Install Kubernetes Platform on Standard with Dedicated Storage

This section describes the steps to install the StarlingX Kubernetes platform on a Standard with Dedicated Storage deployment configuration.

Create bootable USB

Refer to Bootable USB for instructions on how to create a bootable USB with the StarlingX ISO on your system.

Install software on controller-0

  1. Insert the bootable USB into a bootable USB port on the host you are configuring as controller-0.

  2. Power on the host.

  3. Attach to a console, ensure the host boots from the USB, and wait for the StarlingX Installer Menus.

  4. Make the following menu selections in the installer:

    1. First menu: Select ‘Standard Controller Configuration’

    2. Second menu: Select ‘Graphical Console’ or ‘Textual Console’ depending on your terminal access to the console port.

  5. Wait for non-interactive install of software to complete and server to reboot. This can take 5-10 minutes, depending on the performance of the server.

Bootstrap system on controller-0

  1. Login using the username / password of “sysadmin” / “sysadmin”.

    When logging in for the first time, you will be forced to change the password.

    Login: sysadmin
    Password:
    Changing password for sysadmin.
    (current) UNIX Password: sysadmin
    New Password:
    (repeat) New Password:
    
  2. Verify and/or configure IP connectivity.

    External connectivity is required to run the Ansible bootstrap playbook. The StarlingX boot image will DHCP out all interfaces so the server may have obtained an IP address and have external IP connectivity if a DHCP server is present in your environment. Verify this using the ip addr and ping 8.8.8.8 commands.

    Otherwise, manually configure an IP address and default IP route. Use the PORT, IP-ADDRESS/SUBNET-LENGTH and GATEWAY-IP-ADDRESS applicable to your deployment environment.

    sudo ip address add <IP-ADDRESS>/<SUBNET-LENGTH> dev <PORT>
    sudo ip link set up dev <PORT>
    sudo ip route add default via <GATEWAY-IP-ADDRESS> dev <PORT>
    ping 8.8.8.8
    
  3. Specify user configuration overrides for the Ansible bootstrap playbook.

    Ansible is used to bootstrap StarlingX on controller-0. Key files for Ansible configuration are:

    /etc/ansible/hosts

    The default Ansible inventory file. Contains a single host: localhost.

    /usr/share/ansible/stx-ansible/playbooks/bootstrap.yml

    The Ansible bootstrap playbook.

    /usr/share/ansible/stx-ansible/playbooks/host_vars/bootstrap/default.yml

    The default configuration values for the bootstrap playbook.

    sysadmin home directory ($HOME)

    The default location where Ansible looks for and imports user configuration override files for hosts. For example: $HOME/<hostname>.yml.

    Important

    Some Ansible bootstrap parameters cannot be changed or are very difficult to change after installation is complete.

    Review the set of install-time-only parameters before installation and confirm that your values for these parameters are correct for the desired installation.

    Refer to Ansible install-time-only parameters for details.

    Specify the user configuration override file for the Ansible bootstrap playbook using one of the following methods:

    Note

    This Ansible Overrides file for the Bootstrap Playbook ($HOME/localhost.yml) contains security sensitive information, use the ansible-vault create $HOME/localhost.yml command to create it. You will be prompted for a password to protect/encrypt the file. Use the ansible-vault edit $HOME/localhost.yml command if the file needs to be edited after it is created.

    1. Use a copy of the default.yml file listed above to provide your overrides.

      The default.yml file lists all available parameters for bootstrap configuration with a brief description for each parameter in the file comments.

      To use this method, run the ansible-vault create $HOME/localhost.yml command and copy the contents of the default.yml file into the ansible-vault editor, and edit the configurable values as required.

