6.8 KiB
Testing Standards in K3s
Testing in K3s comes in 5 forms:
This document will explain when each test should be written and how each test should be generated, formatted, and run.
Note: all shell commands given are relative to the root k3s repo directory.
Unit Tests
Unit tests should be written when a component or function of a package needs testing. Unit tests should be used for "white box" testing.
Framework
All unit tests in K3s follow a Table Driven Test style. Specifically, K3s unit tests are automatically generated using the gotests tool. This is built into the Go vscode extension, has documented integrations for other popular editors, or can be run via command line. Additionally, a set of custom templates are provided to extend the generated test's functionality. To use these templates, call:
gotests --template_dir=<PATH_TO_K3S>/contrib/gotests_templates
Or in vscode, edit the Go extension setting Go: Generate Tests Flags
and add --template_dir=<PATH_TO_K3S>/contrib/gotests_templates as an item.
To facilitate unit test creation, see tests/util/runtime.go helper functions.
Format
All unit tests should be placed within the package of the file they test.
All unit test files should be named: <FILE_UNDER_TEST>_test.go.
All unit test functions should be named: Test_Unit<FUNCTION_TO_TEST> or Test_Unit<RECEIVER>_<METHOD_TO_TEST>.
See the etcd unit test as an example.
Running
go test ./pkg/... -run Unit
Note: As unit tests call functions directly, they are the primary drivers of K3s's code coverage metric.
Integration Tests
Integration tests should be used to test a specific functionality of k3s that exists across multiple Go packages, either via exported function calls, or more often, CLI comands. Integration tests should be used for "black box" testing.
See integration/README.md for more info.
Smoke Tests
Smoke tests are a collection of tests defined under the tests path at the root of this repository.
The sub-directories therein contain fixtures for running simple clusters to assert correct behavior for "happy path" scenarios. These fixtures are mostly self-contained Vagrantfiles describing single-node installations that are easily spun up with Vagrant for the libvirt and virtualbox providers:
- Install Script ➡️ on proposed changes to install.sh
- CentOS 7 (stand-in for RHEL 7)
- Rocky Linux 8 (stand-in for RHEL 8)
- Leap 15.3 (stand-in for SLES)
- MicroOS (stand-in for SLE-Micro)
- Ubuntu 20.04 (Focal Fossa)
- Control Groups ➡️ on any code change
- mode=unified (cgroups v2)
- Fedora 34 (rootfull + rootless)
- mode=unified (cgroups v2)
- Snapshotter ➡️ on any code change
When adding new installer test(s) please copy the prevalent style for the Vagrantfile.
Ideally, the boxes used for additional assertions will support the default virtualbox provider which
enables them to be used by our Github Actions Workflow(s). See:
Framework
If you are new to Vagrant, Hashicorp has written some pretty decent introductory tutorials and docs, see:
- https://learn.hashicorp.com/collections/vagrant/getting-started
- https://www.vagrantup.com/docs/installation
Plugins and Providers
The libvirt and vmware_desktop providers cannot be used without first installing the relevant plugins
which are vagrant-libvirt and
vagrant-vmware-desktop, respectively.
Much like the default virtualbox provider these will do
nothing useful without also installing the relevant server runtimes and/or client programs.
Environment Variables
These can be set on the CLI or exported before invoking Vagrant:
TEST_VM_CPUS(default ➡️ 2)
The number of vCPU for the guest to use.TEST_VM_MEMORY(default ➡️ 2048)
The number of megabytes of memory for the guest to use.TEST_VM_BOOT_TIMEOUT(default ➡️ 600)
The time in seconds that Vagrant will wait for the machine to boot and be accessible.
Running
The Install Script tests can be run by changing to the fixture directory and invoking vagrant up, e.g.:
cd tests/install/rocky-8
vagrant up
# the following provisioners are optional. the do not run by default but are invoked
# explicitly by github actions workflow to avoid certain timeout issues on slow runners
vagrant provision --provision-with=k3s-wait-for-node
vagrant provision --provision-with=k3s-wait-for-coredns
vagrant provision --provision-with=k3s-wait-for-local-storage
vagrant provision --provision-with=k3s-wait-for-metrics-server
vagrant provision --provision-with=k3s-wait-for-traefik
vagrant provision --provision-with=k3s-status
vagrant provision --provision-with=k3s-procps
The Control Groups and Snapshotter tests require that k3s binary is built at dist/artifacts/k3s.
They are invoked similarly, i.e. vagrant up, but with different sets of named shell provisioners.
Take a look at the individual Vagrantfiles and/or the Github Actions workflows that harness them to get
an idea of how they can be invoked.
Performance Tests
Performance tests use Terraform to test large scale deployments of K3s clusters.
See perf/README.md for more info.
End-to-End (E2E) Tests
E2E tests cover multi-node K3s configuration and administration: bringup, update, teardown etc. across a wide range of operating systems. E2E tests are run nightly as part of K3s quality assurance (QA).
See e2e/README.md for more info.
Contributing New Or Updated Tests
We gladly accept new and updated tests of all types. If you wish to create
a new test or update an existing test, please submit a PR with a title that includes the words <NAME_OF_TEST> (Created/Updated).