What
The ghost-cluster bluefin-qa-pipeline Argo workflow (podman nested-systemd + qecore-headless path) runs the smoke suite for bluefin PR #837 and ends with:
test-lane(0:smoke): Failed, exit 1result-summary.txt: "No results generated" (the default fallback)results.json: never written
validate-suites (the suite-name gate) passes; the failure is downstream in the qecore/behave execution.
Evidence
Two runs, both Failed at the workflow level, validate-suites Succeeded:
blu-837-...-kvlxb Failed (test-lane(0:smoke) exit 1)
blu-837-...-ljb9t Failed (test-lane(0:smoke) exit 1)
Node message: main: Error (exit code 1), output result: "No results generated".
Fix hypothesis
The run-container-tests template reaches the else printf 'Execution failed' >/tmp/results/result-summary.txt branch, meaning /tmp/results/results.json was never produced. Likely causes:
qecore-headless exits non-zero before behave writes results.jsonbehave --outfile /tmp/results/results.json never runs because the GNOME Shell AT-SPI wait or gnome-ponytail-daemon fails first- The behave run crashes before opening the output file
Pod logs are GC'd (podGC: OnWorkflowSuccess), so the fastest next step is to reproduce with podGC disabled and capture qecore-headless stderr.
PR #319 adds a release-gate calling testsuite/e2e.yml — a different path (QEMU on GHA ubuntu-latest, not the cluster podman path), so this bug does not directly block #319's gate. But bluefin#837 is the consumer-validation PR for #319, and its cluster smoke CI is red because of this testsuite-infra issue.
Automatable: yes — the fix is in the testsuite harness, not the image.
Filed from a pi agent session reviewing projectbluefin/actions open PRs.
What
The ghost-cluster
bluefin-qa-pipelineArgo workflow (podman nested-systemd + qecore-headless path) runs the smoke suite for bluefin PR #837 and ends with:test-lane(0:smoke): Failed, exit 1result-summary.txt:"No results generated"(the default fallback)results.json: never writtenvalidate-suites(the suite-name gate) passes; the failure is downstream in the qecore/behave execution.Evidence
Two runs, both
Failedat the workflow level,validate-suitesSucceeded:Node message:
main: Error (exit code 1), outputresult: "No results generated".Fix hypothesis
The
run-container-teststemplate reaches theelse printf 'Execution failed' >/tmp/results/result-summary.txtbranch, meaning/tmp/results/results.jsonwas never produced. Likely causes:qecore-headlessexits non-zero before behave writesresults.jsonbehave --outfile /tmp/results/results.jsonnever runs because the GNOME Shell AT-SPI wait orgnome-ponytail-daemonfails firstPod logs are GC'd (
podGC: OnWorkflowSuccess), so the fastest next step is to reproduce with podGC disabled and capture qecore-headless stderr.Why it matters for projectbluefin/actions#319
PR #319 adds a release-gate calling
testsuite/e2e.yml— a different path (QEMU on GHA ubuntu-latest, not the cluster podman path), so this bug does not directly block #319's gate. But bluefin#837 is the consumer-validation PR for #319, and its cluster smoke CI is red because of this testsuite-infra issue.Automatable: yes — the fix is in the testsuite harness, not the image.
Filed from a pi agent session reviewing projectbluefin/actions open PRs.