yajo]

I've been testing this PR in my local dev workflow.

Without it, I was almost about to dump molecule-vagrant. It became too complicated.

With it, it works like a charm. VMs spin up faster. If I need to debug anything or connect to one of them, I just have to cd to the scenario ephemeral directory and run vagrant ssh server0.

Non-deep code review looks good also.

Thanks! This is a must have. 😃

[apatard]

With it, it works like a charm. VMs spin up faster. If I need to debug anything or connect to one of them, I just have to cd to the scenario ephemeral directory and run vagrant ssh server0.

hm. molecule login is not working for you ?

[yajo]

hm. molecule login is not working for you ?

I didn't know it existed 😅

It's working great! (I mean: without the PR)

[yajo]

In any case, I still think this different approach is a good enhancement. It is more similar to what one would do if not using molecule, and is faster.

One problem I noticed is that, when using this PR, machines were created without a network interface.

When using a4bd8b1, they got created with networking.

My platforms definition is:

platforms:
  - &server
    box: generic/ubuntu2004
    groups:
      - k8s_server
      - k8s_node
    name: server0
  - <<: *server
    name: server1
  - <<: *server
    name: server2

[apatard]

@yajo Networking should work. Can you share your molecule.yml so that I can try to reproduce and fix over the week ? It would be bad to see this PR merged if there's a know bug...

[yajo]

Here it is:

dependency:
  # name: galaxy
  # options:
  #   role-file: requirements.yaml
  #   requirements-file: requirements.yaml
  name: shell
  command: ansible-galaxy install -r requirements.yaml
driver:
  name: vagrant
  provider:
    name: libvirt
platforms:
  - &server
    box: generic/ubuntu2004
    groups:
      - k8s_server
      - k8s_node
    name: server0
  - <<: *server
    name: server1
  - <<: *server
    name: server2
provisioner:
  name: ansible
  connection_options:
    ansible_become: true
  config_options:
    defaults:
      vault_password_file: ${MOLECULE_PROJECT_DIRECTORY}/.vault_password.txt
  inventory:
    links:
      group_vars: ../../group_vars
verifier:
  name: ansible

[apatard]

@yajo tried quickly your setup and I can use molecule login to connect so the network seems to work. Can you have a look at the vagrant .err and .out log files to see if there's a clue ?

[yajo]

It seems related to using libvirt.qemu_use_session = true which seems to have some problems on creating network interfaces. Not something specific to this PR.

[yajo]

Definitely the issues I'm experiencing are lower in the stack. Check vagrant-libvirt/vagrant-libvirt#1342 if you're interested. But nothing related to this PR.