Input parameters for Local Repositories

• Input all required values in input/software_config.json.

Parameters for Software Configuration

Parameter	Details
cluster_os_type string Required	The operating system running on the cluster. Value: rhel, rocky
cluster_os_version string Required	The OS Version that will be provisoned on compute nodes. For RHEL and Rocky Linux, the accepted value is 8.8.
repo_config string Required	The type of offline configuration user needs. When the value is set to always, Omnia creates a local repository/registry on the OIM hosting all the packages/images required for the cluster. When the value is set to partial, Omnia creates a local repository/registry on the OIM hosting all the packages/images except those listed in the user_repo_url/user_registry in input/local_repo_config.yml. When the value is set to never, Omnia does not create a local repository/registry. All the packages/images are directly downloaded on the cluster. Note After local_repo.yml has run, the value of repo_config in input/software_config.json cannot be updated without running the oim_cleanup.yml script first. Irrespective of the value of repo_config, all local repositories that are not available as images, debian packages, or RPMs will be downloaded and configured locally on the OIM. Additionally, AMD GPU drivers, CUDA, and OFED are also downloaded by default if they are mentioned in the input/software_config.json. Accepted values: always partial <- Default never
softwares JSON list Required	A JSON list of required software and (optionally) the software version. The following software should be listed with a version in the list: BeeGFS, AMD GPU, Kubernetes, CUDA, OFED, UCX, and ROCm. A minimum of one software should be provided in the list for local_repo.yml to execute correctly. The <os_type>_software_config.json will have the basic softwares present in it. To add additional software stacks, add the software under input/software_config.json. For the list of all applicable softwares based on your <cluster_os_type>, refer the templates at examples/template_<os>_software_config.json. For example, examples/template_rhel_software_config.json. Note The accepted names for software is taken from input/config/<cluster_os_type>/<cluster_os_version>.

• Sample version for RHEL/Rocky Linux:

Note

For Rocky Linux OS, the cluster_os_type in the below sample will be rocky.

{
    "cluster_os_type": "rhel",
    "cluster_os_version": "8.8",
    "repo_config": "partial",
    "softwares": [
        {"name": "amdgpu", "version": "6.3.1"},
        {"name": "cuda", "version": "12.8.0"},
        {"name": "ofed", "version": "24.01-0.3.3.1"},
        {"name": "freeipa"},
        {"name": "openldap"},
        {"name": "secure_login_node"},
        {"name": "nfs"},
        {"name": "beegfs", "version": "7.4.5"},
        {"name": "slurm"},
        {"name": "k8s", "version":"1.31.4"},
        {"name": "jupyter"},
        {"name": "kubeflow"},
        {"name": "kserve"},
        {"name": "pytorch"},
        {"name": "tensorflow"},
        {"name": "vllm"},
        {"name": "telemetry"},
        {"name": "intel_benchmarks", "version": "2024.1.0"},
        {"name": "amd_benchmarks"},
        {"name": "utils"},
        {"name": "ucx", "version": "1.15.0"},
        {"name": "openmpi", "version": "4.1.6"},
        {"name": "csi_driver_powerscale", "version":"v2.13.0"}
    ],

    "amdgpu": [
        {"name": "rocm", "version": "6.3.1" }
    ],
    "vllm": [
        {"name": "vllm_amd"},
        {"name": "vllm_nvidia"}
    ],
    "pytorch": [
        {"name": "pytorch_cpu"},
        {"name": "pytorch_amd"},
        {"name": "pytorch_nvidia"}
    ],
    "tensorflow": [
        {"name": "tensorflow_cpu"},
        {"name": "tensorflow_amd"},
        {"name": "tensorflow_nvidia"}
    ]

