Files
cozystack/hack/e2e-apps
Aleksei Sviridkin 699e429187 test(e2e): make the tenant hairpin DNS check tolerate a node recycle
The ouroboros hairpin assertion resolves a name from inside the tenant
cluster via a single `--restart=Never` netshoot Pod and waits for it to
reach Succeeded. On the memory-constrained single-node sandbox a tenant
worker node is routinely recycled mid-test: its kubelet heartbeat lapses,
the tenant's CAPI MachineHealthCheck marks Ready=Unknown after 30s and
deletes the Machine, its KubeVirt VM and the Node, then provisions a
replacement. A bare Pod bound to the deleted node is removed by the node
controller and, having restartPolicy Never, is never recreated -- so the
probe silently vanishes, typically while its large image is still pulling
and it has never left Pending. The single-shot wait then times out and
reports "dnscheck pod did not reach Succeeded phase (last seen: <empty>)",
which is indistinguishable from a real DNS failure and flakes PRs that do
not touch virt at all.

Keep the DNS resolution asserted exactly once and fail-fast: a probe that
runs to phase Failed is the reconciliation regression this test exists to
catch and fails the test immediately. Recreate only the probe Pod, and
only for the pure-infrastructure node recycle -- when the Pod vanishes and
the node it ran on is confirmed gone or no longer Ready, retry on a
surviving node within a bounded deadline. A probe that disappears while
its node is still Ready is an unexpected deletion and fails loud rather
than being retried, so no pod-churn regression is masked, and a transient
API error while polling is not mistaken for a deleted Pod. The in-Pod dig
loop that the assertion exercises is unchanged.

Assisted-By: Claude <noreply@anthropic.com>
Signed-off-by: Aleksei Sviridkin <f@lex.la>
2026-07-08 15:06:40 +03:00
..