Merge branch 'master' of https://gitlink.org.cn/JointCloud/pcm-coordinator

Former-commit-id: e31d516e8d87623c8471fb45c383834cb6ad21b3

api/desc/monitoring/pcm-monitoring.api

@@ -11,6 +11,12 @@ type CreateAlertRuleReq {
     AlertType string `json:"alertType"`
 }
 
+type DeleteAlertRuleReq {
+    Id int64 `form:"id"`
+    ClusterName string `form:"clusterName"`
+    Name string `form:"name"`
+}
+
 type (
     AlertRulesReq {
         AlertType string `form:"alertType"`
@@ -104,5 +110,4 @@ Link{
     Category {
         name string `json:"name"`
     }
-
 )

api/desc/pcm.api

@@ -1031,6 +1031,9 @@ service pcm {
 	@handler alertRulesHandler
 	get /monitoring/alert/rule (AlertRulesReq) returns (AlertRulesResp)
 
+	@handler DeleteAlertRuleHandler
+	delete /cloud/alert/rule (DeleteAlertRuleReq)
+
 	@doc "cluster resource load"
 	@handler clustersLoadHandler
 	get /monitoring/cluster/load (clustersLoadReq) returns (clustersLoadResp)

api/internal/handler/monitoring/deletealertrulehandler.go

@@ -0,0 +1,25 @@
+package monitoring
+
+import (
+	"gitlink.org.cn/JointCloud/pcm-coordinator/pkg/repository/result"
+	"net/http"
+
+	"github.com/zeromicro/go-zero/rest/httpx"
+	"gitlink.org.cn/JointCloud/pcm-coordinator/api/internal/logic/monitoring"
+	"gitlink.org.cn/JointCloud/pcm-coordinator/api/internal/svc"
+	"gitlink.org.cn/JointCloud/pcm-coordinator/api/internal/types"
+)
+
+func DeleteAlertRuleHandler(svcCtx *svc.ServiceContext) http.HandlerFunc {
+	return func(w http.ResponseWriter, r *http.Request) {
+		var req types.DeleteAlertRuleReq
+		if err := httpx.Parse(r, &req); err != nil {
+			httpx.ErrorCtx(r.Context(), w, err)
+			return
+		}
+
+		l := monitoring.NewDeleteAlertRuleLogic(r.Context(), svcCtx)
+		err := l.DeleteAlertRule(&req)
+		result.HttpResult(r, w, nil, err)
+	}
+}

api/internal/handler/routes.go

@@ -1302,6 +1302,11 @@ func RegisterHandlers(server *rest.Server, serverCtx *svc.ServiceContext) {
 				Path:    "/monitoring/alert/rule",
 				Handler: monitoring.AlertRulesHandler(serverCtx),
 			},
+			{
+				Method:  http.MethodDelete,
+				Path:    "/cloud/alert/rule",
+				Handler: monitoring.DeleteAlertRuleHandler(serverCtx),
+			},
 			{
 				Method:  http.MethodGet,
 				Path:    "/monitoring/cluster/load",

api/internal/logic/monitoring/createalertrulelogic.go

@@ -46,7 +46,7 @@ func (l *CreateAlertRuleLogic) CreateAlertRule(req *types.CreateAlertRuleReq) er
 		return tx.Error
 	}
 
-	// query server http url.
+	// query cluster http url.
 	var server string
 	l.svcCtx.DbEngin.Raw("select ta.server from t_adapter ta,t_cluster tc where ta.id = tc.adapter_id and tc.name = ?", &req.ClusterName).Scan(&server)
 

api/internal/logic/monitoring/deletealertrulelogic.go

@@ -0,0 +1,66 @@
+package monitoring
+
+import (
+	"context"
+
+	"gitlink.org.cn/JointCloud/pcm-coordinator/api/internal/svc"
+	"gitlink.org.cn/JointCloud/pcm-coordinator/api/internal/types"
+
+	"github.com/zeromicro/go-zero/core/logx"
+)
+
+type DeleteAlertRuleLogic struct {
+	logx.Logger
+	ctx    context.Context
+	svcCtx *svc.ServiceContext
+}
+
+func NewDeleteAlertRuleLogic(ctx context.Context, svcCtx *svc.ServiceContext) *DeleteAlertRuleLogic {
+	return &DeleteAlertRuleLogic{
+		Logger: logx.WithContext(ctx),
+		ctx:    ctx,
+		svcCtx: svcCtx,
+	}
+}
+
+func (l *DeleteAlertRuleLogic) DeleteAlertRule(req *types.DeleteAlertRuleReq) error {
+	// Delete data from the database
+	l.svcCtx.DbEngin.Delete(&types.AlertRule{}, "id = ?", req.Id)
+
+	// query cluster http url.
+	var server string
+	l.svcCtx.DbEngin.Raw("select ta.server from t_adapter ta,t_cluster tc where ta.id = tc.adapter_id and tc.name = ?", &req.ClusterName).Scan(&server)
+
+	// create prometheus rule
+	response, err := l.svcCtx.HttpClient.R().
+		SetBody(&CrdStruct{
+			ClusterName: req.ClusterName,
+			Name:        req.Name,
+			Grv: Grv{
+				Group:    "monitoring.coreos.com",
+				Version:  "v1",
+				Resource: "prometheusrules",
+			},
+		}).
+		ForceContentType("application/json").
+		Delete(server + "/api/v1/crd")
+	if err != nil {
+		return err
+	}
+	if err != nil || response.IsError() {
+		return err
+	}
+	return nil
+}
+
+type Grv struct {
+	Group    string `json:"group"`
+	Version  string `json:"version"`
+	Resource string `json:"resource"`
+}
+
+type CrdStruct struct {
+	ClusterName string `json:"clusterName"`
+	Grv         Grv    `json:"grv"`
+	Name        string `json:"name"`
+}

api/internal/logic/monitoring/schedulesituationlogic.go

@@ -33,28 +33,28 @@ func (l *ScheduleSituationLogic) ScheduleSituation() (resp *types.ScheduleSituat
 
