Skip to content

fix(release): compare host-less image refs on digest alone - #3905

Merged
myasnikovdaniil merged 2 commits into
release-1.6from
fix/verify-promoted-hostless-refs
Aug 19, 2026
Merged

fix(release): compare host-less image refs on digest alone#3905
myasnikovdaniil merged 2 commits into
release-1.6from
fix/verify-promoted-hostless-refs

Conversation

@myasnikovdaniil

Copy link
Copy Markdown
Contributor

Fixes the false promotion changed the container repository/digest set that blocks #3904.

normalized_refs() in hack/verify-promoted-packages.sh gives a host-less ref a repository equal to its own tag. hack/lib/image-refs.sh emits two entries per shape-3 map, the correctly joined ref plus shape 1's recursive scrape of the bare tag: <tag>@sha256:... scalar, and on that bare scalar repo="${image%:*}" is a no-op because there is no / and no : in it. Promotion rewrites exactly those tags, so v1.6.2-rc.1 to v1.6.2 shows up as six changed repositories whose digests are identical on both sides. promote-retag.sh never hits this because it drops host-less refs through its ownership filter, the verifier had no equivalent.

Host-less refs now compare on digest alone, which is the whole of the container identity such a ref carries. Checked against the real published artifacts: before the fix 61 entries each side differing by those six pairs, after it 60 each side with the sets equal and 48 container digests identical.

Two things this deliberately does not do, both written as comments at the fix. Not dropping host-less refs outright, because packages/system/kuberture/values.yaml carries an image: map with a tag: and no repository:, so shape 1 is the only rule that ever sees its digest and a drop would silently stop proving it unchanged. And not reusing promote-retag.sh's ${REGISTRY}/ ownership filter, because the verify job's REGISTRY names the private build registry while both artifacts under comparison live on the public one, so that filter drops all 48 refs, and the emptiness guard above tests the raw collection rather than the filtered set, so the check would pass by comparing two empty sets.

Second commit adds the fixture that was missing. The existing suite writes a host-bearing string and exercises a passing rc to stable rewrite, so it read as coverage while never building a shape-3 split map, which the library itself calls the dominant shape. Reverting the fix reddens the new case on the digest-set assertion, and replacing the fix with an outright skip of host-less refs makes the verifier exit 0 on a genuinely changed digest.

Once this is on release-1.6 the check on #3904 re-runs against this copy, so no rc.2 re-cut is needed. main carries the same script and wants the same fix.

hack/verify-promoted-packages.sh reported "promotion changed the
container repository/digest set" for a v1.6.2 promotion that moved no
container bytes at all: all 48 real repo@digest refs and all 48 digests
were byte-identical between the rc artifact and the promotion candidate.

The six entries it flagged as changed repositories were not
repositories. They were bare tag strings -- v1.6.2-rc.1 on the rc side,
v1.6.2 on the candidate side -- pairing one-to-one by identical digest.
hack/lib/image-refs.sh emits two entries per shape-3 split map: the
correctly joined repository@digest, plus shape 1's recursive scrape of
the raw .tag scalar, which carries no repository. The library documents
that degeneracy and leaves host-less refs to its callers;
promote-retag.sh drops them, the verifier had only a
*/cozystack-packages exclusion. So `${image%:*}` -- a no-op on a string
holding no colon -- handed the tag over as the repository, and promotion
rewrites exactly those tags.

Set an empty repository for a ref with no `/` ahead of its digest, so it
compares on the digest, which is the whole container identity such a ref
carries and the only part promotion must not move. Dropping these refs
instead would be wrong rather than blunter: packages/system/kuberture
carries an `image:` map with a `tag:` and no `repository:`, so shape 1
is the only rule that sees its digest, and skipping it would silently
stop proving that digest unchanged.

Reproduced over the release-1.6 packages tree with the shape-3 tags
rewritten rc-to-stable: before, the six pairs above diverge; after, both
sides normalize to 60 identical entries, kuberture's digest included and
kube-ovn's real repository still visible through its shape-4 join.

Signed-off-by: Myasnikov Daniil <myasnikovdaniil2001@gmail.com>
No fixture in verify-promoted-packages_test.bats built a shape-3 split
map -- `repository:` on one key, `tag: <tag>@sha256:<digest>` on another
-- which hack/lib/image-refs.sh itself calls the dominant shape in the
tree. The existing fixture's single-string ref carries a registry host,
so its rc-to-stable rewrite normalizes through the other branch and
passes whether or not host-less refs are handled, which is why the suite
looked covered while a clean promotion could still be rejected.

Add two cases. The first promotes a split map whose tag moves from rc to
stable with the repository and digest held constant; reverting the
normalization fix reddens it on
"promotion changed the container repository/digest set", diffing
v9.9.9-rc.3@sha256:eee... against v9.9.9@sha256:eee... -- the same
tag-as-repository pair, with identical digests, that CI reported.

The second pins the property a blunter fix would break. A digest-only
`image:` map with no `repository:` (the kuberture shape) is seen by no
rule but shape 1, so it must still contribute its digest: the case moves
that digest and nothing else, and asserts the rejection comes from the
digest set rather than a neighbouring leg. Replacing the fix with an
outright `continue` over host-less refs makes the verifier exit 0 on that
changed digest and reddens the case at its exit-status assertion.

Negative assertions are counted rather than `!`-negated: under
hack/cozytest.sh a `! grep -q` cannot fail, so it would be a comment
shaped like an assertion.

Signed-off-by: Myasnikov Daniil <myasnikovdaniil2001@gmail.com>
@github-actions github-actions Bot added size/L This PR changes 100-499 lines, ignoring generated files area/release Issues or PRs related to release tooling (changelog, backport, release pipeline) kind/bug Categorizes issue or PR as related to a bug labels Aug 19, 2026
@coderabbitai

coderabbitai Bot commented Aug 19, 2026

Copy link
Copy Markdown
Contributor

Important

Review skipped

Auto reviews are disabled on base/target branches other than the default branch.

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro Plus

Run ID: d47ab5e9-7f96-44f4-88a5-1016fc11380f

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

  • 🔍 Trigger review

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

Comment @coderabbitai help to get the list of available commands.

@myasnikovdaniil
myasnikovdaniil merged commit fc592ef into release-1.6 Aug 19, 2026
12 checks passed
@myasnikovdaniil
myasnikovdaniil deleted the fix/verify-promoted-hostless-refs branch August 19, 2026 06:44
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

area/release Issues or PRs related to release tooling (changelog, backport, release pipeline) kind/bug Categorizes issue or PR as related to a bug size/L This PR changes 100-499 lines, ignoring generated files

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant