Unable to add OCI based GCP's Artifact Registry to Kubeapps

[jeunii]

Description:

I am trying to add an OCI Registry to the App Repository. The OCI registry is GCP's Artifact Registry

OCI registry type is supported.

In GCP I have setup a test repo and can successfully deploy a helm release

helm install hello-chart oci://europe-west4-docker.pkg.dev/sap-cx-pipeline-prod/quickstart-helm-repo/hello-chart --version 0.1.0

NAME: hello-chart
LAST DEPLOYED: Tue Feb 15 14:31:46 2022
NAMESPACE: default
STATUS: deployed
...

I have followed the guide and my config looks like this

I can create the repo in kubeapps successfully and its displayed on the UI

But when I click on it I get the error

Unable to fetch package: Unable to get the available package versions for the package "gcp-artifactory-repo%2Fquickstart-helm-repo%2Fhello-chart" using the plugin "helm.packages": rpc error: code = InvalidArgument desc = Incorrect request.AvailablePackageRef.Identifier, currently just 'foo/bar' patters are supported: gcp-artifactory-repo%2Fquickstart-helm-repo%2Fhello-chart

Steps to reproduce the issue:

1. Create an helm repo of type docker in GCP's artifact registry
2. Create a Service Account in GCP, give it read rights to artifact registry and generate a token
3. Add the repo in App Repository in Kubeapps

Describe the results you received:

I cannot view the helm chart in Catalogue section in Kubeapps. Refreshing the repo creates a cronjob which runs succesfully

kl apprepo-apps-sync-sap-cx-pipeline-prod-vwrzr-5bdnx -f

I0215 20:45:25.984389       1 sync.go:87] Last checksum: be382368038c7bccbc194240d29e4813bbde973c4db7469a3eee2668fc17b538
time="2022-02-15T20:45:26Z" level=info msg="icon not found" name="quickstart-helm-repo%2Fhello-chart"
I0215 20:45:26.416882       1 sync.go:115] Repository synced, shallow=false
I0215 20:45:26.419472       1 sync.go:124] Stored repository update in cache, repo.URL= https://europe-west4-docker.pkg.dev/sap-cx-pipeline-prod/
I0215 20:45:26.419494       1 sync.go:125] Successfully added the package repository sap-cx-pipeline-prod to database

The app repository controller also does not show any error

time="2022-02-15T20:45:23Z" level=info msg="Successfully synced 'apps/sap-cx-pipeline-prod'"
time="2022-02-15T20:45:23Z" level=info msg="Event(v1.ObjectReference{Kind:\"AppRepository\", Namespace:\"apps\", Name:\"sap-cx-pipeline-prod\", UID:\"da1febb8-24d2-4d07-8314-ed583d314623\", APIVersion:\"kubeapps.com/v1alpha1\", ResourceVersion:\"142628072\", FieldPath:\"\"}): type: 'Normal' reason: 'Synced' AppRepository synced successfully"

Describe the results you expected:

Expect to see the helm chart and deploy it.

Additional information you deem important (e.g. issue happens only occasionally):

I also uploaded version 0.2.0 to GCP's artifact repository but it still shows version 0.1.0 in kubeapps UI

In GCP:

In Kubeapps after refreshing the repo

I also see such errors

In GCP

Version of Helm, Kubeapps and Kubernetes:

• Kubeapps version

v2.4.2

[absoludity]

Originally discussed on slack. I've not bisected to see if it's a recent change since switching to the kubeapps-apis server (assume so) or something we've had in the OCI support from the start (doubt it?), but it seems we've enforced a helm package identifier of foo/bar which doesn't work when an OCI repository is itself partitioned with a slash.

[antgamdia]

I had a quick look at the issue (I have repro it, but using Harbor) and the problem seems to be twofold:

1. As Michael pointed out at Return the repo name in a carvel a pkg summary #4716, we are forcing foo/bar patterns
2. The (c *OCIRepoClient) GetChart in this file below is not prepared to unescape the chart name coming from the database:

assets=# SELECT ((info ->> 'ID')) FROM charts WHERE chart_id NOT ILIKE '%bitnami/%';
 oci/kubeapps
 oci/kube%2Fapps

but the URL it requests is:

I0518 14:26:22.178671       1 chart.go:417] XXX repoURL: https://harbor-repo.vmware.com/agamez
I0518 14:26:22.178732       1 chart.go:422] XXX url: https://harbor-repo.vmware.com/agamez
I0518 14:26:22.178782       1 chart.go:425] XXX ref: harbor-repo.vmware.com/agamez/kube%2Fapps:8.1.2-dev0

[antgamdia]

As per the comment at #4739 (comment):

Hello. Thank you for this. In which release should I be able to test this ?

Hi @jeunii, I'm reopening the issue since I didn't mean to close it before checking it with you first.

Since the modified component is kubeapps-apis, you only have to replace the "official" image with the one we build once a new commit gets merged into the main branch. For instance, you can execute:

kubectl scale deploy/kubeapps-internal-kubeappsapis -n kubeapps --replicas=0
kubectl set image deployment/kubeapps-internal-kubeappsapis -n kubeapps kubeappsapis=kubeapps/kubeapps-apis
kubectl scale deploy/kubeapps-internal-kubeappsapis -n kubeapps --replicas=1

If it doesn't work, you can always use our local chart; that is, cloning this project and installing the chart defined under /chart/kubeapps, which includes the latest build.

Let us know if this PR fixed your issue, otherwise, we will still continue to investigate it!

[ppbaena]

Hi @jeunii 👋 , could you take a look on this? Do you have any update related to this issue?