Merge pull request #100573 from openshift-cherrypick-robot/cherry-pick-100332-to-enterprise-4.20

[enterprise-4.20] OSDOCS 16489 Add GPU usage with CMA docs -- NEEDED for 4.20!

Author: mburke5678

modules/nodes-cma-autoscaling-custom-trigger-prom-gpu.adoc

@@ -0,0 +1,40 @@
+// Module included in the following assemblies:
+//
+// * nodes/cma/nodes-cma-autoscaling-custom-trigger.adoc
+
+:_mod-docs-content-type: CONCEPT
+[id="nodes-cma-autoscaling-custom-trigger-prom-gpu_{context}"]
+= Configuring GPU-based autoscaling with Prometheus and DCGM metrics
+
+You can use the Custom Metrics Autoscaler with NVIDIA Data Center GPU Manager (DCGM) metrics to scale workloads based on GPU utilization. This is particularly useful for AI and machine learning workloads that require GPU resources.
+
+.Example scaled object with a Prometheus target for GPU-based autoscaling
+[source,yaml,options="nowrap"]
+----
+apiVersion: keda.sh/v1alpha1
+kind: ScaledObject
+metadata:
+  name: gpu-scaledobject
+  namespace: my-namespace
+spec:
+  scaleTargetRef:
+    kind: Deployment
+    name: gpu-deployment
+  minReplicaCount: 1 <1>
+  maxReplicaCount: 5 <2>
+  triggers:
+  - type: prometheus
+    metadata:
+      serverAddress: https://thanos-querier.openshift-monitoring.svc.cluster.local:9092
+      namespace: my-namespace
+      metricName: gpu_utilization
+      threshold: '90' <3>
+      query: SUM(DCGM_FI_DEV_GPU_UTIL{instance=~".+", gpu=~".+"}) <4>
+      authModes: bearer
+    authenticationRef:
+      name: keda-trigger-auth-prometheus
+----
+<1> Specifies the minimum number of replicas to maintain. For GPU workloads, this should not be set to `0` to ensure that metrics continue to be collected.
+<2> Specifies the maximum number of replicas allowed during scale-up operations.
+<3> Specifies the GPU utilization percentage threshold that triggers scaling. When the average GPU utilization exceeds 90%, the autoscaler scales up the deployment.
+<4> Specifies a Prometheus query using NVIDIA DCGM metrics to monitor GPU utilization across all GPU devices. The `DCGM_FI_DEV_GPU_UTIL` metric provides GPU utilization percentages.

nodes/cma/nodes-cma-autoscaling-custom-trigger.adoc

@@ -22,6 +22,7 @@ You can configure a certificate authority xref:../../nodes/cma/nodes-cma-autosca
 // assemblies.
 
 include::modules/nodes-cma-autoscaling-custom-trigger-prom.adoc[leveloffset=+1]
+include::modules/nodes-cma-autoscaling-custom-trigger-prom-gpu.adoc[leveloffset=+2]
 include::modules/nodes-cma-autoscaling-custom-prometheus-config.adoc[leveloffset=+2]
 
 [role="_additional-resources"]