-
Bug
-
Resolution: Won't Do
-
Normal
-
None
-
4.15
-
Moderate
-
No
-
False
-
This PR disabled any replacement for the imagestreams having LookupPolicy.local=true when workloads are deployed and the ownerReferences.Controller=true.
This is generally a good approach when Kubernetes-native controllers control workloads. However, it also disables this behavior for a wide range of workloads. For example, controllers run by operators that create workloads as pods (Tekton, Argo Workflows, and derived) cannot rely anymore on the image-lookup feature.
In that case, we cannot control the parent controller and should continue replacing the image based on the LookupPolicy. We could delegate the external controllers to decide the logic around reconciling the (mutable) image field of podTemplate specs as partially discussed in https://github.com/openshift/openshift-apiserver/issues/339 .
Version-Release number of selected component (if applicable):
4.15 (.... 4.9ish)
How reproducible:
Always
Steps to Reproduce:
1. Create an image stream with LookupPolicy.local=True:
apiVersion: image.openshift.io/v1
kind: ImageStream
metadata:
name: hello-openshift
spec:
lookupPolicy:
local: true
tags:
- name: 1.2.0
from:
kind: DockerImage
name: quay.io/openshifttest/hello-openshift:1.2.0
importPolicy:
importMode: PreserveOriginal
referencePolicy:
type: Local
2. Create a pod with ownerReferences.controller=true and a container's image field set to the image stream tag name without the internal registry URL
apiVersion: v1
kind: Pod
metadata:
name: example
labels:
app: httpd
namespace: okd-fcos
ownerReferences:
- apiVersion: build.openshift.io/v1
kind: Build
name: xxx
uid: 59efca52-10a8-4425-a4bc-3c5915ca189b
controller: true
spec:
securityContext:
runAsNonRoot: true
seccompProfile:
type: RuntimeDefault
containers:
- name: hello
image: hello-openshift:1.2.0
ports:
- containerPort: 8080
securityContext:
allowPrivilegeEscalation: false
capabilities:
drop:
- ALL
Actual results:
ImagePullBackOff
Expected results:
The image field of the container get replaced with the URL of the image including the internal-registry FQDN.
Additional info:
1. We should document this at least in https://docs.openshift.com/container-platform/4.13/openshift_images/using-imagestreams-with-kube-resources.html 2. A possible solution is to allow that replacement when the workload is a pod, and its controller is not one of the Kubernetes native ones (replicaset, statefuleset etc...). - Previous bug: https://bugzilla.redhat.com/show_bug.cgi?id=1925180 - Related issue: https://github.com/openshift/openshift-apiserver/issues/339
