This is a clone of issue OCPBUGS-43308. The following is the description of the original issue:
—
This is a clone of issue OCPBUGS-43051. The following is the description of the original issue:
—
This is a clone of issue OCPBUGS-42783. The following is the description of the original issue:
—
Context
Some ROSA HCP users host their own container registries (e.g., self-hosted Quay servers) that are only accessible from inside of their VPCs. This is often achieved through the use of private DNS zones that resolve non-public domains like quay.mycompany.intranet to non-public IP addresses. The private registries at those addresses then present self-signed SSL certificates to the client that can be validated against the HCP's additional CA trust bundle.

Problem Description
A user of a ROSA HCP cluster with a configuration like the one described above is encountering errors when attempting to import a container image from their private registry into their HCP's internal registry via oc import-image. Originally, these errors showed up in openshift-apiserver logs as DNS resolution errors, i.e., OCPBUGS-36944. After the user upgraded their cluster to 4.14.37 (which fixes OCPBUGS-36944), openshift-apiserver was able to properly resolve the domain name but complains of HTTP 502 Bad Gateway errors. We suspect these 502 Bad Gateway errors are coming from the Konnectivity-agent while it proxies traffic between the control and data planes.

We've confirmed that the private registry is accessible from the HCP data plane (worker nodes) and that the certificate presented by the registry can be validated against the cluster's additional trust bundle. IOW, curl-ing the private registry from a worker node returns a HTTP 200 OK, but doing the same from a control plane node returns a HTTP 502. Notably, this cluster is not configured with a cluster-wide proxy, nor does the user's VPC feature a transparent proxy.

Version-Release number of selected component
OCP v4.14.37

How reproducible
Can be reliably reproduced, although the network config (see Context above) is quite specific

Steps to Reproduce

1. Run the following command from the HCP data plane

   oc import-image imagegroup/imagename:v1.2.3 --from=quay.mycompany.intranet/imagegroup/imagename:v1.2.3 --confirm
   

2. Observe the command output, the resulting ImageStream object, and openshift-apiserver logs

Actual Results

error: tag v1.2.3 failed: Internal error occurred: quay.mycompany.intranet/imagegroup/imagename:v1.2.3: Get "https://quay.mycompany.intranet/v2/": Bad Gateway
imagestream.image.openshift.io/imagename imported with errors

Name:            imagename
Namespace:        mynamespace
Created:        Less than a second ago
Labels:            <none>
Annotations:        openshift.io/image.dockerRepositoryCheck=2024-10-01T12:46:02Z
Image Repository:    default-route-openshift-image-registry.apps.rosa.clustername.abcd.p1.openshiftapps.com/mynamespace/imagename
Image Lookup:        local=false
Unique Images:        0
Tags:            1

v1.2.3
  tagged from quay.mycompany.intranet/imagegroup/imagename:v1.2.3

  ! error: Import failed (InternalError): Internal error occurred: quay.mycompany.intranet/imagegroup/imagename:v1.2.3: Get "https://quay.mycompany.intranet/v2/": Bad Gateway
      Less than a second ago

error: imported completed with errors

Expected Results
Desired container image is imported from private external image registry into cluster's internal image registry without error