Installing the Spyre Operator

Edit online

The Spyre Operator helps your cluster run AI workloads more efficiently. It improves scheduling, manages resources, works with Spyre operators to optimize spyre use, supports AI pipelines, and integrates with IBM AI services.

Search for "IBM Spyre Operator" from the OpenShift Container Platform Web Console, navigate to Operators and select OperatorHub.
Select the Spyre Operator.
Choose the channel and version for the IBM Spyre Operator, and click Install
In the Installed Namespace from the menu, select spyre-operator.
Set the installation mode to all namespaces or a specific namespace.
Set the Installed namespace to spyre-operator or a new namespace.
Set Update Approval to Manual or Automatic.
Click Install.
Wait for the "Installed Operator: ready for use" message and select the View Operator button.

Select the Spyre Cluster Policy tab and then select Create SpyreClusterPolicy.

In the YAML view, enter the policy details:

apiVersion: spyre.ibm.com/v1alpha1
kind: SpyreClusterPolicy
metadata:
  name: spyreclusterpolicy
spec:
  loglevel: info
  devicePlugin:
    repository: "quay.io/ibm-aiu"
    image: "spyre-device-plugin"
    version: 1.3.0
    configPath: /etc/aiu
    configName: senlib_config.json
    initContainer:
      repository: "quay.io/ibm-aiu"
      image: "spyre-device-plugin-init"
      version: 1.3.0
      executePolicy: IfNotPresent
  scheduler:
    repository: "quay.io/ibm-aiu"
    image: "spyre-scheduler"
    version: 1.3.0
  podValidator:
    enabled: true
    repository: "quay.io/ibm-aiu"
    image: "spyre-webhook-validator"
    version: 1.3.0
  cardManagement:
    enabled: false
  metricsExporter:
    enabled: false
    repository: "quay.io/ibm-aiu"
    image: "spyre-exporter"
    version: 1.3.0
  healthChecker:
    enabled: true
    repository: "quay.io/ibm-aiu"
    image: "spyre-health-checker"
    version: 1.3.0

Note: For OCP cluster versions 4.19.10 and later (including 4.20+), use the Spyre Scheduler image version used is 1.2.0.

Field	Type	Description	Default
Device plug-in Configuration
spec.devicePlugin.configName	string	Configuration file name inside the container	senlib_config.json
spec.devicePlugin.configPath	string	Configuration file path inside the container	/etc/aiu
spec.devicePlugin.env	[] object	Environment variables for the device plug-in container	-
spec.devicePlugin.topologyConfigMap	string	Name of ConfigMap containing topology configuration	-
Device plug-in Init Container
spec.devicePlugin.initContainer.executePolicy	enum	When to run the init container: Always or IfNotPresent	IfNotPresent
Pod Validator Configuration
spec.podValidator.enabled	boolean	Enable or disable the pod validation webhook	true
Other Configuration
spec.cardManagement	object	Card management component configuration	Not Supported for s390x
spec.experimentalMode	[]enum	Experimental features to enable. Options: `perDeviceAllocation` (specify devices by PCI address), `topologyAwareAllocation` (policy-based allocation), `externalDeviceReservation` (use scheduler for reservation)	Not Supported for s390x
spec.loglevel	enum	Operator log level: debug, info, or error	info
spec.metricsExporter	object	Metrics exporter component configuration	Not supported for s390x
spec.skipUpdateComponents	[]string	Components skip updating if already deployed. Options: commonInit, devicePlugin, cardManagement, metricsExporter, scheduler, podValidator	-

Note:

Only one SpyreClusterPolicy resource should exist per cluster.
Device plug-in pod is scheduled on all pods where spyre cards are attached.

Once the SpyreClusterPolicy changes are done, click Create.
Check if the SpyreClusterPolicy shows a status state of ready.
Run the following command to check the status of the SpyreNodeState resource for your worker node:
```
oc describe spyrenodestate <worker-node-name> 
```

The output displays details for each attached Spyre card, including health and PCI addresses.

Name:         ocpz-standard-compute-1
Namespace:    
Labels:       <none>
Annotations:  <none>
API Version:  spyre.ibm.com/v1alpha1
Kind:         SpyreNodeState
Metadata:
  Creation Timestamp:  2026-03-02T09:10:22Z
  Generation:          2
  Owner References:
    API Version:           spyre.ibm.com/v1alpha1
    Block Owner Deletion:  true
    Controller:            true
    Kind:                  SpyreClusterPolicy
    Name:                  spyreclusterpolicy
    UID:                   14ed828d-4405-4820-83b5-3634aef63bf2
  Resource Version:        1598150
  UID:                     536c0233-51e8-48ad-8151-a1dd283620cf
Spec:
  Node Name:  ocpz-standard-compute-1
  Spyre SSA Interfaces:
    Health:       healthy
    Pci Address:  0003:00:00.0
    Health:       healthy
    Pci Address:  0001:00:00.0
    Health:       healthy
    Pci Address:  0002:00:00.0
Status:
  Conditions:
    Last Transition Time:  2026-03-02T09:10:24Z
    Message:               All devices are healthy.
    Reason:                AllDevicesHealthy
    Status:                True
    Type:                  DeviceHealthy
Events:                    <none>

Note:

Note: Create only one SpyreClusterPolicy in the entire Red Hat OpenShift cluster.
Only one SpyreClusterPolicy resource should exist per cluster.

Deploying the Spyre Operator manually through CLI

Note: This is an optional step to install the Spyre Operator in case the RHOCP OperatorHub is not used for installation. The Spyre Operator is a part of certified operators; you have to enable certified- operators, before installation.

Create the spyre-operator namespace by running the following command:
```
oc create ns spyre-operator
```

Create a YAML file of kind OperatorGroup, that is similar to the following example:

apiVersion: operators.coreos.com/v1
kind: OperatorGroup
metadata:
  name: spyre-operator-group
  namespace: spyre-operator

Apply the YAML file by running the following command:
```
oc apply -f operator-group.yaml
```

Create a YAML file of kind Subscription, that is similar to the following example:

apiVersion: operators.coreos.com/v1alpha1
kind: Subscription
metadata:
  name: spyre-operator
  namespace: spyre-operator
spec:
  channel: "stable"
  name: spyre-operator
  source: ibm-spyre-operators
  sourceNamespace: openshift-marketplace

Apply the YAML file by running the following command:
```
oc apply -f subscription.yaml
```

Updating the Spyre operator from 1.1.0 to 1.2.1

Select the Subscription Update Channel, click Save.
Verify if the the CatalogSourse and csv are updated.
Recreate the SpyreClusterPolicy to ensure the upgraded cluster is governed correctly.

Updating the Spyre Operator manually through CLI from 1.1.0 to 1.1.1

Edit the YAML to upgrade Operator by changing the subscription
```
oc edit subscription spyre-operator -n spyre-operator
```
Note: To update the operator to 1.1.1 edit channel "stable" to "stable-v1.1"