Page last updated:
This topic explains recommended testing practices for tile developers.
Good testing assures tile developers that their product installs and runs properly on diverse platforms and assures Pivotal Platform platform operators that the tile they install can provide its service successfully on their platform.
Pivotal recommends a pyramid structure for testing, starting with unit tests and stepping up to successively broader and more automated levels of integration. Pivotal uses and recommends Concourse for creating build pipelines that follow this test structure. Other continuous integration tools should also support a pyramid testing approach.
Tile Test Pyramid
For Pivotal Platform tiles, a typical test pyramid progresses as follows:
Unit tests for each tile component (e.g. service components, broker, adapter, and metrics emitter), manual by developer and in automated pipeline.
System tests of the tile’s BOSH release, including:
- Functional tests covering the main features of the service. The main features typically interact with almost all important external integration points, so these tests confirm product functionality.
- Smoke tests (lifecycle tests) for service instances that create and bind a service instance, call it from a test app, check the logs it generates, and delete it. For a typical end-to-end test sequence, see Smoke Tests below.
System tests of tile operation within Ops Manager.
- These include:
- Configuration checks that test every external configurable integration point and connection to remote servers using configured credentials
- Default checks that confirm “happy path” functionality.
- Use the Ops Manager API to verify that property blueprints in the tile metadata are correct and that they translate correctly to the BOSH manifest that Ops Manager generates.
- Use the Om tool to call the Ops Manager API programmatically from Go. Avoid the unsupported opsmgr gem that called the Ops Manager API from Ruby.
- Confirm manually that the tile wires property blueprints to the expected pane and form controls in the UI.
- Test your environment using one of the environments described in Development Environments
- These include:
Note: System tests might incur costs from using third party services, IaaS resources, etc.
Smoke tests are end-to-end lifecycle tests for service instances that you can include as post-deploy errands within a tile and also automate in Concourse or other integration platforms.
A typical smoke test runs as follows:
Create an org and space for the test to run in.
Register the tile’s service broker.
Enable service access for the created org.
Iterate through all service plans (or a subset of them) to do the following:
- Create a service instance for the plan.
- Push a test app.
- Bind the service instance to the app.
- Use the app in a way that exercises the service instance. For a data service, for example, write and read from the service instance.
- Unbind the service instance.
- Delete the service instance.
- Delete the test app.
Delete the service broker.
Delete the test org and space.
The following are general recommendations for designing and running tests on Pivotal Platform tiles:
Clean up after yourself. Leave the environment exactly as it was before the test was run.
Generate verbose logging with lots of contextual data to make troubleshooting easier.
Design test suites for re-usability by making them highly parameterizable. Important parameters include:
- External settings such as domains, creds, and certs
- Plans to test against. For example, the Redis for Pivotal Platform smoke tests use identical code for two different service plans, pre-provisioned and on-demand.
- Timeouts, numbers of retries, and other things that you need to adjust for different environments
- Switches to include or exclude portions of the tests such as generating metrics or backups
Re-use tests that exist already, for example in Concourse.
Use an example CF app that uses your service. This app can serve for testing, demoing your tile capabilities, and as a code example. See the MySQL Test App an example.
When testing manually, using the UI is better than calling the underlying API directly. Use UIs and APIs the way a customer would.