You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
[shared-ci] E2E — composed common suite / e2e / GNOME — common-b failing across projectbluefin/common #1679
Summary: The E2E — composed common suite / e2e / GNOME — common-b check is failing on 3 open pull requests in projectbluefin/common at the same time. When one check fails on this many pull requests, the cause is usually shared (the base branch, a workflow, a pinned action or the runner image), not any single change. The hive opened this issue so the failure is looked at once, here, instead of once per pull request.
Evidence
Check: E2E — composed common suite / e2e / GNOME — common-b
Default branch: the newest run of this check on the base branch is not red. Either these pull requests are behind their base, or something outside the branch changed since that run (a moved action tag, a new runner image, an upstream outage).
Next step
Repair the shared cause once, usually with one pull request against the base branch or the shared workflow, then close this issue. Pull requests that are only behind their base may also need their branch updated.
The hive keeps this description in step with the failing pull requests while the failure lasts (edits do not notify) and does not comment on the pull requests listed here.
Cause, from the failed runs (#1674 run 43579243413, #1665 run 43533700400):
Every test scenario passes (common-b: 36 passed, 0 failed, 20 skipped; common-a: 53 passed, 0 failed). What fails is the SELinux denial gate that testsuite runs after the scenarios. It reports 20 copies of unexpected SELinux denial: { mac_admin } comm=chcon tuned_t -> tuned_t:capability2.
utah's own e2e has the same denial class. utah's Post-Testing E2E runs have failed since at least 11-13 4:29 PM EST with 88 copies of { mac_admin } comm=chcon unconfined_service_t per run. Their common-a/common-b jobs pass every scenario too (one smoke-a scenario failed on utah-nvidia-gaming at 4:29 PM, which is separate). utah#1198 traces that one: bootc status, run from inside a script, probes for mac_admin with chcon and is denied outside bootc's own domain. utah#1203 (Kyle) is the fix there.
This repo's e2e base is stale.pr-e2e.yml composes on BASE_IMAGE: ghcr.io/projectbluefin/bluefin:testing. That image was last built 2026-09-17 from projectbluefin/bluefin, which is archived. So these jobs test common's layer on a two-month-old base, and the tuned_t domain in the denial comes from that old base, not from this repo (common ships no tuned config and calls chcon only in just overlay). I haven't confirmed which tuned script runs the chcon; the raw AVC lines are in the jobs' results/selinux artifacts, not the log.
common has the same pattern utah#1198 fixes.system_files/shared/usr/libexec/ublue-booted-image:30 runs bootc status --json from a script. Whatever unit runs that script will hit the same probe denial once the base stops masking it.
Options for whoever owns common CI (not decided here):
Point BASE_IMAGE at a current base (for example ghcr.io/projectbluefin/utah:testing). That only goes green once utah#1203 lands, since utah has the same denial today.
Fix ublue-booted-image the same way utah#1198 asks (read the booted ref without bootc, or run bare bootc as the unit's ExecStart).
Allowlisting in testsuite is possible, but its own guidance argues against it.
Summary: The shared desktop tests are failing because the image they start from records security warnings outside this repository’s changes. A replacement image has a separate fix awaiting review, so switching images now would not resolve the failures.
The issue investigation identifies tuned_t denials in the archived Bluefin base and the same denial class in Utah. The proposed current base, ghcr.io/projectbluefin/utah:testing, still depends on projectbluefin/utah#1203 and a subsequent image build; that PR is open. Once available, the composed suite needs to be rerun against the updated image to establish whether any denials remain.
Closing as stale and superseded by #1688. The common-b shard check no longer exists: #1687 (merged 11-14 2:49 PM EST) moved PR E2E to Dakota, which runs the common suite as one job (GNOME — common). The only PR still showing a red common-b result is #1665, and that's left over from a run before #1687. The current failure on every PR is tracked in #1688.
Activity
Danathar commentedon Nov 14, 2026
Cause, from the failed runs (#1674 run 43579243413, #1665 run 43533700400):
unexpected SELinux denial: { mac_admin } comm=chcon tuned_t -> tuned_t:capability2.selinux_allowlist.yamlis empty on purpose: the gate fails on any enforced denial, and the file says to fix denials in the image that owns them.{ mac_admin } comm=chcon unconfined_service_tper run. Their common-a/common-b jobs pass every scenario too (one smoke-a scenario failed on utah-nvidia-gaming at 4:29 PM, which is separate). utah#1198 traces that one:bootc status, run from inside a script, probes formac_adminwithchconand is denied outside bootc's own domain. utah#1203 (Kyle) is the fix there.pr-e2e.ymlcomposes onBASE_IMAGE: ghcr.io/projectbluefin/bluefin:testing. That image was last built 2026-09-17 from projectbluefin/bluefin, which is archived. So these jobs test common's layer on a two-month-old base, and thetuned_tdomain in the denial comes from that old base, not from this repo (common ships no tuned config and callschcononly injust overlay). I haven't confirmed which tuned script runs the chcon; the raw AVC lines are in the jobs'results/selinuxartifacts, not the log.system_files/shared/usr/libexec/ublue-booted-image:30runsbootc status --jsonfrom a script. Whatever unit runs that script will hit the same probe denial once the base stops masking it.Options for whoever owns common CI (not decided here):
BASE_IMAGEat a current base (for exampleghcr.io/projectbluefin/utah:testing). That only goes green once utah#1203 lands, since utah has the same denial today.ublue-booted-imagethe same way utah#1198 asks (read the booted ref without bootc, or run bare bootc as the unit's ExecStart).Same cause as #1680 (common-a).
/triage accepted
Karibhou commentedon Nov 14, 2026
Summary: The shared desktop tests are failing because the image they start from records security warnings outside this repository’s changes. A replacement image has a separate fix awaiting review, so switching images now would not resolve the failures.
The issue investigation identifies
tuned_tdenials in the archived Bluefin base and the same denial class in Utah. The proposed current base,ghcr.io/projectbluefin/utah:testing, still depends on projectbluefin/utah#1203 and a subsequent image build; that PR is open. Once available, the composed suite needs to be rerun against the updated image to establish whether any denials remain.— hive: agent=scanner backend=omp model=openai-codex/gpt-6.1-terra
Danathar commentedon Nov 15, 2026
Closing as stale and superseded by #1688. The
common-bshard check no longer exists: #1687 (merged 11-14 2:49 PM EST) moved PR E2E to Dakota, which runs the common suite as one job (GNOME — common). The only PR still showing a redcommon-bresult is #1665, and that's left over from a run before #1687. The current failure on every PR is tracked in #1688.