    2. Create a minimal user configuration override file.

      To use this method, create your override file with the ansible-vault create $HOME/localhost.yml command and provide the minimum required parameters for the deployment configuration as shown in the example below. Use the OAM IP SUBNET and IP ADDRESSing applicable to your deployment environment.

      cd ~
      
      cat <<EOF > localhost.yml
      
      system_mode: duplex
      
      dns_servers:
        - 8.8.8.8
        - 8.8.4.4
      
      external_oam_subnet: <OAM-IP-SUBNET>/<OAM-IP-SUBNET-LENGTH>
      external_oam_gateway_address: <OAM-GATEWAY-IP-ADDRESS>
      external_oam_floating_address: <OAM-FLOATING-IP-ADDRESS>
      external_oam_node_0_address: <OAM-CONTROLLER-0-IP-ADDRESS>
      external_oam_node_1_address: <OAM-CONTROLLER-1-IP-ADDRESS>
      
      admin_username: admin
      admin_password: <admin-password>
      ansible_become_pass: <sysadmin-password>
      
      # OPTIONALLY provide a ROOT CA certificate and key for k8s root ca,
      # if not specified, one will be auto-generated,
      # see ‘Kubernetes Root CA Certificate’ in Security Guide for details.
      k8s_root_ca_cert: < your_root_ca_cert.pem >
      k8s_root_ca_key: < your_root_ca_key.pem >
      apiserver_cert_sans:
        - < your_hostname_for_oam_floating.your_domain >
      
      EOF
      

      In either of the above options, the bootstrap playbook’s default values will pull all container images required for the StarlingX Platform from Docker hub.

      If you have setup a private Docker registry to use for bootstrapping then you will need to add the following lines in $HOME/localhost.yml:

      docker_registries:
        quay.io:
           url: myprivateregistry.abc.com:9001/quay.io
        docker.elastic.co:
           url: myprivateregistry.abc.com:9001/docker.elastic.co
        gcr.io:
           url: myprivateregistry.abc.com:9001/gcr.io
        ghcr.io:
           url: myprivateregistry.abc.com:9001/gcr.io
        k8s.gcr.io:
           url: myprivateregistry.abc.com:9001/k8s.ghcr.io
        docker.io:
           url: myprivateregistry.abc.com:9001/docker.io
        defaults:
           type: docker
           username: <your_myprivateregistry.abc.com_username>
           password: <your_myprivateregistry.abc.com_password>
      
      # Add the CA Certificate that signed myprivateregistry.abc.com’s
      # certificate as a Trusted CA
      ssl_ca_cert: /home/sysadmin/myprivateregistry.abc.com-ca-cert.pem
      

      See Use a Private Docker Registry for more information.

      If a firewall is blocking access to Docker hub or your private registry from your StarlingX deployment, you will need to add the following lines in $HOME/localhost.yml (see Docker Proxy Configuration for more details about Docker proxy settings):

      # Add these lines to configure Docker to use a proxy server
      docker_http_proxy: http://my.proxy.com:1080
      docker_https_proxy: https://my.proxy.com:1443
      docker_no_proxy:
         - 1.2.3.4
      

      Refer to Ansible Bootstrap Configurations for information on additional Ansible bootstrap configurations for advanced Ansible bootstrap scenarios.

  4. Run the Ansible bootstrap playbook:

    Note

    Before running the Ansible bootstrap playbook, it is important that you ensure that controller-0 server time is synchronized correctly. Run the following command:

    # check the current server time
    $ date
    
    # if the current server time is not correct, update with NTP
    
          # first add nameserver for DNS resolution
          $ echo "nameserver 8.8.8.8" >> /etc/resolv.conf
          $ echo "nameserver 8.8.4.4" >> /etc/resolv.conf
    
          # run ntpdate
          $ sudo ntpdate 0.pool.ntp.org 1.pool.ntp.org
    
    ansible-playbook --ask-vault-pass /usr/share/ansible/stx-ansible/playbooks/bootstrap.yml
    

    Wait for Ansible bootstrap playbook to complete. This can take 5-10 minutes, depending on the performance of the host machine.

Configure controller-0

  1. Acquire admin credentials:

    source /etc/platform/openrc
    
  2. Configure the OAM interface of controller-0 and specify the attached network as “oam”.

    The following example configures the OAM interface on a physical untagged ethernet port, use the OAM port name that is applicable to your deployment environment, for example eth0:

    OAM_IF=<OAM-PORT>
    system host-if-modify controller-0 $OAM_IF -c platform
    system interface-network-assign controller-0 $OAM_IF oam
    

    To configure a vlan or aggregated ethernet interface, see Node Interfaces.

