feat: add server metrics (#273)

* feat: expose metrics path & init analyzer errors metrics This commit add metrics path & the analyzer error metrics in the codebase. The changes have been made across all analyzers and include the addition of a new metric with label values for the analyzer's name, analyzed object's name, and namespace. The metric's value is set to the length of the analyzer objects failures. Signed-off-by: Matthis Holleville <[email protected]> * feat: add metric to cronjob & deployment & netpol Signed-off-by: Matthis Holleville <[email protected]> * feat: expose metric to NodeAnalyzer Signed-off-by: Matthis Holleville <[email protected]> --------- Signed-off-by: Matthis Holleville <[email protected]>
k8sgpt-ai · Apr 15, 2023 · a3becc9 · a3becc9
1 parent 03a95e7
commit a3becc9
Show file tree

Hide file tree

Showing 15 changed files with 122 additions and 17 deletions.
diff --git a/pkg/analyzer/analyzer.go b/pkg/analyzer/analyzer.go
@@ -7,6 +7,15 @@ import (
 	"github.com/fatih/color"
 	"github.com/k8sgpt-ai/k8sgpt/pkg/common"
 	"github.com/k8sgpt-ai/k8sgpt/pkg/integration"
+	"github.com/prometheus/client_golang/prometheus"
+	"github.com/prometheus/client_golang/prometheus/promauto"
+)
+
+var (
+	AnalyzerErrorsMetric = promauto.NewGaugeVec(prometheus.GaugeOpts{
+		Name: "analyzer_errors",
+		Help: "Number of errors detected by analyzer",
+	}, []string{"analyzer_name", "object_name", "namespace"})
 )
 
 var coreAnalyzerMap = map[string]common.IAnalyzer{

diff --git a/pkg/analyzer/cronjob.go b/pkg/analyzer/cronjob.go
@@ -13,6 +13,13 @@ import (
 type CronJobAnalyzer struct{}
 
 func (analyzer CronJobAnalyzer) Analyze(a common.Analyzer) ([]common.Result, error) {
+
+	kind := "CronJob"
+
+	AnalyzerErrorsMetric.DeletePartialMatch(map[string]string{
+		"analyzer_name": kind,
+	})
+
 	var results []common.Result
 
 	cronJobList, err := a.Client.GetClient().BatchV1().CronJobs("").List(a.Context, v1.ListOptions{})
@@ -81,14 +88,16 @@ func (analyzer CronJobAnalyzer) Analyze(a common.Analyzer) ([]common.Result, err
 		}
 
 		if len(failures) > 0 {
-			preAnalysis[cronJob.Name] = common.PreAnalysis{
+			preAnalysis[fmt.Sprintf("%s/%s", cronJob.Namespace, cronJob.Name)] = common.PreAnalysis{
 				FailureDetails: failures,
 			}
+			AnalyzerErrorsMetric.WithLabelValues(kind, cronJob.Name, cronJob.Namespace).Set(float64(len(failures)))
+
 		}
 
 		for key, value := range preAnalysis {
 			currentAnalysis := common.Result{
-				Kind:  "CronJob",
+				Kind:  kind,
 				Name:  key,
 				Error: value.FailureDetails,
 			}

diff --git a/pkg/analyzer/cronjob_test.go b/pkg/analyzer/cronjob_test.go
@@ -121,6 +121,6 @@ func TestCronJobBroken(t *testing.T) {
 	}
 
 	assert.Equal(t, len(analysisResults), 1)
-	assert.Equal(t, analysisResults[0].Name, "example-cronjob")
+	assert.Equal(t, analysisResults[0].Name, "default/example-cronjob")
 	assert.Equal(t, analysisResults[0].Kind, "CronJob")
 }
diff --git a/pkg/analyzer/deployment.go b/pkg/analyzer/deployment.go
@@ -17,6 +17,12 @@ type DeploymentAnalyzer struct {
 // Analyze scans all namespaces for Deployments with misconfigurations
 func (d DeploymentAnalyzer) Analyze(a common.Analyzer) ([]common.Result, error) {
 
+	kind := "Deployment"
+
+	AnalyzerErrorsMetric.DeletePartialMatch(map[string]string{
+		"analyzer_name": kind,
+	})
+
 	deployments, err := a.Client.GetClient().AppsV1().Deployments("").List(context.Background(), v1.ListOptions{})
 	if err != nil {
 		return nil, err
@@ -44,13 +50,14 @@ func (d DeploymentAnalyzer) Analyze(a common.Analyzer) ([]common.Result, error)
 				FailureDetails: failures,
 				Deployment:     deployment,
 			}
+			AnalyzerErrorsMetric.WithLabelValues(kind, deployment.Name, deployment.Namespace).Set(float64(len(failures)))
 		}
 
 	}
 
 	for key, value := range preAnalysis {
 		var currentAnalysis = common.Result{
-			Kind:  "Deployment",
+			Kind:  kind,
 			Name:  key,
 			Error: value.FailureDetails,
 		}

diff --git a/pkg/analyzer/hpa.go b/pkg/analyzer/hpa.go
@@ -12,6 +12,12 @@ type HpaAnalyzer struct{}
 
 func (HpaAnalyzer) Analyze(a common.Analyzer) ([]common.Result, error) {
 
+	kind := "HorizontalPodAutoscaler"
+
+	AnalyzerErrorsMetric.DeletePartialMatch(map[string]string{
+		"analyzer_name": kind,
+	})
+
 	list, err := a.Client.GetClient().AutoscalingV1().HorizontalPodAutoscalers(a.Namespace).List(a.Context, metav1.ListOptions{})
 	if err != nil {
 		return nil, err
@@ -71,13 +77,14 @@ func (HpaAnalyzer) Analyze(a common.Analyzer) ([]common.Result, error) {
 				HorizontalPodAutoscalers: hpa,
 				FailureDetails:           failures,
 			}
+			AnalyzerErrorsMetric.WithLabelValues(kind, hpa.Name, hpa.Namespace).Set(float64(len(failures)))
 		}
 
 	}
 
 	for key, value := range preAnalysis {
 		var currentAnalysis = common.Result{
-			Kind:  "HorizontalPodAutoscaler",
+			Kind:  kind,
 			Name:  key,
 			Error: value.FailureDetails,
 		}

diff --git a/pkg/analyzer/ingress.go b/pkg/analyzer/ingress.go
@@ -12,6 +12,12 @@ type IngressAnalyzer struct{}
 
 func (IngressAnalyzer) Analyze(a common.Analyzer) ([]common.Result, error) {
 
+	kind := "Ingress"
+
+	AnalyzerErrorsMetric.DeletePartialMatch(map[string]string{
+		"analyzer_name": kind,
+	})
+
 	list, err := a.Client.GetClient().NetworkingV1().Ingresses(a.Namespace).List(a.Context, metav1.ListOptions{})
 	if err != nil {
 		return nil, err
@@ -107,13 +113,15 @@ func (IngressAnalyzer) Analyze(a common.Analyzer) ([]common.Result, error) {
 				Ingress:        ing,
 				FailureDetails: failures,
 			}
+			AnalyzerErrorsMetric.WithLabelValues(kind, ing.Name, ing.Namespace).Set(float64(len(failures)))
+
 		}
 
 	}
 
 	for key, value := range preAnalysis {
 		var currentAnalysis = common.Result{
-			Kind:  "Ingress",
+			Kind:  kind,
 			Name:  key,
 			Error: value.FailureDetails,
 		}

diff --git a/pkg/analyzer/netpol.go b/pkg/analyzer/netpol.go
@@ -11,6 +11,13 @@ import (
 type NetworkPolicyAnalyzer struct{}
 
 func (NetworkPolicyAnalyzer) Analyze(a common.Analyzer) ([]common.Result, error) {
+
+	kind := "NetworkPolicy"
+
+	AnalyzerErrorsMetric.DeletePartialMatch(map[string]string{
+		"analyzer_name": kind,
+	})
+
 	// get all network policies in the namespace
 	policies, err := a.Client.GetClient().NetworkingV1().
 		NetworkPolicies(a.Namespace).List(a.Context, metav1.ListOptions{})
@@ -54,16 +61,18 @@ func (NetworkPolicyAnalyzer) Analyze(a common.Analyzer) ([]common.Result, error)
 		}
 
 		if len(failures) > 0 {
-			preAnalysis[policy.Name] = common.PreAnalysis{
+			preAnalysis[fmt.Sprintf("%s/%s", policy.Namespace, policy.Name)] = common.PreAnalysis{
 				FailureDetails: failures,
 				NetworkPolicy:  policy,
 			}
+			AnalyzerErrorsMetric.WithLabelValues(kind, policy.Name, policy.Namespace).Set(float64(len(failures)))
+
 		}
 	}
 
 	for key, value := range preAnalysis {
 		currentAnalysis := common.Result{
-			Kind:  "NetworkPolicy",
+			Kind:  kind,
 			Name:  key,
 			Error: value.FailureDetails,
 		}

diff --git a/pkg/analyzer/node.go b/pkg/analyzer/node.go
@@ -2,6 +2,7 @@ package analyzer
 
 import (
 	"fmt"
+
 	v1 "k8s.io/api/core/v1"
 
