Repository navigation
Description
Summary: The project's on-demand test run ("Manual Test Run") can no longer install the Dakota operating-system image onto the virtual disk it tests, so it fails before any test starts. Every manual run since November 1 has failed this way, which means nobody can currently use it to check a change against Dakota — including the check that test PR #1116 is waiting for.
What happens
manual.yml → e2e.yml job "GNOME 50 — common", step Install OCI image and configure disk (.github/workflows/e2e.yml:396-460 @ 2ff4def) runs bootc install to-disk --via-loopback /data/disk.raw --filesystem ext4 --wipe [--bootloader systemd] inside the image, then assumes an ostree layout: it mounts ${LOOP}p3 as root (e2e.yml:441) and looks for /mnt/root/ostree/deploy/default/deploy/* (e2e.yml:448). With current Dakota images the install fails before that layout exists:
| Run | Date | Image | Error from bootc install to-disk |
|---|---|---|---|
| 42611596358 | 2026-11-09 | ghcr.io/projectbluefin/dakota@sha256:dc3f6aee7a9fbd8373e8cf9d0dfbfd80e6beaf4b19922a7c0daf4725d3086e43 (:testing that day) |
error: Installing to disk: Creating rootfs: root partition 3 has type <none>; expected 4f68bce3-e8cd-4db1-96e7-fbcaf984b709 |
| 42221191543 | 2026-11-06 | dakota@sha256:e9918fd7… |
error: Installing to disk: bootupd is required for ostree-based installs |
| 42129446403 | 2026-11-05 | dakota@sha256:e9918fd7… |
same as above |
All manual.yml runs since 2026-11-01 (42124147912, 42124153048, 42126521360, 41259347169, and the three above) concluded failure. The 11-09 run was dispatched with suites=common, test_ref=ae22434cfe05ce9c119f2e508634d599201ab327; it never reached behave (no desktop screenshot, "runner container may not have loaded").
e2e.yml:22 makes ghcr.io/projectbluefin/dakota:testing the default image, so this likely affects any caller that tests Dakota through this workflow, not only manual runs. (Not checked: whether Dakota's own promotion gate currently runs this path.)
Likely cause (verify first)
The step was written for ostree-based images whose bootc install fails late at the bootloader (see the comment at e2e.yml:392-395) and then falls back to direct kernel boot. Current Dakota images appear to install differently (composefs/verity — bootc status on Dakota shows a Verity: line), and bootc install to-disk now either refuses (no bootupd) or rejects the partition layout it is given. Confirm by reading the full bootc-install.log from run 42611596358 and the Dakota image's bootc install config (/usr/lib/bootc/install/*.toml in the image) before changing anything.
Acceptance criteria
manual.ymlwithimage=ghcr.io/projectbluefin/dakota:testing(pinned by digest) andsuites=commongets past Install OCI image and configure disk, boots, and runs the common features (pass/fail of individual scenarios is out of scope here).- Ostree-based images that work today still work (e.g. a Bluefin image used by another caller of
e2e.yml). - The step fails with a clear message naming the image and the bootc error if the install genuinely cannot proceed — never a later, unrelated "no screenshot" failure.
Test plan
- Reproduce: dispatch
manual.ymlonmainwith the Dakota digest above andsuites=common; confirm the same error. - Apply the fix; dispatch again with the same inputs; the job reaches behave and uploads a desktop screenshot.
- Dispatch once with an ostree-based image that passed before 11-01 to confirm no regression.
Out of scope
- Whether the two immutable-OS scenarios in test: prepare system checks for a fresh image #1116 pass (that PR's own gate; re-run it after this is fixed).
- Changing Dakota's image or disk layout.
- The
@bluefinscenarios skipping on non-Bluefin images (expected).
Activity
fix(e2e): retry bootc install when it fails before deploying
hivecommons-hive commentedon Nov 9, 2026
PR #1125 covers this issue: #1125
The PR body says this is partial/non-closing work (
Refs/Part ofor leave-open wording), so Hive will not add a closing keyword. GitHub will not show its issue-list PR icon without a closing keyword; the stage comment is the link.fix(e2e): retry bootc install when it fails before deploying (#1125)