zarf init into existing multi-architecture cluster

[CJLove]

Hi all,
I've successfully done zarf init in a VM and initialized a k3s cluster there and that's one path forward for getting familiar with zarf.

I also wanted to try zarf init connecting to an existing cluster that has a mix of arm and amd nodes. The zarf-docker-registry pod got into CrashLoopBackoff state; presumably because of an architecture mismatch. Is it safe to assume this ins't a supported use case yet?

Thanks!
Chris

[brandtkeller]

Hey @CJLove , I believe you are correct. Thinking through the init process I don't believe our image and node selection logic currently accounts for architecture (during the injection process).

We intend to make the init-package more configurable and provide for the ability to make the components more finely grained for explicit deployment patterns.

Does your cluster have a custom scheduler that handles scheduling workloads to nodes based on architecture or is that something you do with configuration explicitly?

[CJLove]

@brandtkeller - I have nodes labeled to indicate the architecture (e.g. arm64 or x86_64), so if I'm not deploying multi-arch containers I can use node selector to choose a nodes with the applicable architecture. Being able to do something like that with the init process would be sufficient.

Even with some edge scenarios I deal with in real life with a mix of nodes with different architectures we would be building containers targeting a specific architecture and through labeling or other mechanisms would have control over where zarf packages are getting deployed in the cluster.

[brandtkeller]

Appreciate the context - I believe we could break init configuration down further to meet the sufficient usecase and then further iterate on a mulit-arch containers solution. Would you mind opening an issue for us to track this request?

[CJLove]

Done. Created #3817