Enable EFA support

[ytsssun]

Issue number:

Related #1031

Description of changes:

1. Note: rdma-core will be vended as part of the release package in core-kit. PR Enable rdma core in release bottlerocket-core-kit#252.
2. Added setting generator for metadata.settings.kernel.sysctl."vm/nr_hugepages" in shared default
3. Added migration for setting generator.

Testing done:

Tested migration:

• Test upgrade:

kernel.sysctl setting
Before upgrade:

[ssm-user@control]$ apiclient get settings.kernel
{
  "settings": {
    "kernel": {
      "lockdown": "none"
    }
  }
}

Upgrade:

[ssm-user@control]$ apiclient get settings.kernel
{
  "settings": {
    "kernel": {
      "lockdown": "none",
      "sysctl": {
        "vm/nr_hugepages": "3520"
      }
    }
  }
}

setting generator for settings.kernel.sysctl.vm/nr_hugepages:
Before:

{
    "settings.network.hostname": "netdog generate-hostname",
    "settings.updates.seed": "bork seed",
    "settings.boot": "/usr/bin/prairiedog generate-boot-settings",
    "settings.updates.targets-base-url": "schnauzer-v2 render --requires 'aws@v1' --requires 'updates@v1(helpers=[tuf-prefix])' --template '{{ tuf-prefix settings.aws.region }}/targets/'",
    "settings.host-containers.admin.user-data": "shibaken generate-admin-userdata",
    "settings.host-containers.control.source": "schnauzer-v2 render --requires 'aws@v1(helpers=[ecr-prefix])' --template '{{ ecr-prefix settings.aws.region }}/bottlerocket-control:v0.7.17'",
    "settings.metrics.send-metrics": "shibaken is-partition --partition aws --partition aws-us-gov",
    "settings.aws.config": "schnauzer-v2 render --requires 'aws@v1(helpers=[aws-config])' --template '{{ aws-config settings.aws.config settings.aws.profile }}'",
    "settings.host-containers.admin.source": "schnauzer-v2 render --requires 'aws@v1(helpers=[ecr-prefix])' --template '{{ ecr-prefix settings.aws.region }}/bottlerocket-admin:v0.11.13'",
    "settings.updates.metadata-base-url": "schnauzer-v2 render --requires 'aws@v1' --requires 'updates@v1(helpers=[metadata-prefix, tuf-prefix])' --template '{{ tuf-prefix settings.aws.region }}{{ metadata-prefix settings.aws.region }}/2020-07-07/{{ os.variant_id }}/{{ os.arch }}/'"
}

After upgrades:

{
  ...
  "settings.kernel.sysctl.vm/nr_hugepages": "corndog generate-hugepages-setting", <--- New
  ...
}

• Test downgrade

Verified that settings remains:

[root@admin]# apiclient get settings.kernel
{
  "settings": {
    "kernel": {
      "lockdown": "none",
      "sysctl": {
        "vm/nr_hugepages": "3520"
      }
    }
  }
}

Settings generator is removed after downgrade.

Terms of contribution:

By submitting this pull request, I agree that this contribution is dual-licensed under the terms of both the Apache License, version 2.0, and the MIT license.

[bcressey]

We should require rdma-core over in the core kit's release package instead. You can affinitize it to just aws-* variants though.

As a general rule, I don't like adding to the set of packages all variants need to list to achieve the baseline functionality level.

[bcressey]

Is there a reason the ECS variants were omitted from this list? That should be superseded by the change I suggested, but I'm still curious.

[ytsssun]

We should require rdma-core over in the core kit's release package instead. You can affinitize it to just aws-* variants though.

Got it, it makes sense to put the rdma-core under release package. I could use a pointer on how to affinitize it to just aws-* though.

Is there a reason the ECS variants were omitted from this list? That should be superseded by the change I suggested, but I'm still curious.

Mainly because ECS does not officially support EFA? However, I can totally include that to the aws-ecs-2* variants. As for variants in general, I only wanted to introduce the rdma-core package to the variants that uses kernel-6.1 where EFA drivers are included.

[ytsssun]

As per @bcressey , I added the rdma-core package in release so that it is enabled for aws-* variants. bottlerocket-os/bottlerocket-core-kit#252

Will update this PR once the core-kit PR is merged and getting released.

[bcressey]

LGTM after it's rebased to fix Release.toml and Twoliter.toml.