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
Cover the capability store in the kind capability test #133
tests/capabilities/run.sh starts the rendered runner pod in kind and checks the capabilities really work, but only with copies into each pod. The capability store from #132 isn't covered there: its entries are named for each image's digest, and the test builds its tool and library images locally and loads them into kind, so they have no registry digest to pin.
Running a local registry in the test (push the built images, reference them by the digest the push returns) would let it render a profile with capability_store: true and check what only a real pod shows: the hostPath store and its preparation under the pod's real securityContext, the publish and sysroot merge with busybox, the tool cache's links resolving for @actions/tool-cache, the sysroot subPath mount, and a second pod reusing the entries without copying. The unit tests run the same scripts with sh against a temporary store, which covers the logic but not the mounts or the identities.
Covered by the capability store merge: the kind test now pushes its images to a local registry so they have digests, runs a store profile, and checks the publish, a second pod reusing the entries, the read-only mounts, writable tool trees and the pruner.
Activity
Mearman commentedon Oct 22, 2026
Covered by the capability store merge: the kind test now pushes its images to a local registry so they have digests, runs a store profile, and checks the publish, a second pod reusing the entries, the read-only mounts, writable tool trees and the pruner.