 	// hpc
 	var hpcLinks []string
-	tx = l.svcCtx.DbEngin.Raw("SELECT  GROUP_CONCAT(cluster_id SEPARATOR ',') as cluster_ids FROM task_hpc WHERE deleted_at IS NULL GROUP BY task_id HAVING COUNT(*) > 1;").Scan(&hpcLinks)
+	tx = l.svcCtx.DbEngin.Raw("SELECT distinct GROUP_CONCAT( distinct cluster_id SEPARATOR ',') as cluster_ids FROM task_hpc WHERE deleted_at IS NULL GROUP BY task_id HAVING COUNT(*) > 1;").Scan(&hpcLinks)
 	if tx.Error != nil {
 		return nil, tx.Error
 	}
 	LinksHandler(hpcLinks, resp)
 	// cloud
 	var cloudLinks []string
-	tx = l.svcCtx.DbEngin.Raw("SELECT  GROUP_CONCAT(cluster_id SEPARATOR ',') as cluster_ids FROM task_cloud  WHERE deleted_at IS NULL GROUP BY task_id HAVING COUNT(*) > 1;").Scan(&cloudLinks)
+	tx = l.svcCtx.DbEngin.Raw("SELECT distinct  GROUP_CONCAT(distinct cluster_id SEPARATOR ',') as cluster_ids FROM task_cloud  WHERE deleted_at IS NULL GROUP BY task_id HAVING COUNT(*) > 1;").Scan(&cloudLinks)
 	if tx.Error != nil {
 		return nil, tx.Error
 	}
 	LinksHandler(cloudLinks, resp)
 	// ai
 	var aiLinks []string
-	tx = l.svcCtx.DbEngin.Raw("SELECT  GROUP_CONCAT(cluster_id SEPARATOR ',') as cluster_ids FROM task_ai WHERE deleted_at IS NULL GROUP BY task_id HAVING COUNT(*) > 1;").Scan(&aiLinks)
+	tx = l.svcCtx.DbEngin.Raw("SELECT distinct GROUP_CONCAT(distinct cluster_id SEPARATOR ',') as cluster_ids FROM task_ai WHERE deleted_at IS NULL GROUP BY task_id HAVING COUNT(*) > 1;").Scan(&aiLinks)
 	if tx.Error != nil {
 		return nil, tx.Error
 	}
 	LinksHandler(aiLinks, resp)
 	// vm
 	var vmLinks []string
-	tx = l.svcCtx.DbEngin.Raw("SELECT  GROUP_CONCAT(cluster_id SEPARATOR ',') as cluster_ids FROM task_vm WHERE deleted_at IS NULL GROUP BY task_id HAVING COUNT(*) > 1;").Scan(&vmLinks)
+	tx = l.svcCtx.DbEngin.Raw("SELECT distinct GROUP_CONCAT(distinct cluster_id SEPARATOR ',') as cluster_ids FROM task_vm WHERE deleted_at IS NULL GROUP BY task_id HAVING COUNT(*) > 1;").Scan(&vmLinks)
 	if tx.Error != nil {
 		return nil, tx.Error
 	}

api/internal/types/types.go

@@ -1164,7 +1164,7 @@ type CommitHpcTaskReq struct {
 	Description string            `json:"description,optional"`
 	TenantId    int64             `json:"tenantId,optional"`
 	TaskId      int64             `json:"taskId,optional"`
-	AdapterIds  []string          `json:"adapterIds"`
+	AdapterIds  []string          `json:"adapterId"`
 	MatchLabels map[string]string `json:"matchLabels,optional"`
 	CardCount   int64             `json:"cardCount,optional"`
 	WorkDir     string            `json:"workDir,optional"` //paratera:workingDir
@@ -5750,6 +5750,12 @@ type CreateAlertRuleReq struct {
 	AlertType   string `json:"alertType"`
 }
 
+type DeleteAlertRuleReq struct {
+	Id          int64  `form:"id"`
+	ClusterName string `form:"clusterName"`
+	Name        string `form:"name"`
+}
+
 type AlertRulesReq struct {
 	AlertType string `form:"alertType"`
 	AdapterId string `form:"adapterId,optional"`

pkg/tracker/tracker.go

@@ -78,6 +78,18 @@ var (
 		Name: "cluster_pod_total",
 		Help: "Cluster Pod total.",
 	}, []string{"cluster_name", "adapter_id"})
+	ClusterCpuCoreHoursGauge = prometheus.NewGaugeVec(prometheus.GaugeOpts{
+		Name: "cluster_cpu_core_hours",
+		Help: "Cluster Cpu Core Hours.",
+	}, []string{"cluster_name", "adapter_id"})
+	ClusterCardsAvailGauge = prometheus.NewGaugeVec(prometheus.GaugeOpts{
+		Name: "cluster_cards_avail",
+		Help: "Cluster Cards Available.",
+	}, []string{"cluster_name", "adapter_id"})
+	ClusterGpuAvailGauge = prometheus.NewGaugeVec(prometheus.GaugeOpts{
+		Name: "cluster_gpu_avail",
+		Help: "Cluster Gpu Available.",
+	}, []string{"cluster_name", "adapter_id"})
 
 	metrics = []prometheus.Collector{
 		ClusterCpuUtilisationGauge,