监控修改

Former-commit-id: 6546b9888afc383fc497b7b4990f489125cb4b5d

api/desc/cloud/pcm-cloud.api

@@ -11,10 +11,12 @@ info(
 type ControllerMetricsReq {
     Metrics []string `form:"metrics"`
     ParticipantId int64 `form:"participantId"`
-    Pods string `form:"pods"`
+    Pod string `form:"pod,optional"`
+    WorkloadName string `form:"workloadName,optional"`
     Steps string `form:"steps"`
     Start string `form:"start"`
     End string `form:"end"`
+    Level string `form:"level,optional"`
 }
 
 type (
@@ -29,26 +31,26 @@ type (
         Name string `form:"name"`                 // 名称
     }
 
-     CloudResp {
+    CloudResp {
         Code string `json:"code"`
-        Msg  string `json:"msg"`
+        Msg string `json:"msg"`
         Data string `json:"data"`
     }
 
     TenantInfo {
-        Id          int64     `json:"id"`           // id
+        Id int64 `json:"id"`                     // id
-        TenantName  string    `json:"tenantName"`  // 租户名称
+        TenantName string `json:"tenantName"`    // 租户名称
-        TenantDesc  string    `json:"tenantDesc"`  // 描述信息
+        TenantDesc string `json:"tenantDesc"`    // 描述信息
-        Clusters    string    `json:"clusters"`     // 集群名称，用","分割
+        Clusters string `json:"clusters"`        // 集群名称，用","分割
-        Type        int64     `json:"type"`         // 租户所属(0数算，1超算，2智算）
+        Type int64 `json:"type"`                 // 租户所属(0数算，1超算，2智算）
-        DeletedFlag int64     `json:"deletedFlag"` // 是否删除
+        DeletedFlag int64 `json:"deletedFlag"`   // 是否删除
-        CreatedBy   int64     `json:"createdBy"`   // 创建人
+        CreatedBy int64 `json:"createdBy"`       // 创建人
-        CreatedTime string     `json:"createdTime"` // 创建时间
+        CreatedTime string `json:"createdTime"`  // 创建时间
-        UpdatedBy   int64     `json:"updatedBy"`   // 更新人
+        UpdatedBy int64 `json:"updatedBy"`       // 更新人
-        UpdatedTime string    `json:"updated_time"` // 更新时间
+        UpdatedTime string `json:"updated_time"` // 更新时间
     }
 
-    UpdateTenantReq{
+    UpdateTenantReq {
         Tenants [] TenantInfo `json:"tenants"`
     }
 )

api/internal/logic/cloud/controllermetricslogic.go

@@ -26,9 +26,15 @@ func NewControllerMetricsLogic(ctx context.Context, svcCtx *svc.ServiceContext)
 
 func (l *ControllerMetricsLogic) ControllerMetrics(req *types.ControllerMetricsReq) (resp *types.ControllerMetricsResp, err error) {
 	resp = &types.ControllerMetricsResp{}
-	metrics := l.svcCtx.PromClient[req.ParticipantId].GetNamedMetricsByTime(req.Metrics, req.Start, req.End, 60*time.Minute, tracker.ControllerOption{
+	if len(req.Pod) != 0 {
-		PodsName: req.Pods,
+		resp.Data = l.svcCtx.PromClient[req.ParticipantId].GetNamedMetricsByTime(req.Metrics, req.Start, req.End, 60*time.Minute, tracker.PodOption{
-	})
+			PodName: req.Pod,
-	resp.Data = metrics
+		})
+	} else {
+
+		resp.Data = l.svcCtx.PromClient[req.ParticipantId].GetNamedMetricsByTime(req.Metrics, req.Start, req.End, 60*time.Minute, tracker.ControllerOption{
+			WorkloadName: req.WorkloadName,
+		})
+	}
 	return resp, nil
 }

api/internal/types/types.go

@@ -3410,10 +3410,12 @@ type ShowNodeDetailsResp struct {
 type ControllerMetricsReq struct {
 	Metrics       []string `form:"metrics"`
 	ParticipantId int64    `form:"participantId"`
-	Pods          string   `form:"pods"`
+	Pod           string   `form:"pod,optional"`
+	WorkloadName  string   `form:"workloadName,optional"`
 	Steps         string   `form:"steps"`
 	Start         string   `form:"start"`
 	End           string   `form:"end"`
+	Level         string   `form:"level,optional"`
 }
 
 type RegisterClusterReq struct {

pkg/tracker/promql.go

@@ -56,8 +56,8 @@ var promQLTemplates = map[string]string{
 	"namespace_ingresses_extensions_count": `sum by (namespace) (kube_ingress_labels{namespace!=""} * on (namespace) group_left(workspace) kube_namespace_labels{$1})`,
 	"namespace_s2ibuilder_count":           `sum by (namespace) (s2i_s2ibuilder_created{namespace!=""} * on (namespace) group_left(workspace) kube_namespace_labels{$1})`,
 
-	"controller_cpu_usage_rate":    `round(sum by (owner_name) (sum by (owner_name, pod) (irate(container_cpu_usage_seconds_total{job="kubelet", $1, image!=""}[5m]))/ sum by (owner_name,pod) (kube_pod_container_resource_limits{resource="cpu"}))/count(kube_pod_info{$2}) by (owner_name),0.0001)`,
+	"controller_cpu_usage_rate":    `sum(  node_namespace_pod_container:container_cpu_usage_seconds_total:sum_irate{}* on(namespace,pod)  group_left(workload) namespace_workload_pod:kube_pod_owner:relabel{$1}) by (workload)/sum(  kube_pod_container_resource_limits{job="kube-state-metrics", resource="cpu"}* on(namespace,pod)  group_left(workload) namespace_workload_pod:kube_pod_owner:relabel{ }) by (workload)`,
-	"controller_memory_usage_rate": `round(sum by (owner_name) (sum by (owner_name, pod) (irate(container_memory_usage_bytes{job="kubelet", $1, image!=""}[5m]))/ sum by (owner_name,pod) (kube_pod_container_resource_limits{resource="memory"}))/count(kube_pod_info{$2}) by (owner_name),0.0001)`,
+	"controller_memory_usage_rate": `sum(    container_memory_working_set_bytes{job="kubelet", metrics_path="/metrics/cadvisor",  container!="", image!=""}  * on(namespace,pod)    group_left(workload) namespace_workload_pod:kube_pod_owner:relabel{$1}) by (workload)/sum(  kube_pod_container_resource_limits{job="kube-state-metrics",  resource="memory"}* on(namespace,pod)  group_left(workload) namespace_workload_pod:kube_pod_owner:relabel{ }) by (workload)`,
 	// pod
 	"pod_cpu_usage":              `round(sum by (namespace, pod) (irate(container_cpu_usage_seconds_total{job="kubelet", pod!="", image!=""}[5m])) * on (namespace, pod) group_left(owner_kind,owner_name) kube_pod_owner{$1} * on (namespace, pod) group_left(node) kube_pod_info{$2}, 0.001)`,
 	"pod_cpu_usage_rate":         `sum(node_namespace_pod_container:container_cpu_usage_seconds_total:sum_irate{  $1}) by (pod) / sum(kube_pod_container_resource_limits{ $1,unit="core"}) by (pod)`,
@@ -147,82 +147,17 @@ func makeNamespaceMetricExpr(tmpl string, o QueryOptions) string {
 }
 
 func makeControllerMetricExpr(tmpl string, o QueryOptions) string {
-	var podName string
+	var workload string
 
-	podName = fmt.Sprintf(`pod=~"%s"`, o.PodName)
+	workload = fmt.Sprintf(`workload="%s"`, o.WorkloadName)
-	return strings.NewReplacer("$1", podName, "$2", podName).Replace(tmpl)
+	return strings.NewReplacer("$1", workload).Replace(tmpl)
 }
 
 func makePodMetricExpr(tmpl string, o QueryOptions) string {
-	var podSelector, workloadSelector string
+	var podName string
 
