The Rancher Test Framework provides tools for writing integrations and validation tests. The framework manages interactions with the external services being tested and aids in cleaning up resources after a test is completed. The framework is organized into three disciplines: framework, clients, and extensions. The framework consists of a few core libraries used to make it easy to write homologous tests. See extensions and clients for more details.
To run rancher integration tests you will need:
- a running instance of rancher with an accessible url
- a rancher access token for the admin user
- golang 1.20
- k3d
Validation tests have the same requirements as integration tests however different suites may need credentials for cloud providers. Check the configs for your test suite for details.
Integration tests - Don't require any external configuration or access to any other external services. These should also be short running as they will run on every PR within CI. Integration tests do not need access keys to cloud providers to run.
Validation tests - Require access to external services, and needs to a config file to run them.
Extensions are functions that complete common operations used by tests. Extensions should not require much configuration or support multiple behaviors. Extensions should be simple enough that they can be used in many tests without needing to be tested themselves.
Clients are used to create the resources in the tests. The advantage of the cients is they register the cleanups with sessions, making resource cleanup trivial. There are three clients; the rancher client, dynamic client, and the k3d client. The rancher client is used to create resources on a rancher instance. Given a host url, and a bearer token the rancher client communicates directly with the rancher instance. The dynamic client is used as means of communication on a downstream cluster, and used for kubernete resource management on that cluster. The k3d client is used to create a minimal k3d cluster, this is to give the integration tests the ability to test against a downstream cluster.
Wait is used to monitor resources, and wait for specified conditions. There are multiple ways to wait for a resource. There is WatchWait that uses the watch.Interface of a resource to wait until the check function returns true. For more generic polling or waiting, the "k8s.io/apimachinery/pkg/util/wait" package can be used.
Sessions are used to track resources created by tests. A session allows cleanup functions to be registered while it is open. Once a session is closed the cleanup functions will be called latest to oldest. Sessions should be closed after a set of tests that use the same resources is completed. This eliminates the need for each test to create and tear down its own resources allowing for more efficient reuse of some resources. When pared with a client sessions are a powerful tool that can track and cleanup any resource a tests creates with no additional work from the developer.
Configuration is loaded from the yaml or json file described in CATTLE_TEST_CONFIG. Configuration objects are loaded from their associated key in the configuration file. Default values can also be set on configuration objects.
Tests should be created under tests/v2/integration or tests/v2/validation, see Integration vs Validation for details. Tests can be grouped into files and packages however the developer sees fit. When grouping tests a developer should make sure the test will be easy to find for the next developer and that it is grouped with other tests that would be run at the same time. Within the files tests should be grouped into Suites. A suite should share clients and a session across all it's tests. A Suite's tests should all be testing the same functionality and reuse suite resources. For example if we were writing tests that ensure project roles have appropriate access to project resources, a suite could contain all tests for every role and share a project across all tests. This will save time because each test will not need to create a project. Tests should always use the framework clients when possible, this will ensure that any resources created by tests will be cleaned up and not interfere with other tests. A test is responsible for cleaning up any resource it creates fully before the next test is run. See Sessions for how to track resources in tests.