e2e test fails if there are resouces created by the previous test

[MIBc]

Describe the feature:

The e2e test may fail in sometime, for example failing download docker image. When start test again, it will fail for some existed resource. In this situation, I have to clear resource manually.
Is there a command or shell to clear all resource the test has created?

[rhuss]

AFAIR we create all resources in a temporary namespace and then just remove that namespace after the test. So a kubectl delete ns <temp-namespace> should be enough. (@navidshaikh please me correct me when I'm wrong)

Or are you talking about cluster-wide resources ?

[MIBc]

I try delete ns. But clusterrole and clusterrolebinding still left.

[navidshaikh]

That's correct, though there are multiple test namespaces kne2etests[0-9], and the SA, ClusterRole and ClusterRoleBindings are introduced recently. We do need tear down functionality.

[rhuss]

correct, ClusterRole and ClusterRoleBindings are cluster-wide resources which need to be cleaned up seperately. I wonder, whether those cluster-wide resources (if different) work nicely with parallel running tests ? Or ar their names also randomized (so that they don't interfer which each other) ?

An idea: We could use the suffix of the namespace under test for the cluster role names and then just delete all clusterroles with those prefix. This would be generic enough and doesn't need constant maintenance.

[maximilien]

I am also seeing this issue consistently as well with ClusterRole and ClusterRoleBindings.

The kne2etestsX namespaces from previous runs are not deleted (sometimes) on failures, so will cause errors in subsequent runs of e2e. I think we need to have a clean up of these namespaces at the start of the e2e run. Like a cleanup to ensure all is well. The danger is if there is a running e2e, which we can claim is not supported (e.g., running e2e multiple times at once). Thoughts?

[MIBc]

The danger is if there is a running e2e, which we can claim is not supported (e.g., running e2e multiple times at once). Thoughts?

-- It must fail . If a test is running, another test will meet resource already exists error.