}

For a list of accepted values in softwares, go to input/config/<cluster_os_type>/<cluster_os_version> and view the list of JSON files available. The filenames present in this location (without the * .json extension) are a list of accepted software names. The repositories to be downloaded for each software are listed the corresponding JSON file. For example, for a cluster running RHEL 8.8, go to input/config/rhel/8.8/ and view the file list:

amdgpu.json
bcm_roce.json
beegfs.json
cuda.json
jupyter.json
k8s.json
kserve.json
kubeflow.json
nfs.json
ofed.json
openldap.json
pytorch.json
tensorflow.json
vllm.json

For a list of repositories (and their types) configured for AMD GPUs, view the amdgpu.json file:

{
  "amdgpu": {
    "cluster": [
        {"package": "linux-headers-$(uname -r)", "type": "deb", "repo_name": "jammy"},
        {"package": "linux-modules-extra-$(uname -r)", "type": "deb", "repo_name": "jammy"},
        {"package": "amdgpu-dkms", "type": "deb", "repo_name": "amdgpu"}
    ]
  },
  "rocm": {
    "cluster": [
      {"package": "rocm", "type": "rpm", "repo_name": "rocm"}
      {"package": "rocm-validation-suite", "type": "rpm", "repo_name": "rocm"}
    ]
  }
}

Note

To configure a locally available repository that does not have a pre-defined json file, click here.

• Input the required values in input/local_repo_config.yml.

