Files
cozystack/hack/testdata
Aleksei Sviridkin 9efe78ce44 test(e2e): hold the tenant-cluster install wait to an ordering rule (#3802)
A test that creates a tenant `Kubernetes` cluster must not end while
that release is still installing. The kind drops the install wait on its
main resources, but the chart's blocking post-install hook still holds
the install open, so a test that asserts and ends hands the rest of it
to teardown. Teardown then reds on a step that tests nothing. Reading
the cluster mid-install is fine. Ending mid-install is not.

`hack/e2e-chainsaw/_lib/run-kubernetes.sh` already waits for the parent
release before its tenant-cluster assertions, and that wait is older
than this change. Nothing held it there. This adds the guard that does.

The guard reads the script, not Chainsaw fixtures: the cluster comes
from a heredoc, so a guard walking `apply.file` would find nothing to
check and pass. It requires the CR, then a readiness wait on the parent
release, then the assertions the test ends on. The wait has to block,
and it has to be unconditional in the body that holds it. One more case
holds this script to being the only creator of that kind, since a second
one would be governed by no rule here.

The ordering is read within one function body, so splitting
`run_kubernetes_test` reds the unit lane until the rule is pointed at
the new shape.

Timeout length stays with review. The floor depends on where the wait
sits, and the script text does not carry that.

Which spellings each matcher takes, which rewrites it refuses instead of
judging, and the blind spots it names are written in the guard's header
next to the code, each with its mechanism and whether it fails loudly or
quietly. Fixtures mark their own lines, so a probe resolves by name
instead of by a line number that drifts.

No runtime code here: the guard, its fixtures, and the conventions that
state the rule.

```release-note
NONE
```
2026-09-22 19:00:52 +04:00
..