Add support for ARM node groups

[gurchik]

Currently eks-operator only supports 3 types of node groups:

• eks.AMITypesAl2X8664 - for x86_64 nodes (default case)
• eks.AMITypesAl2X8664Gpu - for GPU x86_64 nodes (used when the gpu input is set to true)
• eks.AMITypesCustom - used in all other cases, requires the user to create a custom launch template

https://github.com/rancher/eks-operator/blob/3e0b47a570397a23c82f8fb6ce8e8fee34582e0d/pkg/eks/create.go#L273-L281

It would be great if there was also support for eks.AMITypesAl2Arm64. This would allow the creation of Managed Node Groups that are running ARM. Currently the only way to target ARM is to create a custom launch template and specify it with the launch_template input. Creating a custom launch template is not preferable because, among other things, it requires creating a custom userdata script and you lose the ability to perform a managed node group update.

This constant is available in the version of aws-sdk-go being used: https://pkg.go.dev/github.com/aws/aws-sdk-go@v1.45.6/service/eks#pkg-constants

Desirable result would be to have an arm input I can set to true.

[kkaempf]

@davidstauffer 👆🏻

[valaparthvi]

Tested via API on 2.9-head and it works fine.

"nodeGroups": [
    {
        "arm": true,
        "desiredSize": 2,
        "diskSize": 20,
        "ec2SshKey": null,
        "gpu": null,
        "imageId": null,
        "instanceType": "a1.medium",
        "labels": { },
        "launchTemplate": null,
        "maxSize": 3,
        "minSize": 1,
        "nodeRole": "arn:aws:iam::879933548321:role/auto-ekshostcluster-hp-cipze-node--NodeInstanceRole-YyTRKONmeTTW",
        "nodegroupName": "dng",
        "requestSpotInstances": false,
        "resourceTags": { },
        "spotInstanceTypes": null,
        "subnets": [ 2 items
            "subnet-0196175050481fd97",
            "subnet-06bbbb08fbb70e046"
        ],
        "tags": { },
        "type": "/v3/schemas/nodeGroup",
        "userData": null,
        "version": "1.28"
    }
]

1. Add EKS with arm
2. Add nodegroup
3. Delete nodegroup
4. Scale the nodegroup up/down; here I noticed some flakes, for e.g. For an existing nodegroup with InstanceType: a1.large, scaling up returned the following error: [Syncing error] error for cluster [c-vzmmp]: health error for node group [dng] in cluster [c-vzmmp]: Could not launch On-Demand Instances. Unsupported - Your requested instance type (a1.medium) is not supported in your requested Availability Zone (ap-south-1a). Please retry your request by not specifying an Availability Zone or choosing ap-south-1b, ap-south-1c. Launching EC2 instance failed.

Changes that UI will need:

1. UI is missing a checkbox for "ARM".
2. Instance types compatible with ARM are not visible in UI.

Overall I think it works well.

[furkatgofurov7]

@kkaempf can we get help from UI for:

Changes that UI will need:

1. UI is missing a checkbox for "ARM".
2. Instance types compatible with ARM are not visible in UI.

[mjura]

I don's see UI ticket assigned to it, @valaparthvi do you know if UI issue was also created.

[valaparthvi]

There is no UI ticket open. Should I transfer this issue to rancher/dashboard or create a new one? @mjura

[mjura]

There is no UI ticket open. Should I transfer this issue to rancher/dashboard or create a new one? @mjura

Yes, please transfer this issue to rancher/dashboard

[kkaempf]

@gaktive @nwmac - ptal 😉

[nwmac]

Let's not move tickets to the dashboard repo with a milestone that has passed - updating the milestone to 2.11.0 FYI @gaktive - will need to review - this probably needs a JIRA ticket as well.

[gaktive]

Internal reference: SURE-8886