From 4984c628b1aaa04b5415e156ff69d1e62cbb4fff Mon Sep 17 00:00:00 2001 From: Danathar Date: Tue, 10 Nov 2026 16:31:23 -0500 Subject: [PATCH] ci(startup): give the install container the host's udev database bootc 1.17 reads the new root partition's type with lsblk, which gets it from the udev database. The startup lane's install container gets /dev but not /run/udev, so on the current Dakota testing image bootc stops with "root partition 3 has type " before writing anything. The startup checks have failed this way on every run since 2026-11-10, and the v1 gate's boots fail the same way in e2e.yml (#1123). #1125's retries don't help: all three attempts fail identically. Mount /run/udev into the install container. This PR's own startup checks are the test. If they pass, e2e.yml needs the same one-line mount in a separate change, since it is a human-merge release workflow. Refs #1123 Assisted-by: Claude Opus 5.5 via Claude Code Co-Authored-By: Claude Opus 5.5 Signed-off-by: Danathar --- .github/workflows/startup-qualification.yml | 6 +++++- .../ci-ops/e2e-workflow/references/troubleshooting.md | 2 ++ 2 files changed, 7 insertions(+), 1 deletion(-) diff --git a/.github/workflows/startup-qualification.yml b/.github/workflows/startup-qualification.yml index fa5b38ef4..3948289cb 100644 --- a/.github/workflows/startup-qualification.yml +++ b/.github/workflows/startup-qualification.yml @@ -111,9 +111,13 @@ jobs: BOOTLOADER_FLAG="--bootloader systemd" fi set +e + # /run/udev: bootc >=1.17 reads the new root partition's type with + # lsblk, which takes it from the udev database. Without the host's + # /run/udev the container sees no type and bootc aborts with + # "root partition 3 has type " (testsuite#1123). sudo podman run --rm --privileged --pid=host \ --security-opt label=type:unconfined_t \ - -v /var/lib/containers:/var/lib/containers -v /dev:/dev -v "$(pwd):/data" \ + -v /var/lib/containers:/var/lib/containers -v /dev:/dev -v /run/udev:/run/udev -v "$(pwd):/data" \ "${IMAGE}" bootc install to-disk --via-loopback /data/disk.raw \ --filesystem ext4 --wipe ${BOOTLOADER_FLAG} 2>&1 | tee bootc-install.log echo "bootc install exited ${PIPESTATUS[0]} (bootloader step may fail; deployment checked below)" diff --git a/docs/skills/ci-ops/e2e-workflow/references/troubleshooting.md b/docs/skills/ci-ops/e2e-workflow/references/troubleshooting.md index 9f91e4c4c..1f6ab5a18 100644 --- a/docs/skills/ci-ops/e2e-workflow/references/troubleshooting.md +++ b/docs/skills/ci-ops/e2e-workflow/references/troubleshooting.md @@ -82,6 +82,8 @@ still belong only in a fresh disposable guest without publishing credentials. **Fix:** the "Install OCI image and configure disk" step in `e2e.yml` retries `bootc install to-disk` (3 attempts) whenever no ostree deployment was written, and fails with a `bootc install wrote no deployment` annotation naming the image and bootc's `error:` line if every attempt fails. Do not retry when the deployment *was* written: `bootupd is required for ostree-based installs` (bootc 1.17+ with `--bootloader systemd`) is the expected late bootloader failure, and direct kernel boot continues past it. +**Update (2026-11-10, #1123):** on `dakota:testing@sha256:dc3f6aee…` (built 2026-11-09) the error is not a race: every `v1` gate run since 2026-11-09 09:31Z failed all 3 attempts identically (for example run 42807521428), and the `startup` checks fail the same way (run 42842531479). The install container is given `/dev` but not the host's `/run/udev`, so `lsblk` inside it has no udev database to read the partition type from (compare bootc-dev/bootc#2239 and #2169). `startup-qualification.yml` now passes `-v /run/udev:/run/udev` to the install container. If that holds, `e2e.yml`'s install step needs the same mount; that file is a human-merge release workflow, so it is a separate change. + ### SSH never became ready Check the serial log artifact. Common causes: