Normal Description of problem:
As the end user, I used the tag instead of digest for the bundle image so that I don't need to update the `catalog` if any updates, as follows,
jiazha-mac:~ jiazha$ cat catalog/index.yaml defaultChannel: "candidate-v1.0" ... --- image: quay.io/olmqe/nginx-operator-bundle:v0.0.1 name: nginx-operator.v0.0.1 package: nginx-operator properties: - type: olm.bundle.mediatype value: "registry+v1" - type: olm.package value: packageName: nginx-operator version: 0.0.1 schema: olm.bundle ---
But, the "IfNotPresent" imagePullPolicy of the bundle pod couldn't take the bundle image updates. If the previous bundle images have a problem, the users have to retag the bundle image and rebuild the catalog image or remove the preserved old image in the node manually. That's not a good experience. The pain points are as follows,
1, the "IfNotPresent" imagePullPolicy always uses the old bundle image even if there are updates on it.
- lastTransitionTime: "2024-02-05T07:25:24Z" message: 'Failed to unpack the nginx-test-2jzxs1txpexrnyim9qut1b8067n0rvgd62zl33ejj Bundle: convert registry+v1 bundle to plain+v0 bundle: readfile metadata/annotations.yaml: file does not exist' reason: UnpackFailed status: "False" type: HasValidBundle
2, there is no exact image digest display on the problemed bundle, the end user cannot get the exact image they used. As follows,
jiazha-mac:operator-framework-rukpak jiazha$ oc get bundle nginx-test-2jzxs1txpexrnyim9qut1b8067n0rvgd62zl33ejj -o yaml apiVersion: core.rukpak.io/v1alpha1 kind: Bundle metadata: creationTimestamp: "2024-02-05T07:25:21Z" finalizers: - core.rukpak.io/delete-cached-bundle generation: 1 labels: core.rukpak.io/bundle-template-hash: 2jzxs1txpexrnyim9qut1b8067n0rvgd62zl33ejjca7 core.rukpak.io/owner-kind: BundleDeployment core.rukpak.io/owner-name: nginx-test name: nginx-test-2jzxs1txpexrnyim9qut1b8067n0rvgd62zl33ejj ownerReferences: - apiVersion: core.rukpak.io/v1alpha1 blockOwnerDeletion: true controller: true kind: BundleDeployment name: nginx-test uid: dc1dd96b-d235-4aab-9446-563e58b2f0ef resourceVersion: "347445" uid: 377235c6-bc57-4d0d-927f-a865b19bff6d spec: provisionerClassName: core-rukpak-io-registry source: image: ref: quay.io/olmqe/nginx-operator-bundle:v0.0.1 type: image status: conditions: - lastTransitionTime: "2024-02-05T07:25:21Z" message: 'convert registry+v1 bundle to plain+v0 bundle: readfile metadata/annotations.yaml: file does not exist' reason: UnpackFailed status: "False" type: Unpacked observedGeneration: 1 phase: Failing
Version-Release number of selected component (if applicable):
4.15.0
How reproducible:
always
Steps to Reproduce:
1. create a problems bundle image, which lack of the metadata files.
jiazha-mac:~ jiazha$ tree nginx
nginx
├── Dockerfile
├── manifests
│ ├── nginx-operator-controller-manager-metrics-service_v1_service.yaml
│ ├── nginx-operator-manager-config_v1_configmap.yaml
│ ├── nginx-operator-metrics-reader_rbac.authorization.k8s.io_v1_clusterrole.yaml
│ ├── nginx-operator.clusterserviceversion.yaml
│ └── olm.example.com_nginxolms.yaml
jiazha-mac:nginx jiazha$ docker buildx build . --push --platform linux/amd64,linux/arm64,linux/ppc64le,linux/s390x -t quay.io/olmqe/nginx-operator-bundle:v0.0.1
...
2. build catalog image.