  3. Configure the MGMT interface of controller-0 and specify the attached networks of both “mgmt” and “cluster-host”.

    The following example configures the MGMT interface on a physical untagged ethernet port, use the MGMT port name that is applicable to your deployment environment, for example eth1:

    MGMT_IF=<MGMT-PORT>
    
    # De-provision loopback interface and
    # remove mgmt and cluster-host networks from loopback interface
    system host-if-modify controller-0 lo -c none
    IFNET_UUIDS=$(system interface-network-list controller-0 | awk '{if ($6=="lo") print $4;}')
    for UUID in $IFNET_UUIDS; do
        system interface-network-remove ${UUID}
    done
    
    # Configure management interface and assign mgmt and cluster-host networks to it
    system host-if-modify controller-0 $MGMT_IF -c platform
    system interface-network-assign controller-0 $MGMT_IF mgmt
    system interface-network-assign controller-0 $MGMT_IF cluster-host
    

    To configure a vlan or aggregated ethernet interface, see Node Interfaces.

  4. Configure NTP servers for network time synchronization:

    system ntp-modify ntpservers=0.pool.ntp.org,1.pool.ntp.org
    

    To configure PTP instead of NTP, see PTP Server Configuration.

  5. If required, configure Ceph storage backend:

    A persistent storage backend is required if your application requires PVCs.

    Important

    The StarlingX OpenStack application requires PVCs.

    system storage-backend-add ceph --confirmed
    

OpenStack-specific host configuration

Important

These steps are required only if the StarlingX OpenStack application (stx-openstack) will be installed.

  1. For OpenStack only: Assign OpenStack host labels to controller-0 in support of installing the stx-openstack manifest and helm-charts later.

    system host-label-assign controller-0 openstack-control-plane=enabled
    
  2. For OpenStack only: Configure the system setting for the vSwitch.

    StarlingX has OVS (kernel-based) vSwitch configured as default:

    • Runs in a container; defined within the helm charts of stx-openstack manifest.

    • Shares the core(s) assigned to the platform.

    If you require better performance, OVS-DPDK (OVS with the Data Plane Development Kit, which is supported only on bare metal hardware) should be used:

    • Runs directly on the host (it is not containerized). Requires that at least 1 core be assigned/dedicated to the vSwitch function.

    To deploy the default containerized OVS:

    system modify --vswitch_type none
    

    This does not run any vSwitch directly on the host, instead, it uses the containerized OVS defined in the helm charts of stx-openstack manifest.

    To deploy OVS-DPDK, run the following command:

    system modify --vswitch_type ovs-dpdk

    Once vswitch_type is set to OVS-DPDK, any subsequent AIO-controller or worker nodes created will default to automatically assigning 1 vSwitch core for AIO controllers and 2 vSwitch cores (both on numa-node 0; physical NICs are typically on first numa-node) for compute-labeled worker nodes.

    Note

    After controller-0 is unlocked, changing vswitch_type requires locking and unlocking controller-0 to apply the change.

Unlock controller-0

Important

Make sure the Ceph storage backend is configured. If it is not configured, you will not be able to configure storage nodes.

Unlock controller-0 in order to bring it into service:

system host-unlock controller-0

Controller-0 will reboot in order to apply configuration changes and come into service. This can take 5-10 minutes, depending on the performance of the host machine.

Install software on controller-1, storage nodes, and worker nodes

  1. Power on the controller-1 server and force it to network boot with the appropriate BIOS boot options for your particular server.

  2. As controller-1 boots, a message appears on its console instructing you to configure the personality of the node.

  3. On the console of controller-0, list hosts to see newly discovered controller-1 host (hostname=None):

    system host-list
    +----+--------------+-------------+----------------+-------------+--------------+
    | id | hostname     | personality | administrative | operational | availability |
    +----+--------------+-------------+----------------+-------------+--------------+
    | 1  | controller-0 | controller  | unlocked       | enabled     | available    |
    | 2  | None         | None        | locked         | disabled    | offline      |
    +----+--------------+-------------+----------------+-------------+--------------+
    
  4. Using the host id, set the personality of this host to ‘controller’:

    system host-update 2 personality=controller
    

    This initiates the install of software on controller-1. This can take 5-10 minutes, depending on the performance of the host machine.

  5. While waiting for the previous step to complete, power on the storage-0 and storage-1 servers. Set the personality to ‘storage’ and assign a unique hostname for each.

    For example, power on storage-0 and wait for the new host (hostname=None) to be discovered by checking ‘system host-list’:

    system host-update 3 personality=storage
    

    Repeat for storage-1. Power on storage-1 and wait for the new host (hostname=None) to be discovered by checking ‘system host-list’:

    system host-update 4 personality=storage
    

    This initiates the software installation on storage-0 and storage-1. This can take 5-10 minutes, depending on the performance of the host machine.

  6. While waiting for the previous step to complete, power on the worker nodes. Set the personality to ‘worker’ and assign a unique hostname for each.

    For example, power on worker-0 and wait for the new host (hostname=None) to be discovered by checking ‘system host-list’:

    system host-update 5 personality=worker hostname=worker-0
    

    Repeat for worker-1. Power on worker-1 and wait for the new host (hostname=None) to be discovered by checking ‘system host-list’:

    system host-update 6 personality=worker hostname=worker-1
    

    This initiates the install of software on worker-0 and worker-1.

    Note

    A node with Edgeworker personality is also available. See Deploy Edgeworker Nodes for details.

  7. Wait for the software installation on controller-1, storage-0, storage-1, worker-0, and worker-1 to complete, for all servers to reboot, and for all to show as locked/disabled/online in ‘system host-list’.

    system host-list
    +----+--------------+-------------+----------------+-------------+--------------+
    | id | hostname     | personality | administrative | operational | availability |
    +----+--------------+-------------+----------------+-------------+--------------+
    | 1  | controller-0 | controller  | unlocked       | enabled     | available    |
    | 2  | controller-1 | controller  | locked         | disabled    | online       |
    | 3  | storage-0    | storage     | locked         | disabled    | online       |
    | 4  | storage-1    | storage     | locked         | disabled    | online       |
    | 5  | worker-0     | worker      | locked         | disabled    | online       |
    | 6  | worker-1     | worker      | locked         | disabled    | online       |
    +----+--------------+-------------+----------------+-------------+--------------+
    

Configure controller-1

  1. Configure the OAM interface of controller-1 and specify the attached network of “oam”.

    The following example configures the OAM interface on a physical untagged ethernet port, use the OAM port name that is applicable to your deployment environment, for example eth0:

    OAM_IF=<OAM-PORT>
    system host-if-modify controller-1 $OAM_IF -c platform
    system interface-network-assign controller-1 $OAM_IF oam
    

    To configure a vlan or aggregated ethernet interface, see Node Interfaces.

  2. The MGMT interface is partially set up by the network install procedure; configuring the port used for network install as the MGMT port and specifying the attached network of “mgmt”.

    Complete the MGMT interface configuration of controller-1 by specifying the attached network of “cluster-host”.

    system interface-network-assign controller-1 mgmt0 cluster-host
    

OpenStack-specific host configuration

Important

This step is required only if the StarlingX OpenStack application (stx-openstack) will be installed.

For OpenStack only: Assign OpenStack host labels to controller-1 in support of installing the stx-openstack manifest and helm-charts later.

system host-label-assign controller-1 openstack-control-plane=enabled

Unlock controller-1

Unlock controller-1 in order to bring it into service:

system host-unlock controller-1

Controller-1 will reboot in order to apply configuration changes and come into service. This can take 5-10 minutes, depending on the performance of the host machine.

  • For OpenStack only: Due to the additional openstack services’ containers running on the controller host, the size of the docker filesystem needs to be increased from the default size of 30G to 60G.

    # check existing size of docker fs
    system host-fs-list controller-1
    
    # check available space (Avail Size (GiB)) in cgts-vg LVG where docker fs is located
    system host-lvg-list controller-1
    
    # if existing docker fs size + cgts-vg available space is less than
    # 60G, you will need to add a new disk partition to cgts-vg.
    
            # Assuming you have unused space on ROOT DISK, add partition to ROOT DISK.
            # ( if not use another unused disk )
    
            # Get device path of ROOT DISK
            system host-show controller-1 | fgrep rootfs
    
            # Get UUID of ROOT DISK by listing disks
            system host-disk-list controller-1
    
            # Create new PARTITION on ROOT DISK, and take note of new partition’s ‘uuid’ in response
            # Use a partition size such that you’ll be able to increase docker fs size from 30G to 60G
            PARTITION_SIZE=30
            system host disk-partition-add -t lvm_phys_vol controller-1 <root-disk-uuid> ${PARTITION_SIZE}
    
            # Add new partition to ‘cgts-vg’ local volume group
            system host-pv-add controller-1 cgts-vg <NEW_PARTITION_UUID>
            sleep 2    # wait for partition to be added
    
    # Increase docker filesystem to 60G
    system host-fs-modify controller-1 docker=60
    

Note

Controller-0 and controller-1 use IP multicast messaging for synchronization. If loss of synchronization occurs a few minutes after controller-1 becomes available, ensure that the switches and other devices on the management and infrastructure networks are configured with appropriate settings.

In particular, if IGMP snooping is enabled on ToR switches, then a device acting as an IGMP querier is required on the network (on the same VLAN) to prevent nodes from being dropped from the multicast group. The IGMP querier periodically sends IGMP queries to all nodes on the network, and each node sends an IGMP join or report in response. Without an IGMP querier, the nodes do not periodically send IGMP join messages after the initial join sent when the link first goes up, and they are eventually dropped from the multicast group.

Configure storage nodes

  1. Assign the cluster-host network to the MGMT interface for the storage nodes:

    (Note that the MGMT interfaces are partially set up automatically by the network install procedure.)

    for NODE in storage-0 storage-1; do
       system interface-network-assign $NODE mgmt0 cluster-host
    done
    
  2. Add OSDs to storage-0.

    HOST=storage-0
    
    # List host’s disks and identify disks you want to use for CEPH OSDs, taking note of their UUID
    # By default, /dev/sda is being used as system disk and can not be used for OSD.
    system host-disk-list ${HOST}
    
    # Add disk as an OSD storage
    system host-stor-add ${HOST} osd <disk-uuid>
    
    # List OSD storage devices and wait for configuration of newly added OSD to complete.
    system host-stor-list ${HOST}
    
  3. Add OSDs to storage-1.

    HOST=storage-1
    
    # List host’s disks and identify disks you want to use for CEPH OSDs, taking note of their UUID
    # By default, /dev/sda is being used as system disk and can not be used for OSD.
    system host-disk-list ${HOST}
    
    # Add disk as an OSD storage
    system host-stor-add ${HOST} osd <disk-uuid>
    
    # List OSD storage devices and wait for configuration of newly added OSD to complete.
    system host-stor-list ${HOST}
    

Unlock storage nodes

Unlock storage nodes in order to bring them into service:

for STORAGE in storage-0 storage-1; do
   system host-unlock $STORAGE
done

The storage nodes will reboot in order to apply configuration changes and come into service. This can take 5-10 minutes, depending on the performance of the host machine.

Configure worker nodes

  1. The MGMT interfaces are partially set up by the network install procedure; configuring the port used for network install as the MGMT port and specifying the attached network of “mgmt”.

    Complete the MGMT interface configuration of the worker nodes by specifying the attached network of “cluster-host”.

    for NODE in worker-0 worker-1; do
       system interface-network-assign $NODE mgmt0 cluster-host
    done
    

OpenStack-specific host configuration

Important

These steps are required only if the StarlingX OpenStack application (stx-openstack) will be installed.

  1. For OpenStack only: Assign OpenStack host labels to the worker nodes in support of installing the stx-openstack manifest and helm-charts later.

    for NODE in worker-0 worker-1; do
      system host-label-assign $NODE  openstack-compute-node=enabled
      kubectl taint nodes $NODE openstack-compute-node:NoSchedule
      system host-label-assign $NODE  openvswitch=enabled
      system host-label-assign $NODE  sriov=enabled
    done
  2. For OpenStack only: Configure the host settings for the vSwitch.

    If using OVS-DPDK vSwitch, run the following commands:

    Default recommendation for worker node is to use node two cores on numa-node 0 for OVS-DPDK vSwitch; physical NICs are typically on first numa-node. This should have been automatically configured, if not run the following command.

    for NODE in worker-0 worker-1; do
    
        # assign 2 cores on processor/numa-node 0 on worker-node to vswitch
        system host-cpu-modify -f vswitch -p0 2 $NODE
    
    done
    

    When using OVS-DPDK, configure 1G of huge pages for vSwitch memory on each NUMA node on the host. It is recommended to configure 1x 1G huge page (-1G 1) for vSwitch memory on each NUMA node on the host.

    However, due to a limitation with Kubernetes, only a single huge page size is supported on any one host. If your application VMs require 2M huge pages, then configure 500x 2M huge pages (-2M 500) for vSwitch memory on each NUMA node on the host.

    for NODE in worker-0 worker-1; do
    
      # assign 1x 1G huge page on processor/numa-node 0 on worker-node to vswitch
      system host-memory-modify -f vswitch -1G 1 $NODE 0
    
      # assign 1x 1G huge page on processor/numa-node 0 on worker-node to vswitch
      system host-memory-modify -f vswitch -1G 1 $NODE 1
    
    done
    

    Important

    VMs created in an OVS-DPDK environment must be configured to use huge pages to enable networking and must use a flavor with property: hw:mem_page_size=large

    Configure the huge pages for VMs in an OVS-DPDK environment on this host, the following commands are an example that assumes that 1G huge page size is being used on this host:

    for NODE in worker-0 worker-1; do
    
      # assign 10x 1G huge page on processor/numa-node 0 on worker-node to applications
      system host-memory-modify -f application -1G 10 $NODE 0
    
      # assign 10x 1G huge page on processor/numa-node 1 on worker-node to applications
      system host-memory-modify -f application -1G 10 $NODE 1
    
    done
    
  3. For OpenStack only: Setup disk partition for nova-local volume group, needed for stx-openstack nova ephemeral disks.

    for NODE in worker-0 worker-1; do
       system host-lvg-add ${NODE} nova-local
    
       # Get UUID of DISK to create PARTITION to be added to ‘nova-local’ local volume group
       # CEPH OSD Disks can NOT be used
       # For best performance, do NOT use system/root disk, use a separate physical disk.
    
       # List host’s disks and take note of UUID of disk to be used
       system host-disk-list ${NODE}
       # ( if using ROOT DISK, select disk with device_path of
       #   ‘system host-show ${NODE} | fgrep rootfs’   )
    
       # Create new PARTITION on selected disk, and take note of new partition’s ‘uuid’ in response
       # The size of the PARTITION needs to be large enough to hold the aggregate size of
       # all nova ephemeral disks of all VMs that you want to be able to host on this host,
       # but is limited by the size and space available on the physical disk you chose above.
       # The following example uses a small PARTITION size such that you can fit it on the
       # root disk, if that is what you chose above.
       # Additional PARTITION(s) from additional disks can be added later if required.
       PARTITION_SIZE=30
    
       system hostdisk-partition-add -t lvm_phys_vol ${NODE} <disk-uuid> ${PARTITION_SIZE}
    
       # Add new partition to ‘nova-local’ local volume group
       system host-pv-add ${NODE} nova-local <NEW_PARTITION_UUID>
       sleep 2
    done
    
  4. For OpenStack only: Configure data interfaces for worker nodes. Data class interfaces are vswitch interfaces used by vswitch to provide VM virtio vNIC connectivity to OpenStack Neutron Tenant Networks on the underlying assigned Data Network.

    Important

    A compute-labeled worker host MUST have at least one Data class interface.

    • Configure the data interfaces for worker nodes.

      # Execute the following lines with
      export NODE=worker-0
      # and then repeat with
      export NODE=worker-1
      
         # List inventoried host’s ports and identify ports to be used as ‘data’ interfaces,
         # based on displayed linux port name, pci address and device type.
         system host-port-list ${NODE}
      
         # List host’s auto-configured ‘ethernet’ interfaces,
         # find the interfaces corresponding to the ports identified in previous step, and
         # take note of their UUID
         system host-if-list -a ${NODE}
      
         # Modify configuration for these interfaces
         # Configuring them as ‘data’ class interfaces, MTU of 1500 and named data#
         system host-if-modify -m 1500 -n data0 -c data ${NODE} <data0-if-uuid>
         system host-if-modify -m 1500 -n data1 -c data ${NODE} <data1-if-uuid>
      
         # Create Data Networks that vswitch 'data' interfaces will be connected to
         DATANET0='datanet0'
         DATANET1='datanet1'
         system datanetwork-add ${DATANET0} vlan
         system datanetwork-add ${DATANET1} vlan
      
         # Assign Data Networks to Data Interfaces
         system interface-datanetwork-assign ${NODE} <data0-if-uuid> ${DATANET0}
         system interface-datanetwork-assign ${NODE} <data1-if-uuid> ${DATANET1}
      

Optionally Configure PCI-SRIOV Interfaces

  1. Optionally, configure pci-sriov interfaces for worker nodes.

    This step is optional for Kubernetes. Do this step if using SR-IOV network attachments in hosted application containers.

    This step is optional for OpenStack. Do this step if using SR-IOV vNICs in hosted application VMs. Note that pci-sriov interfaces can have the same Data Networks assigned to them as vswitch data interfaces.

    • Configure the pci-sriov interfaces for worker nodes.

      # Execute the following lines with
      export NODE=worker-0
      # and then repeat with
      export NODE=worker-1
      
         # List inventoried host’s ports and identify ports to be used as ‘pci-sriov’ interfaces,
         # based on displayed linux port name, pci address and device type.
         system host-port-list ${NODE}
      
         # List host’s auto-configured ‘ethernet’ interfaces,
         # find the interfaces corresponding to the ports identified in previous step, and
         # take note of their UUID
         system host-if-list -a ${NODE}
      
         # Modify configuration for these interfaces
         # Configuring them as ‘pci-sriov’ class interfaces, MTU of 1500 and named sriov#
         system host-if-modify -m 1500 -n sriov0 -c pci-sriov ${NODE} <sriov0-if-uuid> -N <num_vfs>
         system host-if-modify -m 1500 -n sriov1 -c pci-sriov ${NODE} <sriov1-if-uuid> -N <num_vfs>
      
         # If not created already, create Data Networks that the 'pci-sriov'
         # interfaces will be connected to
         DATANET0='datanet0'
         DATANET1='datanet1'
         system datanetwork-add ${DATANET0} vlan
         system datanetwork-add ${DATANET1} vlan
      
         # Assign Data Networks to PCI-SRIOV Interfaces
         system interface-datanetwork-assign ${NODE} <sriov0-if-uuid> ${DATANET0}
         system interface-datanetwork-assign ${NODE} <sriov1-if-uuid> ${DATANET1}
      
    • For Kubernetes only: To enable using SR-IOV network attachments for the above interfaces in Kubernetes hosted application containers:

      • Configure the Kubernetes SR-IOV device plugin.

        for NODE in worker-0 worker-1; do
           system host-label-assign $NODE sriovdp=enabled
        done
        
      • If planning on running DPDK in Kubernetes hosted application containers on this host, configure the number of 1G Huge pages required on both NUMA nodes.

        for NODE in worker-0 worker-1; do
        
           # assign 10x 1G huge page on processor/numa-node 0 on worker-node to applications
           system host-memory-modify -f application $NODE 0 -1G 10
        
           # assign 10x 1G huge page on processor/numa-node 1 on worker-node to applications
           system host-memory-modify -f application $NODE 1 -1G 10
        
        done
        

Unlock worker nodes

Unlock worker nodes in order to bring them into service:

for NODE in worker-0 worker-1; do
   system host-unlock $NODE
done

The worker nodes will reboot in order to apply configuration changes and come into service. This can take 5-10 minutes, depending on the performance of the host machine.

Next steps

Your Kubernetes cluster is now up and running.

For instructions on how to access StarlingX Kubernetes see Access StarlingX Kubernetes R7.0.

For instructions on how to install and access StarlingX OpenStack see StarlingX OpenStack.