How to share platforms between scenarios to avoid duplication-induced errors?

[ng-bsy]

I've got a role I want to call N times with different parameters on a fixed set of hosts and test/verify the different outcomes.

It is my understanding, that I have to use N different scenarios for this, calling the role with according parameters in converge.yml and verifying the outcomes in verify.yml

Every parameter - except the scenario name - in molecule.yml of each scenario is exactly the same, as far as I am concerned, though I can't be sure, because I had to copy them N times, make some adjustments, like adding another platform/host, and may have forgotten to adjust some of them.

To avoid problems like this which are ultimately the result of unnecessary duplication I would like to share the platforms & other parameters between these N scenarios.

I found in the documentation, that for inventory I could use the links parameter.
But how would I do that for platforms and maybe other parameters? Searches in documentation and the web have been fruitless to the day.

Or maybe I'm wrong in using multiple scenarios?

[mafalb]

If going the multiple scenario route -c, --base-config could be one answer to your question. you still need a molecule.yml in your scenario directory, but it could be empty if you don't need to override.

another idea would be to parameterize your scenario from outside with environment variables and stick with one scenario
e.g.:

$ ARG1=bla ARG2=bla molecule test

in your molecule.yml you would have

provisioner:
  name: ansible
  ansible_args:
    - -e arg1='${ARG1}' -e arg2='${ARG2}'

you could also put the shell variable in an env file.

$ cat pathtoenvfile
ARG1: bla
ARG2: bla

$ molecule -e pathtoenvfile test

Another method is using one scenario with multiple platforms, but that does not scale well with big N.
That's my point of view, hope that helps a little bit.

[ng-bsy]

Thanks a lot for your answer,
though none of the options seems to be a good fit.

Molecule is actually executed inside a CI pipeline for a whole collection with multiple roles and each of them has multiple scenarios each with their own set of hosts/configs.

So I would prefer a molecule (config) only approach without having to complicate the CI pipeline with additional logic to accommodate for something that, in my opinion, should be handled by molecule.

I am indeed surprised not to have found such a functionality in molecule, as it seems quite integral to me for a testing framework.

[isuftin]

You can put your base config into your project's .config/molecule/config.yml and define your base configuration there. I do this for dependency, driver, provisioner, etc.

The problem with defining your platform there is that once you override any parameters, you end up having to override the whole thing anyway :/

We also run multiple systems in parallel in CI. An example of our scenario-specific platform config looks like (don't sweat the non-standard parameters, we created our own ec2-driver create/destroy scenarios:

platforms:
  - name: '${MOLECULE_SCENARIO_NAME}-${PLAYBOOK_NAME}-${MOLECULE_ANSIBLE_VERSION:-localtest}'
    region: ${AWS_REGION}
    vpc_id: ${AWS_VPC_ID}
    vpc_subnet_id: ${AWS_SUBNET}
    instance_type: ${AWS_INSTANCE_TYPE}
    security_groups: ${AWS_SECURITY_GROUPS}
    aws_iam_role: ${AWS_IAM_ROLE}
    tags:
      project_id: enterprise
      molecule_pipeline_id: ${CI_PIPELINE_ID:-manual}
      ansible_managed: 'true'
      Name: Ansible Molecule - ${MOLECULE_SCENARIO_NAME}-${PLAYBOOK_NAME}-${MOLECULE_ANSIBLE_VERSION:-localtest}
    image: ${AWS_AMI_RHEL_9}
    key_inject_method: ec2
    groups:
      - account_development
      - os_rhel_9
      - aws_ec2
      - molecule
      - vault_molecule
      - hashicorp_vault_molecule
      - all

The only difference between the scanerio platforms is the groups parameter. Everything else is (or should be) either dynamically set via the .env file in the root project directory or driven by the parallel matrix variables in our GitLab ci.

So if we didn't have to define the entire platform due to how Ansible merges dictionaries, we could just get away with:

platforms:
    groups:
      - account_development
      - os_rhel_9
      - aws_ec2
      - molecule
      - vault_molecule
      - hashicorp_vault_molecule
      - all

But I don't think the merging is additive, so we do have to redefine the whole thing. Someone please let me know if I'm wrong. I'll actually play with this today to see if I am ;)

[c33s]

my config with example files and directory tree. hopfully saves someone some time.

• the global config can be at ~/.config/molecule/config.yml
• project_root means the next parent git repo. so you have to create .config on the same level as the .git/vcs folder
• per call config molecule --base-config molecule/common/base_config.yaml test (--base-config can be repated according the manual)

project_root example:

.git
.config/molecule/config.yml

my files:

.config/molecule/config.yml

---
driver:
  name: podman

platforms:
  - name: instance
    image: debian:12
    pre_build_image: true
    privileged: true

provisioner:
  name: ansible
  playbooks:
    prepare: ../common/prepare.yaml
    converge: ../common/converge.yaml

verifier:
  name: testinfra
  options:
    verbose: true

ansible/roles/wireguard directory structure

├── molecule
│     ├── common
│     │     ├── base_config.yaml
│     │     ├── converge.yaml
│     │     └── prepare.yaml
│     ├── custom_network
│     │     ├── molecule.yml
│     │     └── tests
│     │         └── test_default.py
│     └── default
│         ├── molecule.yml
│         └── tests
│             └── test_default.py
└── tasks
      └── main.yaml

default/molecule.yml is empty

custom_network/molecule.yml

---
verifier:
  options:
    extra_vars:
      wireguard_address: 10.99.99.1/24