alert list

Former-commit-id: 454efc36db

api/desc/monitoring/pcm-monitoring.api

@@ -1,7 +1,7 @@
 syntax = "v1"
 
 type CreateAlertRuleReq {
-    CLusterId int64 `json:"clusterId"`
+    CLusterId string `json:"clusterId"`
     ClusterName string `json:"clusterName"`
     Name string `json:"name"`
     PromQL string `json:"promQL"`
@@ -46,8 +46,14 @@ type (
 type (
     alertListReq {
         alertType string `form:"alertType"`
+        adapterId string `form:"adapterId,optional"`
+        clusterId string `form:"clusterId,optional"`
     }
     alertListResp {
         alertMap map[string]interface{} `json:"alertMap"`
     }
-)
+)
+
+type SyncClusterAlertReq {
+    AlertRecordsMap map[string]interface{} `json:"alertRecordsMap"`
+}

api/desc/pcm.api

@@ -969,4 +969,8 @@ service pcm {
 	@doc "alert list"
 	@handler alertListHandler
 	get /monitoring/alert/list (alertListReq) returns (alertListResp)
+
+	@doc "Synchronize Cluster alert Information"
+	@handler syncClusterAlertHandler
+	post /core/syncClusterAlert (SyncClusterAlertReq)
 }

api/internal/handler/monitoring/syncclusteralerthandler.go

@@ -0,0 +1,28 @@
+package monitoring
+
+import (
+	"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 SyncClusterAlertHandler(svcCtx *svc.ServiceContext) http.HandlerFunc {
+	return func(w http.ResponseWriter, r *http.Request) {
+		var req types.SyncClusterAlertReq
+		if err := httpx.Parse(r, &req); err != nil {
+			httpx.ErrorCtx(r.Context(), w, err)
+			return
+		}
+
+		l := monitoring.NewSyncClusterAlertLogic(r.Context(), svcCtx)
+		err := l.SyncClusterAlert(&req)
+		if err != nil {
+			httpx.ErrorCtx(r.Context(), w, err)
+		} else {
+			httpx.Ok(w)
+		}
+	}
+}

api/internal/handler/routes.go

@@ -1217,6 +1217,11 @@ func RegisterHandlers(server *rest.Server, serverCtx *svc.ServiceContext) {
 				Path:    "/monitoring/alert/list",
 				Handler: monitoring.AlertListHandler(serverCtx),
 			},
+			{
+				Method:  http.MethodPost,
+				Path:    "/core/syncClusterAlert",
+				Handler: monitoring.SyncClusterAlertHandler(serverCtx),
+			},
 		},
 		rest.WithPrefix("/pcm/v1"),
 	)

api/internal/logic/monitoring/alertlistlogic.go

@@ -2,9 +2,9 @@ package monitoring
 
 import (
 	"context"
-	"github.com/pkg/errors"
+	"fmt"
 	v1 "github.com/prometheus/client_golang/api/prometheus/v1"
-	tool "gitlink.org.cn/JointCloud/pcm-coordinator/pkg/utils"
+	"k8s.io/apimachinery/pkg/util/json"
 
 	"gitlink.org.cn/JointCloud/pcm-coordinator/api/internal/svc"
 	"gitlink.org.cn/JointCloud/pcm-coordinator/api/internal/types"
@@ -33,31 +33,27 @@ type AlertListResp struct {
 }
 
 func (l *AlertListLogic) AlertList(req *types.AlertListReq) (resp *types.AlertListResp, err error) {
-	// todo: add your logic here and delete this line
-	resp = &types.AlertListResp{}
+	resp = &types.AlertListResp{
+		AlertMap: make(map[string]interface{}),
+	}
 
 	// query server http url.
-	var serverArray []string
-	l.svcCtx.DbEngin.Raw("select ta.server from t_adapter ta,t_cluster tc where ta.id = tc.adapter_id and label = 'kubernetes' and ta.type = ?", req.AlertType).Scan(&serverArray)
-
-	result := make(map[string][]*v1.Alert)
-	for _, server := range serverArray {
-		alertListResp := AlertListResp{}
-		response, err := l.svcCtx.HttpClient.R().
-			SetResult(&alertListResp).
-			ForceContentType("application/json").
-			Get(server + "/api/v1/alert/rule/list")
-		if err != nil {
-			logx.Error(response)
-			return nil, err
-		}
-		if response.IsError() {
-			return nil, errors.New(response.String())
-		}
-		for k, v := range alertListResp.Data {
-			result[k] = v
-		}
+	var clusterArray []string
+
+	sql := "select distinct tc.name from t_adapter ta,t_cluster tc where ta.id = tc.adapter_id and label = 'kubernetes' and ta.type = ? and ta.id = ?"
+	if len(req.AdapterId) > 0 {
+		sql = fmt.Sprintf("select distinct tc.name from t_adapter ta,t_cluster tc where ta.id = tc.adapter_id and label = 'kubernetes' and ta.type = ? and ta.id = %s", req.AdapterId)
+	}
+	if len(req.ClusterId) > 0 {
+		sql = fmt.Sprintf("select distinct tc.name from t_adapter ta,t_cluster tc where ta.id = tc.adapter_id and label = 'kubernetes' and ta.type = ? and tc.id = %s", req.ClusterId)
+	}
+	l.svcCtx.DbEngin.Raw(sql, req.AlertType).Scan(&clusterArray)
+
+	for _, clusterName := range clusterArray {
+		getResult := l.svcCtx.RedisClient.Get(l.ctx, clusterName)
+		var alerts []v1.Alert
+		json.Unmarshal([]byte(getResult.Val()), &alerts)
+		resp.AlertMap[clusterName] = alerts
 	}
-	tool.Convert(result, &resp.AlertMap)
 	return resp, nil
 }

