Omnia

Ansible playbook-based tools for deploying Slurm and Kubernetes clusters for High Performance Computing, Machine Learning, Deep Learning, and High-Performance Data Analytics

This project is maintained by dellhpc

Install the Omnia Control Plane

Typical layout of an HPC cluster supported by Omnia 1.2

Using Omnia 1.2, you can provision and monitor hardware devices such as servers, storage devices, network switches, and InfiniBand switches in an HPC cluster. To enable Omnia to provision or configure the supported hardware devices, Omnia requires the following connections to be made available in your HPC cluster environment.

Typical layout of a HPC cluster

List of all supported Omnia Network Topologies

  1. Dedicated Network This connection option uses the server’s dedicated iDRAC connection for management, while the production data traffic flows through the NICs on the server. This configuration separates the two flows of traffic. For more information on the topology, click on the hyperlinked heading.
    • Connecting a Pass-Through Switch: Provision and configure a 1GBE pass-through switch which will be used as a pass-through uplink switch. One of the NIC on the control plane must be connected to a data port on the pass-through switch and a second connection must be established from a data port on the pass-through switch to the management port of the TOR network switch.

      Note: Omnia is not responsible for provisioning and configuring the pass-through switch.

    • Establishing a management network: From the data ports on the pass-through switch, connect to the following ports:
      • iDRAC ports on manager and compute nodes
      • Management port on the network switches
      • Management port on the PowerVault devices
      • Management port on the InfiniBand switches

      Through this management network, management DHCP assigns IP addresses to the devices in the HPC cluster.

    • Establishing a data network:
      • Connect one of the data ports on the PowerVault device to the NIC of the compute node.
      • For InfiniBand DHCP, connect NIC on the control plane to one of the data ports of the InfiniBand Switch. Next, connect one of the data ports of the InfiniBand switch to the NIC on the compute node.
      • Establishing a host network: For Cobbler DHCP to assign an IP address to the compute node NIC, connect NIC of the control plane to the data port on the network switch. Connect another data port on the network switch to the NIC on the compute node. Omnia will provision OS on the compute nodes using PXE when the iDRAC Enterprise license is missing on any of the compute nodes in the HPC cluster.

    Note: * Cobbler web support has been discontinued from Omnia 1.2 onwards. * Note that the PowerVault NFS server should have separate NICs configured for management, data (Connecting to other compute nodes) and a dedicated data connection to the storage array. * Refer to the Control Plane Pre-Requisites file to ensure smooth running of the control_plane.

  2. Shared LOM A LOM port could be shared with the host operating system production traffic. Also, LOM ports can be dedicated to server management. For example, with a four-port LOM adapter, LOM ports one and two could be used for production data while three and four could be used for iDRAC, VNC, RDP, or other operating system-based management data. For more information on the topology, click on the hyperlinked heading.

Steps to deploy the Omnia Control Plane

  1. On the control plane, change the working directory to the directory where you want to clone the Omnia Git repository.
  2. Clone the Omnia repository using the command:
    git clone https://github.com/dellhpc/omnia.git 
    
  3. Change the directory to omnia using the command: cd omnia
  4. Edit the omnia_config.yml file to:
    • Specify the Kubernetes version which will be installed on the manager and compute nodes in the k8s_version variable. By default, it is set to 1.16.7. Edit this variable to change the version. Supported versions are 1.16.7 and 1.19.3.
    • To configure a login node in the cluster. By default, the login_node_required variable is set to “true”. Using the login node, cluster administrators can provide access to users to log in to the login node to schedule Slurm jobs. However, if you do not want to configure the login node, then you can set the variable to “false”. Without the login node, Slurm jobs can be scheduled only through the manager node.

Note: Ensure that the parameter enable_security_support in telemetry\input_params\base_vars.yml is set to ‘false’ before editing the following variables.

To configure the login node, refer to Install_Omnia.

Note:

  • Supported values for Kubernetes CNI are calico and flannel. The default value of CNI considered by Omnia is calico.
  • The default value of Kubernetes Pod Network CIDR is 10.244.0.0/16. If 10.244.0.0/16 is already in use within your network, select a different Pod Network CIDR. For more information, see https://docs.projectcalico.org/getting-started/kubernetes/quickstart.
  • The default path of the Ansible configuration file is /etc/ansible/. If the file is not present in the default path, then edit the ansible_conf_file_path variable to update the configuration path.
  • If you choose to enable security on the login node, simply follow the steps mentioned here.
  1. Change the directory to control_plane/input_params using the command: cd omnia/control_plane/input_params
  2. Edit the base_vars.yml file to update the required variables.

    Note: The IP address 192.168.25.x is used for PowerVault Storage communications. Therefore, do not use this IP address for other configurations.**

  3. Provided that the host_mapping_file_path is updated as per the provided template, Omnia deploys the control plane and assigns the component roles by executing the omnia.yml file. To deploy the Omnia control plane, run the following command:
    ansible-playbook control_plane.yml
    
  4. If the host_mapping_file_path is not provided, then you must manually assign the component roles through the AWX UI. Go to Assign component roles using AWX UI.

Omnia creates a log file which is available at: /var/log/omnia_control_plane.log.

