On this post I want to show the current process of debugging for the network system role and the reasons to add Pytest as tool for the integration tests, which is my task during the summer!
Intro to the Network Role
Since Linux System Roles is executed with Ansible, it is nice to take a look on what Ansible is. According to their main page: “Ansible is an IT automation tool. It can configure systems, deploy software, and orchestrate more advanced IT tasks such as continuous deployments or zero downtime rolling updates.”
An Ansible Role is a framework that manages collections of tasks and variables independently. Each role has a particular functionality. This said, Linux System Roles are a collection of Ansible Roles that admin the configuration of common GNU/Linux subsystems. The aim of Linux System Roles is to give a consistent API to a Linux distribution and making it consistent across their releases. Therefore, the Network Role is a role capable of configuring the networks of the machines that we want to manage through Ansible.
Integration Testing on the Network Role
My main objective in this period is to be able to improve the testing of the network role. Right now, the system used to make integration testing are ansible-playbooks. These playbooks are the standard way of executing the role, but they present some problems when being used to make these tests.
Current debugging workflow
So, how to proceed if the current integration tests end with errors? First of
all, since there is not a totally reliable source on the output that tells the
developer if the role had an unexpected error. Right now, the best way of
knowing this is returning the exit code of the process with the command
If the output if different than 0, it means that there is something you need to change on your code. To make further debugging, it is convenient to execute the tests with the command:
TEST_DEBUG=1 TEST_SUBJECTS=CentOS-8-GenericCloud-8.1.1911-20200113.3.x86_64.qcow2 ansible-playbook -v -i /usr/share/ansible/inventory/standard-inventory-qcow2 playbooks/tests_ethtool_features.yml --skip-tags tests::cleanup -e network_provider=initscripts
What does this command do?
TEST_DEBUG=1- Indicates not to remove the VM that is created for the testing.
TEST_SUBJECTS=/path/to/virtual/machine- Indicates which VM will be built to test the role.
ansible-playbook -v- The command that execute the playbooks with the flag verbose
-i /path/to/inventory- The inventory flag points to a script that sets the configuration for the testing. This script enables the variables
--skip-tags test::cleanup- This flag indicates the playbook to not execute the playbook blocks that have “test::cleanup” as a tag. Adding it will make the changes stay on the VM. This way we can later enter the VM and see what went wrong.
-e network_provider=initscripts- This selects the provider we want to use to make the network changes. It can be either initscripts or nm.
Executing this command will perform the tests and also output the information we need to later use the VM again, which are:
The ssh command to enter the VM. With this we can enter the VM and check the logs of the provider we used, to get more information on the problems encountered through the execution of the playbook.
The Ansible Inventory that can be used in case we want to execute the tests again with the same VM.
An example is:
[INFO ] standard-inventory-qcow2: ssh -p 3266 -o StrictHostKeyChecking=no -o UserKnownHostsFile=/dev/null -i /tmp/inventory-clouduyfw03ub/identity firstname.lastname@example.org [INFO ] standard-inventory-qcow2: export ANSIBLE_INVENTORY=/tmp/inventory- clouduyfw03ub/inventory
Aiming for a better integration testing
Although integration testing can be done this way, introducing a tool like Pytest can help reduce debugging time and complexity. If Pytest is added, then Ansible would be no longer needed for the testing, since Pytest can simulate the changes performed by the Ansible program. Some of the benefits of using it, among others, would be:
Pytest is a widely known tool, so using it for integration testing would make the project more accesible to new developers.
The use of different Pytest features like fixtures can easily reduce writing complexity of tests and expand the coverage of the tests at the same time.
The introduction time to the project has helped me realize why it is important to improve the testing of the roles. The following weeks will be a complete challenge to me, but I think that the outcome of it will be completely worth it, both for me and for the project. =)