Parameters for Local Repository Configuration

Parameter	Details
repo_store_path string Required	The intended file path for offline repository data. Ensure the disk partition has enough space. Ensure that 755 permission is given to repo_store_path if user intends to use nfs share mount for repo_store_path. Default value: "/opt/omnia_repo"
user_repo_url JSON List Optional	This variable accepts the repository urls of the user which contains the packages required for the cluster. When repo_config is always, the given list will be configured on the OIM and packages required for cluster will be downloaded into a local repository. When repo_config is partial, a local repository is created on the OIM containing packages that are not part of the user’s repository. When repo_config is never, no local repository is created and packages are downloaded on all cluster nodes. ‘url’ defines the baseurl for the repository. ‘gpgkey’ defines gpgkey for the repository. If ‘gpgkey’ is omitted then gpgcheck=0 is set for that repository. Sample value: - {url: "http://crb.com/CRB/x86_64/os/",gpgkey: "http://crb.com/CRB/x86_64/os/RPM-GPG-KEY"}
user_registry JSON List Optional	This variable accepts the registry url along with port of the user which contains the images required for cluster. When repo_config is always, the list given in user_registry will be configured on the OIM and packages required for cluster will be downloaded into a local repository. If the same repository is available in both the user_repo_url and the user_registry, the repository will be configured using the values in user_registry. When repo_config is partial, a local registry is created on the OIM containing packages that are not part of the user_registry. Images listed in user_registry are directly configured as a mirror on compute nodes. Compute nodes are expected to connect to the URLs in the user_registry via http_proxy. When repo_config is never, no local registry is created and packages/images are downloaded on all cluster nodes. ‘host’ defines the URL and path to the registry. ‘cert_path’ defines the absolute path where the security certificates for each registry. If this path is not provided, insecure registries are configured. Sample value: - { host: 10.11.0.100:5001, cert_path: "/home/ca.crt" } - { host: registryhostname.registry.test, cert_path: "" }
rhel_os_url string Required	Mandatory when cluster_os_type is rhel in softwares_config.json. User has to provide the code ready builder url in order to download the packages. When repo_config is “always”, the given rhel_os_url will be configured in the OIM and packages required for cluster will be downloaded. When repo_config is “partial” or “never”, the packages required for cluster which were coming from rhel_repo_url will not be downloaded. The rhel_os_url is configured via proxy in compute nodes. Example: If cluster_os_type is rhel, rhel_os_url might be “- {url: “http://crb.com/CRB/x86_64/os/”, gpgkey: “http://crb.com/CRB/x86_64/os/RPM-GPG-KEY”}” Caution Omnia does not support adding a Red Hat subscription URL to the rhel_os_url parameter. Adding it results in an error during local_repo.yml playbook execution. For more information, click here.
omnia_repo_url_rhel JSON List Required	A list of all the repo urls from where rpms will be downloaded for Omnia features on RHEL clusters. ‘url’ defines the baseurl for the repository. ‘gpgkey’ defines gpgkey for the repository. If ‘gpgkey’ is omitted, then gpgcheck=0 is set for that repository This value is not validated by Omnia. Any errors can cause Omnia to fail. Ensure that all URLs listed below are reachable to the OIM. Default value: - { url: "https://download.docker.com/linux/centos/$releasever/$basearch/stable", gpgkey: "https://download.docker.com/linux/centos/gpg" } - { url: "https://repo.radeon.com/rocm/rhel8/{{ rocm_version }}/main", gpgkey: "https://repo.radeon.com/rocm/rocm.gpg.key" } - { url: "https://download.fedoraproject.org/pub/epel/8/Everything/$basearch", gpgkey: "https://dl.fedoraproject.org/pub/epel/RPM-GPG-KEY-EPEL-8" } - { url: "https://repo.radeon.com/amdgpu/{{ amdgpu_version }}/rhel/{{ cluster_os_version }}/main/x86_64", gpgkey: "https://repo.radeon.com/rocm/rocm.gpg.key" } - { url: "https://www.beegfs.io/release/beegfs_{{beegfs_version}}/dists/rhel8", gpgkey: "https://www.beegfs.io/release/beegfs_{{beegfs_version}}/gpg/GPG-KEY-beegfs" } - { url: "https://yum.repos.intel.com/oneapi", gpgkey: "https://yum.repos.intel.com/intel-gpg-keys/GPG-PUB-KEY-INTEL-SW-PRODUCTS.PUB" } - { url: "https://ltb-project.org/rpm/openldap25/$releasever/$basearch", gpgkey: ""}
omnia_repo_url_rocky JSON List Required	A list of all the repo urls from where rpms will be downloaded for Omnia features on Rocky clusters. ‘url’ defines the baseurl for the repository. ‘gpgkey’ defines gpgkey for the repository. If ‘gpgkey’ is omitted, then gpgcheck=0 is set for that repository This value is not validated by Omnia. Any errors can cause Omnia to fail. Ensure that all URLs listed below are reachable to the OIM. Default value: - { url: "https://download.docker.com/linux/centos/$releasever/$basearch/stable", gpgkey: "https://download.docker.com/linux/centos/gpg" } - { url: "https://repo.radeon.com/rocm/rhel8/{{ rocm_version }}/main", gpgkey: "https://repo.radeon.com/rocm/rocm.gpg.key" } - { url: "https://download.fedoraproject.org/pub/epel/8/Everything/$basearch", gpgkey: "https://dl.fedoraproject.org/pub/epel/RPM-GPG-KEY-EPEL-8" } - { url: "https://repo.radeon.com/amdgpu/{{ amdgpu_version }}/rhel/{{ cluster_os_version }}/main/x86_64", gpgkey: "https://repo.radeon.com/rocm/rocm.gpg.key" } - { url: "https://www.beegfs.io/release/beegfs_{{beegfs_version}}/dists/rhel8", gpgkey: "https://www.beegfs.io/release/beegfs_{{beegfs_version}}/gpg/GPG-KEY-beegfs" } - { url: "https://yum.repos.intel.com/oneapi", gpgkey: "https://yum.repos.intel.com/intel-gpg-keys/GPG-PUB-KEY-INTEL-SW-PRODUCTS.PUB" } - { url: "https://ltb-project.org/rpm/openldap25/$releasever/$basearch", gpgkey: ""} - { url: "http://dl.rockylinux.org/$contentdir/$releasever/PowerTools/$basearch/os/", gpgkey: ""}

• Input docker_username and docker_password in input/provision_config_credentials.yml to avoid image pullback errors.

If you have any feedback about Omnia documentation, please reach out at omnia.readme@dell.com.