[BUG] Trying to run amd64 images on M1 arm fails

[IBMRob]

General information

• OS: macOS
• Hypervisor: default
• Did you run crc setup before starting it (Yes/No)? Yes
• Running CRC on: Laptop

CRC version

convery@Robs-M1 ibm-appconnect % crc version
CRC version: 2.12.0+ea98bb41
OpenShift version: 4.11.18
Podman version: 4.2.0

CRC status

DEBU CRC version: 2.12.0+ea98bb41                 
DEBU OpenShift version: 4.11.18                   
DEBU Podman version: 4.2.0                        
DEBU Running 'crc status'                         
CRC VM:          Running
OpenShift:       Running (v4.11.18)
RAM Usage:       8.041GB of 9.353GB
Disk Usage:      15.14GB of 32.74GB (Inside the CRC VM)
Cache Usage:     73.63GB
Cache Directory: /Users/convery/.crc/cache

CRC config

- consent-telemetry                     : no

Host Operating System

ProductName:		macOS
ProductVersion:		13.1
BuildVersion:		22C65

Steps to reproduce

1. Add a catalog source which doesn't have an ARM version but does have an amd64 version i.e.

apiVersion: operators.coreos.com/v1alpha1
kind: CatalogSource
metadata:
 name: ibm-operator-catalog
 namespace: openshift-marketplace
spec:
 displayName: IBM Operator Catalog
 publisher: IBM
 sourceType: grpc
 image: icr.io/cpopen/ibm-operator-catalog
 updateStrategy:
   registryPoll:
     interval: 45m

Expected

I was hoping that it would be possible to get OCP to fall back and use the amd64 images like its possible on docker i.e. I can pull the image using docker pull -platform=linux/amd64 icr.io/cpopen/ibm-operator-catalog and then run it as a normal image using docker run icr.io/cpopen/ibm-operator-catalog

Is it possible to get it to run amd64 containers?

Actual

Pull fails due to Failed to pull image "icr.io/cpopen/ibm-operator-catalog": rpc error: code = Unknown desc = choosing image instance: no image found in manifest list for architecture arm64, variant "v8", OS linux

Logs

Before gather the logs try following if that fix your issue

$ crc delete -f
$ crc cleanup
$ crc setup
$ crc start --log-level debug

Please consider posting the output of crc start --log-level debug on http://gist.github.com/ and post the link in the issue.

[praveenkumar]

@IBMRob This looks like an icr.io/cpopen/ibm-operator-catalog image maintainer should create it multiarch support image so image would be pull as per arch and iirc this is the right way for having workload on different arch for OpenShift.

This is not a CRC issue.

[IBMRob]

While I agree in an ideal world everyone out there would build arm versions of their products but we know thats not going to happen thus why Apple made sure that it could run amd64 code and the reason that other container offerings such as Docker made sure that it supported also running amd64 binary images thus why we are looking for crc to provide a similar capability

Or the ability to run the whole amd64 variant of CRC on ARM. Is there a way to force the crc cli to run the amd64 version and install the amd64 containers?

[gbraad]

This is not a CRC issue.

To clarify, we have the userspace qemu binaries for the podman images, but not for OpenShift. This is something that I will have to take forward to their engineering.

We might wanna experiment with this, but might end up in a 'grey area' as it is hard to track what might cause an issue. However, I can imagine an ARM-based cluster might run a Java (x86-based) container.

[mariomac]

As an Openshift Local/CRC user, being able to setup an AMD64 Local Openshift on Mac M1 would be awesome for development purposes.

Thank you for such an amazing tool!

[cameronlukeroberts]

I would like to echo the above comment. This would be really useful for us.

Thank you 😄

[cfergeau]

Running a full openshift amd64 instance won't be possible on the M1. Running an openshift arm64 instance, and starting some amd64 images on it might be doable, but I suspect we'd be hitting some openshift limitations here and there.

[cameronlukeroberts]

Interesting!

Running an openshift arm64 instance, and starting some amd64 images on it might be doable

I think this would be useful if it were to work. Is it a large piece of work to enable the above?