When server instances are provisioned in the cloud, they are rarely set up manually, but automated using provisioning scripts. the provisioning scripts describe how the server should be set up. this is normal code, as we know it from normal software development, except that this code is specialized on the domain infrastructure. In software development, static code analysis (“linting”) and automated tests have established themselves as means for good maintainability of the production code. Why not apply this good practice to “Infrastructure as code” as well? This article uses the provisioning tool Ansible to describe how automated quality assurance for provisioning scripts can look like.
Brief Introduction to Ansible Ansible
describes itself as a tool for configuration management, software distribution, and for executing ad hoc commands1. Ansible is written in Python. To use it, the developer does not necessarily need to have Python knowledge, since the Ansible scripts are written in YAML. YAML is an abstraction of JSON with the difference that it is more readable2. The ansible scripts are called “Playbooks”. In order to use Ansible, Python must be installed on the target machines and access to these machines must be possible via SSH. Ansible must only be installed on the machine running the Playbooks (“host machine”); it must be a Linux system. For target machines running Windows, Ansible provides rudimentary support. When the Playbooks written in YAML are executed, Ansible translates them into Python scripts and executes them on the target machines. They are then removed from the target machines. A typical Ansible Playbook looks like this:
How a QA Pipeline Works for Ansible Playbooks
In software development, static code analysis and automated tests are performed continuously in a pipeline on a CI server. This concept is to be reused for Ansible Playbooks. First, the developer stores the provisioning scripts in a version control system. Changes to the scripts that are checked in to the version control system trigger a pipeline. A pipeline for provisioning scripts runs through several steps (see also figure 1): First, a lint checks the code to see whether it is syntactically correct and whether it follows best practice. If the lint has nothing to complain about, a test environment is prepared. The provisioning scripts run against these test environments. If they have run without errors, tests are run to verify that everything has been configured as expected. At the end, the test environment is destroyed again.
Figure 1. All steps in a pipeline
Lint are tools that perform static code analysis.
In the following, the individual steps are presented in more detail and how they are then integrated into a pipeline. This is demonstrated using a git repository that includes an Ansible playbook, scripts to build the test environment, the tests, and a description of the pipeline. The Ansible playbook will install an OpenJDK 8 and a Tomcat instance. The complete project can be found on Github4. The starting point is the Git repository, which initially only receives the Ansible Playbook.
Static code analysis with ansible-lint
First, a lint checks whether the Ansible Playbooks are syntactically correct and whether they are written according to best practice rules. For Ansible Playbooks there is the lint
ansible-lint5. It is executed on the CLI (listing2):
In this example,
ansible-lint finds several rule violations. If single rules should not be checked, there are two ways to exclude them. The rule can be switched off either globally for the whole project, or for individual cases (so-called false-positive cases). For the global setting a
.ansible-lint file is placed in the root directory of the project.
In this configuration file you maintain an
In this configuration file further behavior can be configured, e.g. in which path they are stored. More information can be found on the project page4. If the developer wants to exclude false-positive cases from the check, she leaves a comment in the Ansible Playbook.
If the developer lacks further rules, then she can define further rules herself with the help of Python scripts (see also documentation5). Listing 3.
Setting up and Destroying the Test Environment with Terraform
After the lint has successfully passed, the test environment for the Ansible Playbooks should be built. This is done with the help of Terraform6 and the Hetzner Cloud7. Terraform helps developers to provision cloud infrastructure with code. Before the developer can get started with the Terraform script, she must store a public SSH key in the Hetzner Cloud account and generate an API token. The Public SSH Key is later stored in the server instance to be created, so that Ansible can connect to this instance.
With the help of Terraform, the developer describes which server type she wants, which operating system, in which location the server instance is hosted and what is to be provisioned as the basic setup. For this purpose, the developer creates a
testinfrastrucuture.tf file in the root directory of the project (listing 4).
As a basic setup, the developer specifies which public SSH key should be stored on the server (
ssh_keys) and that Python should be installed (
provisioner remote-exec). The public SSH key and Python are needed so that later Ansible can execute its scripts on this server. Since the test server instance is to be operated in the Hetzner Cloud, Terraform must install the required provider plug-in. The developer calls
terraform init in the folder containing
testinfrastructure.tf. Then everything is ready to provision the server instance.
The developer must give the
apply command the variable
hcloud_token with the API token, which the developer generated before in the Hetzner Cloud console. As soon as the server is available, the developer can execute the Ansible Playbooks and the tests against this server. Regardless of the success of the Playbooks or the tests, the server instance is destroyed with the help of Terraform. The
destroy command also requires the API token.