-	// For monitoriong pods of the specific workload
+	podName = fmt.Sprintf(`pod="%s"`, o.PodName)
-	// GET /namespaces/{namespace}/workloads/{kind}/{workload}/pods
+	return strings.NewReplacer("$1", podName).Replace(tmpl)
-
-	// For monitoring pods in the specific namespace
-	// GET /namespaces/{namespace}/workloads/{kind}/{workload}/pods or
-	// GET /namespaces/{namespace}/pods/{pod} or
-	// GET /namespaces/{namespace}/pods
-	if o.Namespace != "" {
-		if o.PodName != "" {
-			podSelector = fmt.Sprintf(`pod="%s", namespace="%s"`, o.PodName, o.Namespace)
-		} else {
-			podSelector = fmt.Sprintf(`pod=~"%s", namespace="%s"`, o.ResourceFilter, o.Namespace)
-		}
-	} else {
-		var namespaces, pods []string
-		if o.NamespacedResourcesFilter != "" {
-			for _, np := range strings.Split(o.NamespacedResourcesFilter, "|") {
-				if nparr := strings.SplitN(np, "/", 2); len(nparr) > 1 {
-					namespaces = append(namespaces, nparr[0])
-					pods = append(pods, nparr[1])
-				} else {
-					pods = append(pods, np)
-				}
-			}
-		}
-		// For monitoring pods on the specific node
-		// GET /nodes/{node}/pods/{pod}
-		// GET /nodes/{node}/pods
-		if o.NodeName != "" {
-			if o.PodName != "" {
-				if nparr := strings.SplitN(o.PodName, "/", 2); len(nparr) > 1 {
-					podSelector = fmt.Sprintf(`namespace="%s",pod="%s", node="%s"`, nparr[0], nparr[1], o.NodeName)
-				} else {
-					podSelector = fmt.Sprintf(`pod="%s", node="%s"`, o.PodName, o.NodeName)
-				}
-			} else {
-				var ps []string
-				ps = append(ps, fmt.Sprintf(`node="%s"`, o.NodeName))
-				if o.ResourceFilter != "" {
-					ps = append(ps, fmt.Sprintf(`pod=~"%s"`, o.ResourceFilter))
-				}
-
-				if len(namespaces) > 0 {
-					ps = append(ps, fmt.Sprintf(`namespace=~"%s"`, strings.Join(namespaces, "|")))
-				}
-				if len(pods) > 0 {
-					ps = append(ps, fmt.Sprintf(`pod=~"%s"`, strings.Join(pods, "|")))
-				}
-				podSelector = strings.Join(ps, ",")
-			}
-		} else {
-			// For monitoring pods in the whole cluster
-			// Get /pods
-			var ps []string
-			if len(namespaces) > 0 {
-				ps = append(ps, fmt.Sprintf(`namespace=~"%s"`, strings.Join(namespaces, "|")))
-			}
-			if len(pods) > 0 {
-				ps = append(ps, fmt.Sprintf(`pod=~"%s"`, strings.Join(pods, "|")))
-			}
-			if len(ps) > 0 {
-				podSelector = strings.Join(ps, ",")
-			}
-		}
-	}
-
-	return strings.NewReplacer("$1", workloadSelector, "$2", podSelector).Replace(tmpl)
 }
 
 func makeContainerMetricExpr(tmpl string, o QueryOptions) string {

pkg/tracker/queryoptions.go

@@ -72,6 +72,7 @@ type QueryOptions struct {
 	WorkspaceName             string
 	Namespace                 string
 	WorkloadKind              string
+	WorkloadName              string
 	OwnerName                 string
 	PodName                   string
 	PodsName                  string
@@ -250,11 +251,10 @@ type PodOption struct {
 }
 
 type ControllerOption struct {
-	PodsName  string
+	Namespace    string
-	Namespace string
+	Kind         string
-	Kind      string
+	WorkloadName string
-	OwnerName string
+	Level        string
-	Level     string
 }
 
 func (po PodOption) Apply(o *QueryOptions) {
@@ -272,8 +272,7 @@ func (co ControllerOption) Apply(o *QueryOptions) {
 	o.Level = LevelController
 	o.Namespace = co.Namespace
 	o.WorkloadKind = co.Kind
-	o.OwnerName = co.OwnerName
+	o.WorkloadName = co.WorkloadName
-	o.PodName = co.PodsName
 
 }