From 8cb8e68a24e2b20711b63b8518633a2bf8a1d76c Mon Sep 17 00:00:00 2001
From: isaaguilar <isa@aguilarstory.com>
Date: Tue, 18 Jul 2023 11:06:03 -0400
Subject: [PATCH] add vcluster health checks

---
 pkg/api/api.go      |  2 ++
 pkg/api/resource.go | 62 +++++++++++++++++++++++++++++++++++++++++++++
 2 files changed, 64 insertions(+)

diff --git a/pkg/api/api.go b/pkg/api/api.go
index e3c262d..16b9f2c 100644
--- a/pkg/api/api.go
+++ b/pkg/api/api.go
@@ -34,6 +34,8 @@ func (h APIHandler) RegisterRoutes() {
 
 	cluster := routes.Group("/cluster")
 	cluster.POST("/", h.AddCluster) // Resource from Add/Update/Delete event
+	cluster.GET("/:cluster_name/health", h.VClusterHealth)
+	cluster.GET("/:cluster_name/tfohealth", h.VClusterTFOHealth)
 	cluster.PUT("/:cluster_name/sync-dependencies", h.SyncEvent)
 	cluster.POST("/:cluster_name/event", h.ResourceEvent) // routes.GET("/cluster-name/:cluster_name", h.GetCluster) // to be removed
 	cluster.PUT("/:cluster_name/event", h.ResourceEvent)
diff --git a/pkg/api/resource.go b/pkg/api/resource.go
index 2cfe05c..563156b 100644
--- a/pkg/api/resource.go
+++ b/pkg/api/resource.go
@@ -229,6 +229,68 @@ func (h APIHandler) getClusterID(clusterName string) uint {
 	return clusters[0].ID
 }
 
+func (h APIHandler) VClusterHealth(c *gin.Context) {
+	clusterName := c.Param("cluster_name")
+	clusterID := h.getClusterID(clusterName)
+	if clusterID == 0 {
+		c.JSON(http.StatusUnprocessableEntity, response(http.StatusUnprocessableEntity, fmt.Sprintf("cluster_name '%s' not found", clusterName), nil))
+		return
+	}
+	config, err := getVclusterConfig(h.clientset, "internal", clusterName)
+	if err != nil {
+		c.JSON(http.StatusUnprocessableEntity, response(http.StatusUnprocessableEntity, err.Error(), nil))
+		return
+	}
+	clientset := kubernetes.NewForConfigOrDie(config)
+	n, err := clientset.CoreV1().Namespaces().List(c, metav1.ListOptions{})
+	if err != nil {
+		c.JSON(http.StatusUnprocessableEntity, response(http.StatusUnprocessableEntity, err.Error(), nil))
+		return
+	}
+	if len(n.Items) == 0 {
+		c.JSON(http.StatusUnprocessableEntity, response(http.StatusUnprocessableEntity, "No namespaces available", nil))
+		return
+	}
+	c.JSON(http.StatusNoContent, nil)
+}
+
+func (h APIHandler) VClusterTFOHealth(c *gin.Context) {
+	clusterName := c.Param("cluster_name")
+	clusterID := h.getClusterID(clusterName)
+	if clusterID == 0 {
+		c.JSON(http.StatusUnprocessableEntity, response(http.StatusUnprocessableEntity, fmt.Sprintf("cluster_name '%s' not found", clusterName), nil))
+		return
+	}
+	config, err := getVclusterConfig(h.clientset, "internal", clusterName)
+	if err != nil {
+		c.JSON(http.StatusUnprocessableEntity, response(http.StatusUnprocessableEntity, err.Error(), nil))
+		return
+	}
+	tfoclientset := tfo.NewForConfigOrDie(config)
+	if _, err = tfoclientset.TfV1beta1().Terraforms("").List(c, metav1.ListOptions{}); err != nil {
+		// tfo client cannot query crds and therefore tfo health is not ready
+		c.JSON(http.StatusUnprocessableEntity, response(http.StatusUnprocessableEntity, err.Error(), nil))
+		return
+	}
+
+	// Following checks if tfo is running. TFO must be installed similar to the bundled packages in the
+	// terraform-operator github repo.
+	clientset := kubernetes.NewForConfigOrDie(config)
+	n, err := clientset.CoreV1().Pods("tf-system").List(c, metav1.ListOptions{
+		LabelSelector: "app=terraform-operator,component=controller",
+		FieldSelector: "status.phase=Running",
+	})
+	if err != nil {
+		c.JSON(http.StatusUnprocessableEntity, response(http.StatusUnprocessableEntity, err.Error(), nil))
+		return
+	}
+	if len(n.Items) == 0 {
+		c.JSON(http.StatusUnprocessableEntity, response(http.StatusUnprocessableEntity, "Terraform operator controller is not running", nil))
+		return
+	}
+	c.JSON(http.StatusNoContent, nil)
+}
+
 // ResourcePoll is a short poll that checks and returns resources created from the tf resource's workflow
 // that have the correct label and annotation value.
 func (h APIHandler) ResourcePoll(c *gin.Context) {