Created attachment 1864162 [details]
AWS cluster provisioning

Description of the problem:

I am trying to provision an amd64 AWS cluster from an ACM hub running on ppc64le.
This used to work by using a multi-arch ocp-release image (as documented here https://access.redhat.com/documentation/en-us/red_hat_advanced_cluster_management_for_kubernetes/2.4/html/release_notes/red-hat-advanced-cluster-management-for-kubernetes-release-notes#no-create-clusters-across-architectures).

With the support of ARM64 in AWS, they have added additional architecture checks in the OCP 4.9 installer:
https://github.com/openshift/installer/blob/325ac4a55ab80e5720829ff9d9b175eb682eb35b/pkg/types/aws/validation/machinepool.go#L15
https://github.com/openshift/installer/blob/325ac4a55ab80e5720829ff9d9b175eb682eb35b/pkg/types/validation/machinepools.go#L78

This breaks cross-architecture cluster provisioning with ACM, even when using a multi-arch ocp-release image:

time="2022-03-02T12:30:13Z" level=info msg="running openshift-install create manifests" installID=mlvd92q4
time="2022-03-02T12:30:13Z" level=info msg="running openshift-install binary" args="[create manifests]" installID=mlvd92q4
time="2022-03-02T12:30:14Z" level=info msg="found file" installID=mlvd92q4 path=/output/.openshift_install.log
time="2022-03-02T12:30:14Z" level=info msg="all files found, ready to proceed" installID=mlvd92q4
time="2022-03-02T12:30:14Z" level=debug msg="OpenShift Installer v4.9.0"
time="2022-03-02T12:30:14Z" level=debug msg="Built from commit e2850d67bbbf1a86101f665d39e1ca7f13965487"
time="2022-03-02T12:30:14Z" level=debug msg="Fetching Master Machines..."
time="2022-03-02T12:30:14Z" level=debug msg="Loading Master Machines..."
time="2022-03-02T12:30:14Z" level=debug msg="  Loading Cluster ID..."
time="2022-03-02T12:30:14Z" level=debug msg="    Loading Install Config..."
time="2022-03-02T12:30:14Z" level=debug msg="      Loading SSH Key..."
time="2022-03-02T12:30:14Z" level=debug msg="      Loading Base Domain..."
time="2022-03-02T12:30:14Z" level=debug msg="        Loading Platform..."
time="2022-03-02T12:30:14Z" level=debug msg="      Loading Cluster Name..."
time="2022-03-02T12:30:14Z" level=debug msg="        Loading Base Domain..."
time="2022-03-02T12:30:14Z" level=debug msg="        Loading Platform..."
time="2022-03-02T12:30:14Z" level=debug msg="      Loading Networking..."
time="2022-03-02T12:30:14Z" level=debug msg="        Loading Platform..."
time="2022-03-02T12:30:14Z" level=debug msg="      Loading Pull Secret..."
time="2022-03-02T12:30:14Z" level=debug msg="      Loading Platform..."
time="2022-03-02T12:30:14Z" level=fatal msg="failed to fetch Master Machines: failed to load asset \"Install Config\": invalid \"install-config.yaml\" file: [controlPlane.architecture: Unsupported value: \"ppc64le\": supported values: \"amd64\", \"arm64\", compute[0].architecture: Unsupported value: \"ppc64le\": supported values: \"amd64\", \"arm64\"]"
time="2022-03-02T12:30:15Z" level=error msg="error after waiting for command completion" error="exit status 1" installID=mlvd92q4
time="2022-03-02T12:30:15Z" level=error msg="error generating installer assets" error="exit status 1" installID=mlvd92q4


The reason is that ACM does not specify a target architecture in the install-config.yaml file. It thus defaults to the archictecture of the hub (in this case ppc64le) which is not an architecture supported by AWS.

This works fine if I use a multi-arch OCP 4.8 ocp-release image because there was no such architecture check in OCP 4.8 for AWS.

A possible workaround is to edit the install-config.yaml file in the ACM console before creating the cluster, by adding the architecture: amd64 attribute to the control plane and compute nodes like in the attached screenshot.


Release version:
2.5

Operator snapshot version:
2.5.0-DOWNSTREAM-2022-02-16-19-52-49

OCP version:
OCP 4.9.23

Browser Info:

Steps to reproduce:
1. Deploy ACM 2.5 on ppc64le hub.
2. Create an AWS cluster from ACM console (using a multi-arch OCP 4.9 release image).
3. Check the cluster creation logs.
4. See error about unsupported ppc64le architecture.

Actual results:
AWS cluster provisioning fails.

Expected results:
We should be able to provision AWS amd64 or arm64 clusters from ACM running on ppc64le (or s390x) when using a multi-arch ocp-release image. There should be a way to specify the target architecture other than editing the YAML.

Additional info: