Extend Capacity with Worker Nodes

This section describes the steps to extend capacity with worker nodes on a StarlingX All-in-one Duplex deployment configuration.

Install software on worker nodes

  1. Power on the worker node servers and force them to network boot with the appropriate BIOS boot options for your particular server.

  2. As the worker nodes boot, a message appears on their console instructing you to configure the personality of the node.

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

    system host-list
+----+--------------+-------------+----------------+-------------+--------------+
| id | hostname     | personality | administrative | operational | availability |
+----+--------------+-------------+----------------+-------------+--------------+
| 1  | controller-0 | controller  | unlocked       | enabled     | available    |
| 2  | controller-1 | controller  | unlocked       | enabled     | available    |
| 3  | None         | None        | locked         | disabled    | offline      |
| 4  | None         | None        | locked         | disabled    | offline      |
+----+--------------+-------------+----------------+-------------+--------------+

  4. Using the host id, set the personality of this host to ‘worker’:

    system host-update 3 personality=worker hostname=worker-0
system host-update 4 personality=worker hostname=worker-1

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

    Note

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

  5. Wait for the install of software on the worker nodes to complete, for the worker nodes to reboot, and for both 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  | unlocked       | enabled     | available    |
| 3  | worker-0     | worker      | locked         | disabled    | online       |
| 4  | worker-1     | worker      | locked         | disabled    | online       |
+----+--------------+-------------+----------------+-------------+--------------+

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 (|prefix|-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 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, assuming 1G huge page size is being used on this host, with the following commands:

    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 host-disk-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 already created, 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 to apply configuration changes and come into service. This can take 5-10 minutes, depending on the performance of the host machine.