Skip to content

Tests in PR explained

PR Continuous Integration

Operators submitted to this repo are automatically tested on a Kubernetes cluster before being merged. The Kubernetes distribution used for testing depends on which directory the operator is submitted to. Ideally, all tests should pass before merging.

You can test operators locally using following documentation.

CI test scripts

Test scripts are written in Ansible and located in our upstream-community branch.

There are 3 test types. A list of tests is shown in the following table.

Test type Description
kiwi Full operator test
lemon Full test of operator to be deployed from scratch
orange Full test of operator to be deployed with existing bundles in quay registry
all kiwi,lemon,orange

[kiwi] - Full operator test

Full operator tests - Building bundle image - from packagemanifest format - from bundle format - Sanity check of operator version (when multiple only last test is done) - Validation using operator-sdk validate - Building a temporary catalog with one operator version in it - Deployment of an operator on kind (k8s) cluster (only for kuberbetes-operator)

[lemon] - Test of operator to be deployed from scratch

Test if deployment is possible from the scratch. It means creating bundle images and index image.

  • Build all bundle images
  • Build catalog

[orange] - Test of operator to be deployed with existing bundles in quay registry

Test if an operator can be added to index from existing bundles from production (quay.io)

  • Build a current operator version locally
  • Use older versions from quay.io
  • Build catalog

OLM

Deployment with the OLM involves creating several required manifest files to create CustomResourceDefinitions (CRD's) and the operators' Deployment using its ClusterServiceVersion (CSV) in-cluster. test-operator will create a operator-registry Docker image containing the operators' bundled manifests, and CatalogSource and Subscription manifest that allows the OLM to find the registry image and deploy a particular CSV from the registry, respectively.

Failure to successfully deploy an operator using the OLM results in a test failure, as all operators are expected to be deployable in this manner.

Scorecard

The Operator SDK scorecard suggests modifications applicable to an operator based on development best practices. The scorecard runs static checks on operator manifests and runtime tests to ensure an operator is using cluster resources correctly. A Custom Resource (CR) is created by the scorecard for use in runtime tests, so alm-examples must be populated.

The scorecard utility runs through multiple test scenarios, some of which are required and others are optional. Currently the tests are configured like this.

Mandatory tests that need to pass for the PR to be accepted:

  • checkspectest - verifies that the CRs have a spec section

  • writingintocrshaseffecttest - verifies that writing into the CR causes the Operator to issue requests against the Kubernetes API server

Recommended tests that should pass in order to have a well-behaved operator:

  • checkstatustest - verifies whether the CRs status block gets updated by the Operator to indicate reconciliation.

See the scorecard test documentation for more information.

test-operator injects a scorecard proxy container and volume into an operator's CSV manifest before deployment; this is necessary to get API server logs, from which the scorecard determines runtime test results. These modifications are not persistent, as they're only needed for testing.

Note: no explicit number of points or percentage is necessary to achieve before merging yet. These are suggestions to improve your operator.

Upstream operators

Operators submitted to the upstream-community-operators/ directory are tested against a KIND instance deployed on a [Travis CI][travis-ci] environment. The OLM is installed locally in this case.

OpenShift operators

Operators submitted to the community-operators/ directory are tested against an OpenShift 4.0 cluster deployed on AWS using the ci-operator.