Cloned-restore fails on Openshift clusters with ImagePull error
Description: Cloned-restore Job failed to be executed
In Openshift 4.x, pods can pull images with the use of ImageStreams. ImageStreamTag represents an image that is retrieved by tag name from an ImageStream. If these are hard-coded in a service account, pods in the target namespace will fail to come up with an ImagePullError.
e.g.
Namespace = testify-ps
SA = default
Target namespace = testify-ps-restore
Error: Failed to pull image
Cloned-restore jobs time-out and eventually fail due to an ImagePullError. Upon further investigation, the root cause is because the default SA (in the target namespace) does not have ImageStreams for the pod to pull an image from a private registry.
e.g. Failed to pull image "image-registry.openshift-image-registry.svc:5000/testify-ps/runner@sha256:9317d61bb7943fb2b3f0d62c37aab233a07a3e0ccc67b104a7e3e69cb778dea0": rpc error: code = Unknown desc = Error reading manifest sha256:9317d61bb7943fb2b3f0d62c37aab233a07a3e0ccc67b104a7e3e69cb778dea0 in image-registry.openshift-image-registry.svc:5000/testify-ps/runner: unauthorized: authentication required
Resolution
There are two ways to overcome such failures:
- Identify missing ImageStreamTag(s), then go through steps 2, 3 or 4:
e.g. Source Namespace
$ oc get istag -n testify-ps
NAME IMAGE REFERENCE UPDATED
mgmt:develop image-registry.openshift-image-registry.svc:5000/testify-ps/mgmt@sha256:092577fb9c72f7631b08c1f96658b34a5f928953f59ff8e1d38e91a5a64ef918 6 days ago
mongo:4.2 mongo@sha256:8c48baa1571469d7f5ae6d603b92b8027ada5eb39826c009cb33a13b46864908 6 days ago
runner:develop image-registry.openshift-image-registry.svc:5000/testify-ps/runner@
Target Namespace
$ oc get istage -n testify-ps-restore
NAME IMAGE REFERENCE UPDATED
mongo:4.2 mongo@sha256:8c48baa1571469d7f5ae6d603b92b8027ada5eb39826c009cb33a13b46864908 2 days ago - Manually add them to the default SA in the target namespace and restart pods.
e.g
$ oc project testify-ps-restore
$ oc edit serviceaccount default
<Add the missing mongo:4.2 secret>
$ oc delete pod <pod name> - Run the script below to pull ImageStreams from one namespace to another.
$ oc policy add-role-to-user \
More details can be found here
system:image-puller system:serviceaccuont:testify-ps:default \
-namespace=testify-ps-restore - Alternatively, the K10 transforms feature can be used to add missing secrets to the default SA as part of the restore process.