3. deploy it on the cluser. It failed as expected.
jiazha-mac:~ jiazha$ cat catalog.yaml
apiVersion: catalogd.operatorframework.io/v1alpha1
kind: Catalog
metadata:
name: nginx-catalog
spec:
source:
type: image
image:
ref: quay.io/olmqe/nginx-operator-index:v1
jiazha-mac:~ jiazha$ cat operator.yaml
apiVersion: operators.operatorframework.io/v1alpha1
kind: Operator
metadata:
name: nginx-test
spec:
packageName: nginx-operator
channel: candidate-v0.0
version: "0.0.1"
jiazha-mac:operator-framework-rukpak jiazha$ oc get operator.operators.operatorframework.io nginx-test -o yaml
apiVersion: operators.operatorframework.io/v1alpha1
kind: Operator
metadata:
creationTimestamp: "2024-02-05T07:25:21Z"
generation: 1
name: nginx-test
resourceVersion: "347389"
uid: 28315a4d-cf9f-45f0-84e7-1577add5d2fd
spec:
channel: candidate-v0.0
packageName: nginx-operator
upgradeConstraintPolicy: Enforce
version: 0.0.1
status:
conditions:
- lastTransitionTime: "2024-02-05T07:25:21Z"
message: resolved to "quay.io/olmqe/nginx-operator-bundle:v0.0.1"
observedGeneration: 1
reason: Success
status: "True"
type: Resolved
- lastTransitionTime: "2024-02-05T07:25:21Z"
message: bundledeployment status is unknown
observedGeneration: 1
reason: InstallationStatusUnknown
status: Unknown
type: Installed
resolvedBundleResource: quay.io/olmqe/nginx-operator-bundle:v0.0.1
jiazha-mac:operator-framework-rukpak jiazha$ oc get bd nginx-test -o yaml
apiVersion: core.rukpak.io/v1alpha1
kind: BundleDeployment
metadata:
creationTimestamp: "2024-02-05T07:25:21Z"
generation: 2
name: nginx-test
ownerReferences:
- apiVersion: operators.operatorframework.io/v1alpha1
blockOwnerDeletion: true
controller: true
kind: Operator
name: nginx-test
uid: 28315a4d-cf9f-45f0-84e7-1577add5d2fd
resourceVersion: "347446"
uid: dc1dd96b-d235-4aab-9446-563e58b2f0ef
spec:
provisionerClassName: core-rukpak-io-plain
template:
metadata: {}
spec:
provisionerClassName: core-rukpak-io-registry
source:
image:
ref: quay.io/olmqe/nginx-operator-bundle:v0.0.1
type: image
status:
conditions:
- lastTransitionTime: "2024-02-05T07:25:24Z"
message: 'Failed to unpack the nginx-test-2jzxs1txpexrnyim9qut1b8067n0rvgd62zl33ejj
Bundle: convert registry+v1 bundle to plain+v0 bundle: readfile metadata/annotations.yaml:
file does not exist'
reason: UnpackFailed
status: "False"
type: HasValidBundle
observedGeneration: 2
4, fix the bundle image and redeploy it.
jiazha-mac:nginx jiazha$ docker buildx build . --push --platform linux/amd64,linux/arm64,linux/ppc64le,linux/s390x -t quay.io/olmqe/nginx-operator-bundle:v0.0.1
...
Actual results:
Still get the same error due to the bundle pod still reading the previously issued bundle image. It's really confusing and not UX for the end user.
jiazha-mac:~ jiazha$ oc get pods nginx-test-2jzxs1txpexrnyim9qut1b8067n0rvgd62zl33ejj -o yaml
apiVersion: v1
kind: Pod
...
containers:
- command:
- /bin/unpack
- --bundle-dir
- /
image: quay.io/olmqe/nginx-operator-bundle:v0.0.1
imagePullPolicy: IfNotPresent
name: bundle
resources: {}
...
Expected results:
Use the "Always" instead of the "IfNotPresent" imagePullPolicy for the bundle unpack pod so that the end user can update their operators/bundles easily.
Additional info:
- relates to
-
OCPBUGS-30132 OCP 4.15.0 is not correctly refreshing operator catalogs (imagePullPolicy: IfNotPresent)
-
- Closed
-