Running Ansible Playbooks
After the developer has created the server instance with Terraform, she executes the Ansible Playbook against this server instance. In a classical infrastructure, the server instance would have a fixed IP address and the developer would have entered this IP address in the Ansibles static inventory so that Ansible knows which server to connect to static Inventory.
Since the servers in the cloud are assigned a new IP address each time they are deployed, the developer cannot use the static inventory in this case. Since the developer goes against the Hetzner Cloud, she uses the Ansible Inventory Plugin
hcloud. Therefore an
inventory folder with the file
test.hcloud.yml has to be created.
The suffix <code>hcloud.yml</code> is important for the file name. test.hcloud.yml
The developer then has to enter the correct server name in the Playbook under
hosts, which she previously defined in the terraform script. Ansible-Playbook Snippet
When running the Playbooks the developer has to make sure that she gives the correct Private SSH key and that the API token is defined in the system environment variable
The API token can also be defined by the developer in the inventory file
test.hcloud.yml. However, since this file is stored in a version control system, it is not advisable to do this, since no credential should be stored in a VCS.
Functional Tests with Testinfra
After the test environment has been set up and the Playbooks have passed successfully, tests should be run to check if the openjdk package has been installed and if the file
/opt/tomcat/bin/catalina.sh exists on the server. There are some test frameworks for provisioning tools such as ServerSpec8, Goss9 and Testinfra10. The main difference between the test frameworks is the syntax in which the tests are written. For ServerSpec the tests are described in Ruby syntax, for Goss in YAML syntax and for testinfra in Python syntax. The way the test frameworks work is the same. They connect to a server that was previously provisioned with provisioning tools and check whether the server provisioning corresponds to the test descriptions. Here, the tests are written with Testinfra. To do this, the developer creates the
tests folder in the root directory of the project. This is where the tests are stored.
The developer writes the tests in
test_tomcat.py. (listing 5)
Testinfra comes with ready-made modules that simplify the test description. In these tests the developer uses e.g.
host.package to query the Package Manager or
host.file to test the existence of a certain file. Testinfra supports several ways to connect to the server. One way is to reuse Ansible’s connection configuration. Since Ansible uses a dynamic inventory here, and Testinfra cannot read all the information from that dynamic inventory, the developer must explicitly enter some configurations in the Ansible configuration file
ansible.cfg. Listing 6.
This file is stored in the root directory of the project. The developer can then run the tests. (listing 7)
After each step in the pipeline has been considered individually, the steps are to be merged into a pipeline in CI-Server Jenkins. The pipeline is described by the developer in a Jenkins file. This Jenkins file describes four stages and one post-action. One stage each for the code check, build test environment and execute playbooks and the last stage for the execution. In the post-action, the test environment is dismantled, regardless of whether errors occurred in the stages (listing 8).
In order for this pipeline to work, the developer must deposit the API token of the Hetzner Cloud in the Jenkins. So that the token is not stored in plain text, the developer stores it in the
Credential area as
Secret Text and assigns an ID, which he can then retrieve using the
credential method in the Jenkins file (here:
Simplification for Ansible Role
For Ansible Role, a structuring option in Ansible to reuse Playbooks across multiple projects, there is a simplification for this presented pipeline. With Ansible Role, the developer can use Molecule to configure the complete pipeline and the presented tools in one go. She then only needs one command (
molecule test) to execute the complete pipeline. A very good introduction to Molecule is given in the blog post “Test-driven infrastructure development with Ansible & Molecule”11,12 by Jonas Hecht.
This article provides an overview of how to build a quality assurance pipeline for Infrastructure as Code.
3. https://www.sandra-parsick.de/publication/ ansible-fuer-dev/ (German)
4. https://github.com/sparsick/ansible-testing-article/ tree/cloudreport19
12. https://blog.codecentric.de/en/2018/12/ test-driven-infrastructure-ansible-molecule/
Sandra Parsick works as a freelance software developer and consultant in the Java environment. Since 2008 she is working with agile software development in different roles. Her focus is on Java enterprise applications, agile methods, software craftsmanship and the automation of software development processes. She enjoys writing articles and speaking at conferences. In her spare time Sandra Parsick is involved in the Softwerkskammer Ruhrgebiet, a regional group of the Software Craftmanship Community in the German speaking area. Since 2019 she is a member of the Oracle Groundbreaker Ambassador Program.