Host labels

Host auto assignment matches worker pool labels in Satellite clusters to host and zone labels on available hosts.

Keep in mind the following information about the host labels that are used for auto assignment.

Default host labels

When you attach a host to a Satellite location, the host automatically gets labels for cpu, os, and memory (in bytes). You cannot remove these labels. You can include additional host labels, or update the host metadata later. If the host does not include the os label, it is automatically assumed as RHEL7.

Hosts can have more labels than worker pools

For example, a host with cpu, memory, and env host labels can be auto-assigned to a worker pool that has only a cpu host label. Host auto assignment matches the cpu labels. The reverse does not work: if a worker pool has more labels than a host, the host cannot be auto assigned to the worker pool.

Matching is exact

Host labels must equal (=) each other exactly. Even if the host label is a number, no less than (<), greater than (>), or other operators are used for matching.

Example scenario for host auto assignment

Say that you have a Satellite cluster with a default worker pool in us-east-1a and the following host labels.

• cpu=4
• env=prod
• os=RHEL7

Your Satellite location has available (unassigned) hosts with host labels as follows.

• Host A: cpu=4, memory=32, env=prod, zone=us-east-1b os=rhel
• Host B: cpu=4, memory=32, zone=us-east-1a os=RHEL7
• Host C: cpu=4, memory=64, env=prod os=RHEL7
• Host D: cpu=4, memory=64, env=prod os=RHCOS (Red Hat CoreOS)

If you resize the default worker pool to request 3 more worker nodes, only Host C can be automatically assigned, but not Host A or Host B.

• Host A meets the central processing unit (CPU) and env=prod label requests, but can only be assigned in us-east-1b. Because the default worker pool is only in us-east-1a, Host A is not assigned.
• Host B meets the CPU and zone requests. However, the host does not have the env=prod label, and so is not assigned.
• Host C is automatically assigned because it matches the cpu=4 and env=prod host labels, and does not have any zone restrictions. The memory=64 host label is not considered, because the worker pool does not request a memory label.
• Host D meets the CPU, zone, and env=prod label requests, but does not meet the os request of RHEL7, and so is not assigned.

Hosts must be assigned as worker nodes in each zone of the default worker pool in your cluster. If your default worker pool spans multiple zones, ensure that you have hosts with matching labels that can be assigned in each zone.

Automatically assigning hosts

IBM Cloud Satellite automatically assigns hosts to worker pools in clusters that request compute capacity using host labels such as cpu.

Before you begin, make sure that you create a Satellite cluster. If you create a cluster in the CLI, you must specify the --workers option with the number of hosts you want to automatically assign. Then, make sure that you attach hosts to your Satellite location, but do not assign the hosts to any resources.

1. Review the host labels that the worker pools use to request compute capacity. You have several options.

   • Create a worker pool in a Satellite cluster with the host labels that you want to use for auto assignment.
   • Review existing worker pools for their host labels. Note that you cannot update the host labels that a worker pool has. You can review the Host labels by running the ibmcloud oc worker-pool get -c <cluster> --worker-pool <worker_pool> command.
   • Review the host labels that a Satellite-enabled IBM Cloud service cluster uses to request resources from the Satellite-enabled IBM Cloud service instance console.

2. Review the host labels that your available hosts have. Remember that hosts automatically get cpu and memory labels when you attach the host to your Satellite location.

   1. Get the Satellite location name.

      ibmcloud sat location ls
      

   2. List the available (unassigned) hosts in your location, and note the IDs of the hosts that you want to check the labels for.

      ibmcloud sat host ls --location LOCATION_NAME_OR_ID | grep unassigned
      

   3. For each host that you want to check, get the host details and note the Labels in the output.

      ibmcloud sat host get --location LOCATION_NAME_OR_ID --host HOST_NAME_OR_ID
      

      Example output

      ...
      Labels      
      app      webserver   
      cpu      4   
      memory   3874564
      os       RHCOS
      ...
      

3. Update the labels of your available host to match all the labels of the worker pool that you want the host to be available for automatic assignment. You can optionally specify a zone to make sure that the host only gets automatically assigned to that zone.

   The cpu, os, and memory labels on the host are applied automatically and cannot be edited.

   ibmcloud sat host update --location <location_name_or_ID> --host <host_name_or_ID> -hl <key=value> [-hl <key=value>] [--zone <zone1>]
   

Disabling host auto assignment

The following actions disable host auto assignment for a worker pool. Later, you can reenable host auto assignment.

• Manually assign hosts to a worker pool.
• Delete an individual worker node from a worker pool.

Re-enabling host auto assignment

If you disabled host auto assignment, you can re-enable auto assignment.

1. Make sure that you have available hosts with labels that match the host labels of the worker pool.
2. Resize the worker pool to set the requested size per zone, rebalance the worker pool, and enable auto assignment again.

For Satellite clusters, do not use the ibmcloud oc worker-pool rebalance command if you have manually assigned worker nodes to your worker pool. Rebalancing a pool with manually assigned worker nodes removes more than the expected number of worker nodes.