Open nebhale opened 1 month ago
Thank you for the detailed issue @nebhale
This is primarily because kapp-controller
caches images based of their sha reference. And this does not change on relocation.
THis leads to the older images being consumed. We need to have some time/location based invalidation logic in vendir
so that kapp-controller
can benefit from this behaviour.
What steps did you take:
PackageRepository
on a cluster that points to location APackageInstall
for the cert-manager package above and validate that the Pod image locations are from quay.ioimpkg copy
PackageRepository
to point to location B and validate that the Pod image locations are now from location Bimgpkg copy
PackageRepository
to point to location C and validate that the Pod image locations are still from location BWhat happened: Upon the first update of the
PackageRepository
location, the Pod image locations were updated to match the new location. Upon the second update of thePackageRepository
location, the Pod image locations were not updated to match the new location.What did you expect: I expect that every update of a
PackageRepository
location updates the Pod image location to match the newPackageRepository
location.Anything else you would like to add: No
Environment:
kubectl get deployment -n kapp-controller kapp-controller -o yaml
and the annotation iskbld.k14s.io/images
):kubectl version
):Vote on this request
This is an invitation to the community to vote on issues, to help us prioritize our backlog. Use the "smiley face" up to the right of this comment to vote.
👍 "I would like to see this addressed as soon as possible" 👎 "There are other more important things to focus on right now"
We are also happy to receive and review Pull Requests if you want to help working on this issue.