Configurations Performed by Omnia Control Plane

The installation of omnia control plane depends largely on the variables entered in base_vars.yml. These variables decide how many functionalities of Omnia are actually required in your environment. The network configuration performed by Omnia depends on the value of network_interface_type in base_vars.yml. Omnia Control plane then provides the user with the choice of assigning management/communication IPs (device_config_support) to all available servers, switches and powervault devices per the chosen network architecture. When true, all applicable devices are given new IPs. It is recommended that when device_config_support is true, a device mapping file (Example here) is used to keep assigned IPs persistent between control plane reboots. If idrac_support true, the devices are expected to have their own IPs furnished in the filepath mentioned under device_ip_list_path. Having the IPs allows omnia to reach and configure switches, servers and powervaults without disturbing the existing network set up. Users can choose which devices require configuration using the variables ethernet_switch_support, ib_switch_support and powervault_support. Table of all supported parameter combinations and the expected outcome is provided below:


network_interface_type

device_config_support

idrac_support

Outcome

One Touch Config Support

Dedicated

TRUE

TRUE

Omnia will assign IPs to all the management ports of the different devices. iDRAC and PXE provisioning is supported. Here, ethernet, InfiniBand and powervault configurations are supported.

Yes

TRUE

FALSE

An assert failure on control_plane_common will manifest and Omnia Control Plane will fail.

No

FALSE

TRUE

Assuming the device_ip_list is populated, mgmt_container will not be used to assign the IPs to all the mgmt ports as a device_ip_list indicates that IP assignment is already done. However, ethernet, InfiniBand, powervault configurations are supported.

Yes

FALSE

FALSE

No IPs will be assigned by Omnia. Provisioning will only be through PXE.

No

lom

TRUE

TRUE

When roce_nic_ip is populated, Omnia will assign IPs to both the management and data ports. Cobbler/pxe provisioning will be done via the roce_network_nic.

Yes

When roce_nic_ip is not populated, the cobbler container will be used to assign IPs to both the iDRAC management port and the data ports. Both iDRAC and pxe mode of provisioning are supported.

No

FALSE

When roce_nic_ip is populated, management network container will come up, and it will be used to assign the management and data port IPs. This only will provide internet connection, if DNS settings are filled in base_vars.yml.

Along with this , Cobbler PXE provisioning will be done over the high speed data path or roce.

No

When roce_nic_ip is not populated, cobbler container will come up and will be responsible for mgmt. and data IP assignment as well as for providing the DNS configurations( if the parameters are given)

No

Note:

  • When network interface type is lom, idrac_support is assumed to be true irrespective of user input.
  • Omnia will not automatically assign IPs to all devices (powervault or ethernet/Infiniband switches) when network_interface_type is lom. However, if required, users can follow the linked steps.
  • Despite the value of mgmt_network_nic and host_network_nic being the same in LOM environments, the IPs assigned for management and data should not be in the same range. The start and end values of the management IP range and the host IP range cannot be the same.
  • When roce_network_nic is provided, the host_mapping_file_path is disregarded. This means that static IP assignment is not supported when high speed data paths are used for provisioning.

Once all network configuration is complete, Omnia uses AWX to integrate a centralized log system, receive live updates of running jobs, scheduled jobs, etc. AWX can also be used to assign component roles, install kuberenetes, JupyterHub, Kubeflow, Slurm, Prometheus and Grafana.

Once all network configuration is complete, Omnia uses AWX to integrate a centralized log system, receive live updates of running jobs, scheduled jobs, etc. AWX can also be used to assign component roles, install kuberenetes, JupyterHub, Kubeflow, Slurm, Prometheus and Grafana.

Credentials used by Omnia

The file login_vars.yml is populated with all credentials used by Omnia to deploy services.

If you want to view or edit the login_vars.yml file, run the following commands:

  1. cd input_params
  2. ansible-vault view login_vars.yml --vault-password-file .login_vault_key or ansible-vault edit login_vars.yml --vault-password-file .login_vault_key.

Note: It is suggested that you use the ansible-vault view or edit commands and that you do not use the ansible-vault decrypt or encrypt commands. If you have used the ansible-vault decrypt or encrypt commands, provide 644 permission to login_vars.yml.

Default Ansible CLI configurations

If AWX is not set up by the control plane (That is, when awx_web_support in base_vars.yml is set to false) , the following configurations take place:

Warning: When awx_web_support is set to false, all one touch configuration is not supported. That is, omnia.yml execution based on the host_mapping_file will not be completed.

Note:

  • In a LOM setup, running ansible-playbook collect_device_info.yml will only return IPs of iDRAC. To get the IPs of all supported devices, run ansible-playbook collect_node_info.yml.
  • All AWX job templates will fail if awx_web_support in base_vars.yml is set to false irrespective of whether AWX is running with previous configurations.

Default Ansible AWX configurations

If AWX is set up by the control plane (That is, when awx_web_support in base_vars.yml is set to true) , the following configurations take place:

Note: The AWX configurations are automatically performed by Omnia, and Dell Technologies recommends that you do not change the default configurations that are provided by Omnia as the functionality may be impacted.

The AWX UI/ Ansible CLI can be used to run playbooks such as omnia.yml. To set up red hat subscription, Slurm, Kubernetes, JupyterHub , configure devices etc., check out Using Playbooks.

Creating a new cluster

From Omnia 1.2, the cobbler container OS will follow the OS on the control plane but will deploy multiple OS’s based on the provision_os value in base_vars.yml.

Example: In a scenario where the user wishes to deploy Red Hat and Rocky on their multiple servers, below are the steps they would use:

  1. Set provision_os to redhat and iso_file_path to /root/RedHat-8.5-DVD-x86_64-Current.iso.
  2. Run control_plane.yml to provision leap and create a profile called RedHat-x86_64 in the cobbler container.
  3. Set provision_os to rocky and iso_file_path to /root/Rocky-8.x-x86_64-minimal.iso.
  4. Run control_plane.yml to provision rocky and create a profile called rocky-x86_64 in the cobbler container.

Note: All compute nodes in a cluster must run the same OS.