api/internal/logic/monitoring/clustersloadlogic.go

@@ -27,7 +27,7 @@ func NewClustersLoadLogic(ctx context.Context, svcCtx *svc.ServiceContext) *Clus
 
 func (l *ClustersLoadLogic) ClustersLoad(req *types.ClustersLoadReq) (resp *types.ClustersLoadResp, err error) {
 	resp = &types.ClustersLoadResp{}
-	metrics := []string{"cluster_cpu_utilisation", "cluster_cpu_avail", "cluster_cpu_total", "cluster_memory_total", "cluster_memory_avail", "cluster_memory_utilisation", "cluster_disk_utilisation", "cluster_disk_avail", "cluster_disk_total"}
+	metrics := []string{"cluster_cpu_utilisation", "cluster_cpu_avail", "cluster_cpu_total", "cluster_memory_total", "cluster_memory_avail", "cluster_memory_utilisation", "cluster_disk_utilisation", "cluster_disk_avail", "cluster_disk_total", "cluster_pod_utilisation"}
 	result := l.svcCtx.PromClient.GetNamedMetrics(metrics, time.Now(), tracker.ClusterOption{ClusterName: req.ClusterName})
 	resp.Data = result
 	return resp, nil

api/internal/logic/monitoring/syncclusteralertlogic.go

@@ -0,0 +1,44 @@
+package monitoring
+
+import (
+	"context"
+	"gitlink.org.cn/JointCloud/pcm-coordinator/api/internal/svc"
+	"gitlink.org.cn/JointCloud/pcm-coordinator/api/internal/types"
+	"k8s.io/apimachinery/pkg/util/json"
+	"time"
+
+	"github.com/zeromicro/go-zero/core/logx"
+)
+
+type SyncClusterAlertLogic struct {
+	logx.Logger
+	ctx    context.Context
+	svcCtx *svc.ServiceContext
+}
+
+func NewSyncClusterAlertLogic(ctx context.Context, svcCtx *svc.ServiceContext) *SyncClusterAlertLogic {
+	return &SyncClusterAlertLogic{
+		Logger: logx.WithContext(ctx),
+		ctx:    ctx,
+		svcCtx: svcCtx,
+	}
+}
+
+func (l *SyncClusterAlertLogic) SyncClusterAlert(req *types.SyncClusterAlertReq) error {
+
+	if len(req.AlertRecordsMap) != 0 {
+		for k, v := range req.AlertRecordsMap {
+			bytes, err := json.Marshal(v)
+			if err != nil {
+				return err
+			}
+			setCmd := l.svcCtx.RedisClient.Set(l.ctx, k, bytes, 1*time.Minute)
+			if setCmd.Err() != nil {
+				logx.Error(setCmd.Err())
+			}
+
+		}
+	}
+
+	return nil
+}

api/internal/svc/servicecontext.go

@@ -72,6 +72,7 @@ func NewServiceContext(c config.Config) *ServiceContext {
 		panic("InitSnowflake err")
 	}
 	httpClient := resty.New()
+	httpClient.SetTimeout(1 * time.Second)
 	alertClient := tracker.NewAlertClient(c.Monitoring.AlertUrl)
 	if err != nil {
 		logx.Errorf("InitPrometheus err: %v", err)

api/internal/types/types.go

@@ -5453,7 +5453,7 @@ type PushResourceInfoReq struct {
 }
 
 type CreateAlertRuleReq struct {
-	CLusterId   int64  `json:"clusterId"`
+	CLusterId   string `json:"clusterId"`
 	ClusterName string `json:"clusterName"`
 	Name        string `json:"name"`
 	PromQL      string `json:"promQL"`
@@ -5495,8 +5495,14 @@ type NodesLoadTopResp struct {
 
 type AlertListReq struct {
 	AlertType string `form:"alertType"`
+	AdapterId string `form:"adapterId,optional"`
+	ClusterId string `form:"clusterId,optional"`
 }
 
 type AlertListResp struct {
 	AlertMap map[string]interface{} `json:"alertMap"`
 }
+
+type SyncClusterAlertReq struct {
+	AlertRecordsMap map[string]interface{} `json:"alertRecordsMap"`
+}

pkg/tracker/promql.go

@@ -30,6 +30,7 @@ var promQLTemplates = map[string]string{
 	"cluster_cpu_avail":          "cluster_cpu_avail{$1}",
 	"cluster_memory_avail":       "cluster_memory_avail{$1}",
 	"cluster_disk_avail":         "cluster_disk_avail{$1}",
+	"cluster_pod_utilisation":    "cluster_pod_utilisation{$1}",
 
 	// center
 	"center_cpu_utilisation":    "(sum by (adapter_id)(cluster_cpu_total{$1})-sum by (adapter_id)(cluster_cpu_avail{$1}))/sum by (adapter_id)(cluster_cpu_total{$1})",