Repository navigation
ci: Cover btrfs on Fedora 44 - #2644
Conversation
btrfs is Fedora's default root filesystem and supports fs-verity, so composefs installs there get a strict command line, but no CI job has ever installed onto it: the integration matrix only had ext4 and xfs. Add it as a filesystem value and exclude it everywhere except Fedora composefs, keeping one sealed UKI cell (the same plans as the other sealed cells) and one GRUB BLS cell (the full integration set, like the other BLS cells). Using excludes rather than include: entries matters here: an included fedora-44 row would also be added on ci/tier-1 runs, which only build centos-10 packages. Closes: bootc-dev#2633 Generated-by: AI Signed-off-by: Joseph Marrero Corchado <jmarrero@redhat.com>
It was excluded in bb674f1 until 1.16.8, because older releases deadlocked in the upgrade path (fixed by 71ad2d3, in 1.16.8). The published Fedora 44 image has shipped newer bootc for months, so the exclusion only hides coverage now. Run it on btrfs, Fedora's default, which also covers an upgrade from the published bootc on that filesystem. CentOS Stream keeps composefs on ext4. The filesystem is now a matrix axis exported as BOOTC_filesystem, which test-upgrade already passes to the composefs install; ostree keeps the image's default (xfs) as before. Generated-by: AI Signed-off-by: Joseph Marrero Corchado <jmarrero@redhat.com>
install_args only passed --filesystem with the composefs backend, so an ostree run with BOOTC_filesystem=btrfs silently installed on the image's default filesystem instead. Pass it whenever one is set; without one, composefs still defaults to ext4 and ostree to the image's default. The existing ostree CI cells set xfs, which is already their default. The Fedora CoreOS xfs default now only applies when no filesystem was asked for, so an ostree run that sets one doesn't pass both to bcvk. Prep for an ostree btrfs CI cell. Generated-by: AI Signed-off-by: Joseph Marrero Corchado <jmarrero@redhat.com>
Fedora's own ostree installs (the Atomic desktops) sit on btrfs, while the ostree cells only ever ran on xfs. Now that the xtask passes the filesystem to bcvk for ostree, this cell really installs on btrfs. Generated-by: AI Signed-off-by: Joseph Marrero Corchado <jmarrero@redhat.com>
cgwalters
left a comment
There was a problem hiding this comment.
LGTM but I think we've been past the point for quite a while where running the entire matrix on every PR is unreasonable; we keep adding straws to the camel's back here, I can't say this will be the one that breaks it but...
I think we can move about 70% of this matrix to opt-in and postsubmit + automated agentic-assisted reverts.
|
Let's map the minimum tests then we want on each PR? @jmarrero-bot take Colin's last comment and map the entire test matrix here in a new tracker issue in our forge. Then let's evaluate what would be the minimal tests we should include in each PR. Maybe even propose a way to run a set of tests depending of what area of the code is changed. Ideally with would not include agents making the decision and can be deterministic. |
|
On it: jmarrero-forge/tracker#85 (full matrix map, a minimal per-PR set, and a deterministic path-based selection proposal). |
No CI job installed onto btrfs, Fedora's default root filesystem, which supports fs-verity (so composefs gets a strict command line there). This adds it on Fedora 44:
test-integration: composefs btrfs systemd UKI sealed (readonly image-upgrade-reboot, as the other sealed cells), composefs btrfs grub BLS and ostree btrfs grub BLS (both the fullintegrationset). It uses the existingfilesystemaxis plus excludes rather thaninclude:, soci/tier-1runs (centos-10 only) get no fedora-44 row.test-upgrade (fedora-44, composefs)is back, now on btrfs. It was excluded until 1.16.8 for the deadlock fixed in 71ad2d3; the publishedfedora-bootc:44ships 1.17.1.install_argsonly passed--filesystemto bcvk for composefs, so the ostree cell would have silently installed on xfs.Per-PR cost, from the matching cells of run 42243165273: sealed UKI ~35-40 min, composefs BLS ~2.5 h, ostree ~3 h, upgrade ~15 min, about 6.5 runner-hours on top of ~42 for
test-integrationtoday. Wall-clock time doesn't change (the longest existing cell takes 3.5 h). If that is too much, the composefs BLS cell could run a reduced set (readonly image-upgrade-reboot composefs-verity composefs-gc rollback, ~45 min), which needs a smalltest-composefsknob since its plan filter always includesintegration; the ostree cell is its own commit and can be dropped.Matrix expansion (checked with a script that reproduces the 19 cells of that run from main's workflow): the full OS list goes from 19 to 22 integration and from 3 to 4 upgrade jobs;
ci/tier-1stays at 7 and 2, all centos-10. actionlint reports nothing new.Testing, on a 16-core devspace (jmarrero-devspace-42377888679) at a266a79 (7436abd since then only rewords a Justfile comment), running each cell with CI's env and steps (
BOOTC_filesystem=btrfsetc., prebuilt packages,build-fetch,build,test-container-integration, thentest-composefs/test-tmt integration/test-upgrade):test-upgradefedora-44 composefs btrfs: passed (installed by bootc 1.17.1, upgraded, readonly)integration: all 30 plans passed (~1h45 including the build)integration: all 34 plans passed (~2h15 including the build)readonly: passed; bcvk now gets an explicit--filesystem=xfsthere, which is the image's default anywaycargo test -p xtask: 32 passedEach bcvk install ran with
--filesystem=btrfsand formatted the root withmkfs.btrfs, the ostree one included.Fixes #2633
The
Signed-off-by: Joseph Marrero Corchado <jmarrero@redhat.com>on these commits was added on jmarrero's approval of the review draft: jmarrero-forge#15 (comment)Generated-by: https://github.com/jmarrero-bot#llms