Add support for GPU exporters.

vantage-sh · May 30, 2024 · 392babe · 392babe
1 parent d81b300
commit 392babe
Show file tree

Hide file tree

Showing 5 changed files with 64 additions and 2 deletions.
diff --git a/charts/vantage-kubernetes-agent/Chart.yaml b/charts/vantage-kubernetes-agent/Chart.yaml
@@ -2,6 +2,6 @@ apiVersion: v2
 name: vantage-kubernetes-agent
 description: Provisions the Vantage Kubernetes agent.
 type: application
-version: 1.0.33
-appVersion: "1.0.25"
+version: 1.0.34
+appVersion: "1.0.26"
 icon: "https://assets.vantage.sh/www/vantage_avatar-social.jpg"
diff --git a/charts/vantage-kubernetes-agent/templates/application.yaml b/charts/vantage-kubernetes-agent/templates/application.yaml
@@ -93,6 +93,26 @@ spec:
             - name: VANTAGE_REPORT_HTTP_PROXY
               value: "{{ .Values.agent.reportHTTPProxy }}"
             {{- end}}
+            {{- if .Values.agent.gpu.usageMetrics}}
+            - name: VANTAGE_GPU_METRICS
+              value: "true"
+            {{- end}}
+            {{- if .Values.agent.gpu.exporterNamespace}}
+            - name: VANTAGE_GPU_EXPORTER_NAMESPACE
+              value: "{{ .Values.agent.gpu.exporterNamespace }}"
+            {{- end}}
+            {{- if .Values.agent.gpu.exporterServiceName}}
+            - name: VANTAGE_GPU_EXPORTER_SERVICE_NAME
+              value: "{{ .Values.agent.gpu.exporterServiceName }}"
+            {{- end}}
+            {{- if .Values.agent.gpu.exporterPortName}}
+            - name: VANTAGE_GPU_EXPORTER_PORT_NAME
+              value: "{{ .Values.agent.gpu.exporterPortName }}"
+            {{- end}}
+            {{- if .Values.agent.gpu.exporterPath}}
+            - name: VANTAGE_GPU_EXPORTER_PATH
+              value: "{{ .Values.agent.gpu.exporterPath }}"
+            {{- end}}
             - name: VANTAGE_API_TOKEN
               valueFrom:
                 secretKeyRef:

diff --git a/charts/vantage-kubernetes-agent/templates/clusterrole.yaml b/charts/vantage-kubernetes-agent/templates/clusterrole.yaml
@@ -33,6 +33,12 @@ rules:
   - "jobs"
   - "cronjobs"
   verbs: ["get", "watch", "list"]
+{{- if .Values.agent.gpu.usageMetrics}}
+- apiGroups: [""]
+  resources:
+  - "endpoints"
+  verbs: ["get", "watch", "list"]
+{{- end}}
 {{- if .Values.agent.argocdRollouts}}
 - apiGroups: ["argoproj.io"]
   resources:

diff --git a/charts/vantage-kubernetes-agent/values.schema.json b/charts/vantage-kubernetes-agent/values.schema.json
@@ -26,6 +26,26 @@
                 "disableKubeTLSverify": {
                     "type": "string"
                 },
+                "gpu": {
+                    "type": "object",
+                    "properties": {
+                        "exporterNamespace": {
+                            "type": "string"
+                        },
+                        "exporterPath": {
+                            "type": "string"
+                        },
+                        "exporterPortName": {
+                            "type": "string"
+                        },
+                        "exporterServiceName": {
+                            "type": "string"
+                        },
+                        "usageMetrics": {
+                            "type": "boolean"
+                        }
+                    }
+                },
                 "logLevel": {
                     "type": "string"
                 },

diff --git a/charts/vantage-kubernetes-agent/values.yaml b/charts/vantage-kubernetes-agent/values.yaml
@@ -44,6 +44,22 @@ agent:
   # Optional. URL of an HTTP proxy used for external communications during the periodic report. This includes HTTP requests to the Vantage API and AWS S3. ie. http://example.com:3001
   reportHTTPProxy: ""
 
+  gpu:
+    # Optional. Whether the agent should scrape GPU metrics or not. Requires NVIDIA dcgm-exporter to be deployed and configured with DCGM_FI_DEV_FB_USED and DCGM_FI_DEV_FB_TOTAL
+    usageMetrics: false
+    # Optional. The namespace the GPU exporter is deployed in.
+    # Uses agent default if not specified: "gpu-operator"
+    exporterNamespace: ""
+    # Optional. The service name for the GPU exporter. Used to lookup the endpoints.
+    # Uses agent default if not specified: "nvidia-dcgm-exporter"
+    exporterServiceName: ""
+    # Optional. The port name for the metrics port of the gpu exporter.
+    # Uses agent default if not specified: "gpu-metrics"
+    exporterPortName: ""
+    # Optional. The path the metrics endpoint is available under.
+    # Uses agent default if not specified: "/metrics"
+    exporterPath: ""
+
 
 persist:
   mountPath: "/var/lib/vantage-agent"