[Failing Test] ci-kubernetes-node-e2e-containerd-serial failing with rotating specs

Which jobs are failing?

• ci-kubernetes-node-e2e-containerd-serial (prow job name resolved via TestGrid table endpoint .query = kubernetes-ci-logs/logs/ci-kubernetes-node-e2e-containerd-serial)
  • TestGrid tab: https://testgrid.k8s.io/sig-node-containerd#ci-node-e2e-serial

Which tests are failing?

Job-level failure — no individual test row has alerted in TestGrid. The alerting rows are only the harness rows:

• ci-kubernetes-node-e2e-containerd-serial.Overall (fail_count 3)
• kubetest.Node Tests (fail_count 3)

Per-build junit shows the underlying spec failure rotates — a different test fails each run, so no single spec hits the 3-consecutive alert threshold:

• Build 2100188711563038720 (2026-09-16): ubuntu host failed [It] [sig-node] Summary API when querying /stats/summary under pressure [FeatureGate:KubeletPSI] should report Memory pressure in PSI metrics — suite timeout ([TIMEDOUT] A suite timeout occurred ... at summary_test.go:406). cos host: 0 failures.
• Build 2100158262149124096: cos host failed [It] [sig-node] GarbageCollect Garbage Collection Test: Many Restarting Containers Should eventually garbage collect containers when we exceed the number of dead containers per container. ubuntu host: 0 failures.
• Build 2100128061046394880: ubuntu host failed [It] [sig-node] StaticPod when the static pod has init container should be ready after init container is removed and kubelet restarts. cos host: 0 failures.

Since when has it been failing?

Observed failing today (2026-09-16) during the weekly TestGrid review — tab was FAILING at review time with 3 consecutive failures.

Testgrid link

https://testgrid.k8s.io/sig-node-containerd#ci-node-e2e-serial

Reason for failure (if possible)

Truncated failure_message from the TestGrid summary:

error during go run /home/prow/go/src/k8s.io/kubernetes/test/e2e_node/...o/test-infra/jobs/e2e_node/containerd/image-config.yaml: exit status 1

This is the generic kubetest wrapper error — the real failure is the ginkgo spec failure above (different spec each run). Example tail from build 2100188711563038720/build-log.txt:

command [... ginkgo --nodes=8 --skip="" --label-filter="!Flaky && Serial && NodeConformance" ...] failed with error: exit status 1
...
Step 'go run .../run_remote.go ... --image-config-file=/home/prow/go/src/k8s.io/test-infra/jobs/e2e_node/containerd/image-config.yaml' finished in 1h19m10s
...
Something went wrong: encountered 1 errors: [error during go run .../image-config.yaml: exit status 1]

Rotating single-spec failures across both ubuntu and cos hosts with a suite timeout in one run suggests job-level flakiness / timeout pressure rather than one broken test. Sibling job ci-kubernetes-node-e2e-containerd-slow shares the same wrapper signature but fails on different specs (ImageGC, MirrorPod) and started 15 days earlier (2026-09-01), so it is not clustered here — see #141806 discussion.

Anything else we need to know?

• Job history: https://prow.k8s.io/job-history/gs/kubernetes-ci-logs/logs/ci-kubernetes-node-e2e-containerd-serial
• Dashboard: sig-node-containerd only (not release-blocking).
• linked_bugs in TestGrid: empty.
• Related but distinct: open flake #140094 ([Flake] kubetest.Node Tests, same wrapper signature on sibling job ci-kubernetes-node-e2e-containerd, June) and open failing-test #141806 (ImageGC on the -slow sibling). Neither names this -serial job or its failing specs.
• No kind/failing-test + sig/node open issue found naming this job (searched job name, Summary API ... PSI, and label-filtered list).

Relevant SIG(s)

/sig node