Add support for Ray Operator on GKE (#11145) (#7795)

[upstream:78b6e94c2527e11e0011805e7b61fcd2cd7d427e]

Signed-off-by: Modular Magician <[email protected]>

Author: modular-magician

.changelog/11145.txt

@@ -0,0 +1,3 @@
+```release-note:enhancement
+container: added field `ray_operator_config` for `resource_container_cluster`
+```

google-beta/services/container/resource_container_cluster.go

@@ -87,6 +87,7 @@ var (
 		"addons_config.0.config_connector_config",
 		"addons_config.0.gcs_fuse_csi_driver_config",
 		"addons_config.0.stateful_ha_config",
+		"addons_config.0.ray_operator_config",
 		"addons_config.0.istio_config",
 		"addons_config.0.kalm_config",
 	}
@@ -522,6 +523,52 @@ func ResourceContainerCluster() *schema.Resource {
 								},
 							},
 						},
+						"ray_operator_config": {
+							Type:         schema.TypeList,
+							Optional:     true,
+							Computed:     true,
+							AtLeastOneOf: addonsConfigKeys,
+							MaxItems:     3,
+							Description:  `The status of the Ray Operator addon, which enabled management of Ray AI/ML jobs on GKE. Defaults to disabled; set enabled = true to enable.`,
+							Elem: &schema.Resource{
+								Schema: map[string]*schema.Schema{
+									"enabled": {
+										Type:     schema.TypeBool,
+										Required: true,
+									},
+									"ray_cluster_logging_config": {
+										Type:        schema.TypeList,
+										Optional:    true,
+										Computed:    true,
+										MaxItems:    1,
+										Description: `The status of Ray Logging, which scrapes Ray cluster logs to Cloud Logging. Defaults to disabled; set enabled = true to enable.`,
+										Elem: &schema.Resource{
+											Schema: map[string]*schema.Schema{
+												"enabled": {
+													Type:     schema.TypeBool,
+													Required: true,
+												},
+											},
+										},
+									},
+									"ray_cluster_monitoring_config": {
+										Type:        schema.TypeList,
+										Optional:    true,
+										Computed:    true,
+										MaxItems:    1,
+										Description: `The status of Ray Cluster monitoring, which shows Ray cluster metrics in Cloud Console. Defaults to disabled; set enabled = true to enable.`,
+										Elem: &schema.Resource{
+											Schema: map[string]*schema.Schema{
+												"enabled": {
+													Type:     schema.TypeBool,
+													Required: true,
+												},
+											},
+										},
+									},
+								},
+							},
+						},
 					},
 				},
 			},
@@ -4521,6 +4568,28 @@ func expandClusterAddonsConfig(configured interface{}) *container.AddonsConfig {
 		}
 	}
 
+	if v, ok := config["ray_operator_config"]; ok && len(v.([]interface{})) > 0 {
+		addon := v.([]interface{})[0].(map[string]interface{})
+		ac.RayOperatorConfig = &container.RayOperatorConfig{
+			Enabled:         addon["enabled"].(bool),
+			ForceSendFields: []string{"Enabled"},
+		}
+		if v, ok := addon["ray_cluster_logging_config"]; ok && len(v.([]interface{})) > 0 {
+			loggingConfig := v.([]interface{})[0].(map[string]interface{})
+			ac.RayOperatorConfig.RayClusterLoggingConfig = &container.RayClusterLoggingConfig{
+				Enabled:         loggingConfig["enabled"].(bool),
+				ForceSendFields: []string{"Enabled"},
+			}
+		}
+		if v, ok := addon["ray_cluster_monitoring_config"]; ok && len(v.([]interface{})) > 0 {
+			loggingConfig := v.([]interface{})[0].(map[string]interface{})
+			ac.RayOperatorConfig.RayClusterMonitoringConfig = &container.RayClusterMonitoringConfig{
+				Enabled:         loggingConfig["enabled"].(bool),
+				ForceSendFields: []string{"Enabled"},
+			}
+		}
+	}
+
 	if v, ok := config["istio_config"]; ok && len(v.([]interface{})) > 0 {
 		addon := v.([]interface{})[0].(map[string]interface{})
 		ac.IstioConfig = &container.IstioConfig{
@@ -5705,6 +5774,24 @@ func flattenClusterAddonsConfig(c *container.AddonsConfig) []map[string]interfac
 			},
 		}
 	}
+	if c.RayOperatorConfig != nil {
+		rayConfig := c.RayOperatorConfig
+		result["ray_operator_config"] = []map[string]interface{}{
+			{
+				"enabled": rayConfig.Enabled,
+			},
+		}
+		if rayConfig.RayClusterLoggingConfig != nil {
+			result["ray_operator_config"].([]map[string]any)[0]["ray_cluster_logging_config"] = []map[string]interface{}{{
+				"enabled": rayConfig.RayClusterLoggingConfig.Enabled,
+			}}
+		}
+		if rayConfig.RayClusterMonitoringConfig != nil {
+			result["ray_operator_config"].([]map[string]any)[0]["ray_cluster_monitoring_config"] = []map[string]interface{}{{
+				"enabled": rayConfig.RayClusterMonitoringConfig.Enabled,
+			}}
+		}
+	}
 
 	if c.IstioConfig != nil {
 		result["istio_config"] = []map[string]interface{}{

google-beta/services/container/resource_container_cluster_test.go

@@ -5209,6 +5209,9 @@ resource "google_container_cluster" "primary" {
   initial_node_count = 1
 
   min_master_version = "latest"
+  release_channel {
+    channel = "RAPID"
+  }
 
   workload_identity_config {
     workload_pool = "${data.google_project.project.project_id}.svc.id.goog"
@@ -5248,6 +5251,9 @@ resource "google_container_cluster" "primary" {
     stateful_ha_config {
       enabled = false
     }
+    ray_operator_config {
+      enabled = false
+    }
     istio_config {
       disabled = true
       auth     = "AUTH_MUTUAL_TLS"
@@ -5316,6 +5322,15 @@ resource "google_container_cluster" "primary" {
     stateful_ha_config {
       enabled = true
     }
+    ray_operator_config {
+      enabled = true
+      ray_cluster_logging_config {
+        enabled = true
+      }
+      ray_cluster_monitoring_config {
+        enabled = true
+      }
+    }
     istio_config {
       disabled = false
       auth     = "AUTH_NONE"

website/docs/r/container_cluster.html.markdown

@@ -459,6 +459,20 @@ Fleet configuration for the cluster. Structure is [documented below](#nested_fle
     The status of the Stateful HA addon, which provides automatic configurable failover for stateful applications.
     It is disabled by default for Standard clusters. Set `enabled = true` to enable.
 
+*  `ray_operator_config` - (Optional). The status of the [Ray Operator
+   addon](https://cloud.google.com/kubernetes-engine/docs/add-on/ray-on-gke/concepts/overview).
+   It is disabled by default. Set `enabled = true` to enable. The minimum
+   cluster version to enable Ray is 1.30.0-gke.1747000.
+
+   Ray Operator config has optional subfields
+   `ray_cluster_logging_config.enabled` and
+   `ray_cluster_monitoring_config.enabled` which control Ray Cluster logging
+   and monitoring respectively. See [Collect and view logs and metrics for Ray
+   clusters on
+   GKE](https://cloud.google.com/kubernetes-engine/docs/add-on/ray-on-gke/how-to/collect-view-logs-metrics)
+   for more information.
+
+
 This example `addons_config` disables two addons:
 
 ```hcl