Configuring Watson OpenScale after installation
You can configure Watson™ OpenScale by enabling a horizontal pod autoscaler and by scaling out Watson OpenScale services and ETCD manually.
Enabling a horizontal pod autoscaler
IBM Cloud Pak® for Data supports scaling out during deployment, which means that to enable a horizontal autoscaler for Watson OpenScale, you must enable it when it is installed. It cannot be enabled after installation. For the steps to enable the horizontal pod autoscaler, see Scaling services. The following additional parameters for Horizontal Pod Autoscaler for Watson OpenScale are available:
| Prefix/Suffix | minReplicas | maxReplicas | targetCPUUtilizationPercentage | Description |
|---|---|---|---|---|
| bias.autoscaling | 1 | 1 | 80 | Bias service |
| commonApi.autoscaling | 1 | 1 | 80 | Common API service |
| configuration.autoscaling | 1 | 1 | 80 | Configuration service |
| dashboard.autoscaling | 1 | 1 | 80 | Dashboard service |
| datamart.autoscaling | 1 | 1 | 80 | Datamart service |
| explainability.autoscaling | 1 | 1 | 80 | Explainability service |
| fastpath.autoscaling | 1 | 1 | 80 | Fast Path service |
| feedback.autoscaling | 1 | 1 | 80 | Feedback service |
| mlGatewayDiscovery.autoscaling | 1 | 1 | 80 | ML Gateway Discovery service |
| payloadLogging.autoscaling | 1 | 1 | 80 | Payload Logging service |
| payloadLoggingApi.autoscaling | 1 | 1 | 80 | Payload Logging Api service |
| scheduling.autoscaling | 1 | 1 | 80 | Scheduling service |
Scaling out Watson OpenScale services manually with the kubectl commands
The following Watson OpenScale services and their default replicas are available:
| Deployment Names | Default Replicas | Description |
|---|---|---|
| aiopenscale-ibm-aios-bias | 1 | Bias service |
| aiopenscale-ibm-aios-common-api | 1 | Common API service |
| aiopenscale-ibm-aios-configuration | 1 | Configuration service |
| aiopenscale-ibm-aios-dashboard | 1 | Dashboard service |
| aiopenscale-ibm-aios-datamart | 1 | Datamart service |
| aiopenscale-ibm-aios-explainability | 1 | Explainability service |
| aiopenscale-ibm-aios-fast-path | 1 | Fast Path service |
| aiopenscale-ibm-aios-feedback | 1 | Feedback service |
| aiopenscale-ibm-aios-ml-gateway-discovery | 1 | ML Gateway Discovery service |
| aiopenscale-ibm-aios-payload-logging | 1 | Payload Logging service |
| aiopenscale-ibm-aios-payload-logging-api | 1 | Payload Logging Api service |
| aiopenscale-ibm-aios-scheduling | 1 | Scheduling service |
Scaling out ETCD manually with the kubectl commands
You should not scale out to an even number of ETCD pods. Currently, ETCD has 3 pods. If you are using manual storage provisioning, you must manually create persistent volumes required for the new ETCD pods that you want to scale out.
To scale out, use either the kubectl scale or kubectl patch commands as shown in the following code samples:
kubectl -n <namespace> scale StatefulSets aiopenscale-ibm-aios-etcd --replicas=<number of replicas>or
kubectl -n <namespace> patch StatefulSets aiopenscale-ibm-aios-etcd -p '{"spec":{"replicas":<number of replicas>}}'Example:
kubectl -n aiopenscale scale StatefulSets aiopenscale-ibm-aios-etcd --replicas=5or
kubectl -n aiopenscale patch StatefulSets aiopenscale-ibm-aios-etcd -p '{"spec":{"replicas":5}}'