Merge pull request #43 from nebius/feature/dcgm-service-monitor

Allow enabling DCGM ServiceMonitor
nebius · Oct 22, 2024 · 6253b42 · 6253b42
2 parents beb828a + 6e556ec
commit 6253b42
Show file tree

Hide file tree

Showing 3 changed files with 10 additions and 3 deletions.
diff --git a/modules/gpu-operator/helm.tf → modules/gpu-operator/app.tf b/modules/gpu-operator/helm.tf → modules/gpu-operator/app.tf
@@ -4,9 +4,10 @@ resource "nebius_applications_v1alpha1_k8s_release" "gpu-operator" {
 
   application_name = "gpu-operator"
   namespace        = "gpu-operator"
-  product_slug     = "nebius/nvidia-gpu-operator"
+  product_slug     = var.product_slug
 
   set = {
-    "driver.version" : var.driver_version
+    "driver.version" : var.driver_version,
+    "dcgmExporter.serviceMonitor.enabled" : var.enable_dcgm_service_monitor
   }
 }
diff --git a/modules/gpu-operator/variables.tf b/modules/gpu-operator/variables.tf
@@ -15,7 +15,13 @@ variable "product_slug" {
 }
 
 variable "driver_version" {
-  description = "Enable Node Feature Discovery."
+  description = "GPU driver version."
   type        = string
   default     = "550.54.15"
 }
+
+variable "enable_dcgm_service_monitor" {
+  description = "Whether to enable DCGM service monitor."
+  type        = bool
+  default     = false
+}
diff --git a/modules/network-operator/helm.tf → modules/network-operator/app.tf b/modules/network-operator/helm.tf → modules/network-operator/app.tf