 	"github.com/k8sgpt-ai/k8sgpt/pkg/common"
@@ -13,6 +14,12 @@ type NodeAnalyzer struct{}
 
 func (NodeAnalyzer) Analyze(a common.Analyzer) ([]common.Result, error) {
 
+	kind := "Node"
+
+	AnalyzerErrorsMetric.DeletePartialMatch(map[string]string{
+		"analyzer_name": kind,
+	})
+
 	list, err := a.Client.GetClient().CoreV1().Nodes().List(a.Context, metav1.ListOptions{})
 	if err != nil {
 		return nil, err
@@ -42,12 +49,14 @@ func (NodeAnalyzer) Analyze(a common.Analyzer) ([]common.Result, error) {
 				Node:           node,
 				FailureDetails: failures,
 			}
+			AnalyzerErrorsMetric.WithLabelValues(kind, node.Name, "").Set(float64(len(failures)))
+
 		}
 	}
 
 	for key, value := range preAnalysis {
 		var currentAnalysis = common.Result{
-			Kind:  "Node",
+			Kind:  kind,
 			Name:  key,
 			Error: value.FailureDetails,
 		}

diff --git a/pkg/analyzer/pdb.go b/pkg/analyzer/pdb.go
@@ -12,6 +12,12 @@ type PdbAnalyzer struct{}
 
 func (PdbAnalyzer) Analyze(a common.Analyzer) ([]common.Result, error) {
 
+	kind := "PodDisruptionBudget"
+
+	AnalyzerErrorsMetric.DeletePartialMatch(map[string]string{
+		"analyzer_name": kind,
+	})
+
 	list, err := a.Client.GetClient().PolicyV1().PodDisruptionBudgets(a.Namespace).List(a.Context, metav1.ListOptions{})
 	if err != nil {
 		return nil, err
@@ -63,12 +69,13 @@ func (PdbAnalyzer) Analyze(a common.Analyzer) ([]common.Result, error) {
 				PodDisruptionBudget: pdb,
 				FailureDetails:      failures,
 			}
+			AnalyzerErrorsMetric.WithLabelValues(kind, pdb.Name, pdb.Namespace).Set(float64(len(failures)))
 		}
 	}
 
 	for key, value := range preAnalysis {
 		var currentAnalysis = common.Result{
-			Kind:  "PodDisruptionBudget",
+			Kind:  kind,
 			Name:  key,
 			Error: value.FailureDetails,
 		}

diff --git a/pkg/analyzer/pod.go b/pkg/analyzer/pod.go
@@ -12,6 +12,13 @@ type PodAnalyzer struct {
 }
 
 func (PodAnalyzer) Analyze(a common.Analyzer) ([]common.Result, error) {
+
+	kind := "Pod"
+
+	AnalyzerErrorsMetric.DeletePartialMatch(map[string]string{
+		"analyzer_name": kind,
+	})
+
 	// search all namespaces for pods that are not running
 	list, err := a.Client.GetClient().CoreV1().Pods(a.Namespace).List(a.Context, metav1.ListOptions{})
 	if err != nil {
@@ -70,12 +77,13 @@ func (PodAnalyzer) Analyze(a common.Analyzer) ([]common.Result, error) {
 				Pod:            pod,
 				FailureDetails: failures,
 			}
+			AnalyzerErrorsMetric.WithLabelValues(kind, pod.Name, pod.Namespace).Set(float64(len(failures)))
 		}
 	}
 
 	for key, value := range preAnalysis {
 		var currentAnalysis = common.Result{
-			Kind:  "Pod",
+			Kind:  kind,
 			Name:  key,
 			Error: value.FailureDetails,
 		}

diff --git a/pkg/analyzer/pvc.go b/pkg/analyzer/pvc.go
@@ -12,6 +12,12 @@ type PvcAnalyzer struct{}
 
 func (PvcAnalyzer) Analyze(a common.Analyzer) ([]common.Result, error) {
 
+	kind := "PersistentVolumeClaim"
+
+	AnalyzerErrorsMetric.DeletePartialMatch(map[string]string{
+		"analyzer_name": kind,
+	})
+
 	// search all namespaces for pods that are not running
 	list, err := a.Client.GetClient().CoreV1().PersistentVolumeClaims(a.Namespace).List(a.Context, metav1.ListOptions{})
 	if err != nil {
@@ -43,12 +49,13 @@ func (PvcAnalyzer) Analyze(a common.Analyzer) ([]common.Result, error) {
 				PersistentVolumeClaim: pvc,
 				FailureDetails:        failures,
 			}
+			AnalyzerErrorsMetric.WithLabelValues(kind, pvc.Name, pvc.Namespace).Set(float64(len(failures)))
 		}
 	}
 
 	for key, value := range preAnalysis {
 		var currentAnalysis = common.Result{
-			Kind:  "PersistentVolumeClaim",
+			Kind:  kind,
 			Name:  key,
 			Error: value.FailureDetails,
 		}

diff --git a/pkg/analyzer/rs.go b/pkg/analyzer/rs.go
@@ -12,6 +12,12 @@ type ReplicaSetAnalyzer struct{}
 
 func (ReplicaSetAnalyzer) Analyze(a common.Analyzer) ([]common.Result, error) {
 
+	kind := "ReplicaSet"
+
+	AnalyzerErrorsMetric.DeletePartialMatch(map[string]string{
+		"analyzer_name": kind,
+	})
+
 	// search all namespaces for pods that are not running
 	list, err := a.Client.GetClient().AppsV1().ReplicaSets(a.Namespace).List(a.Context, metav1.ListOptions{})
 	if err != nil {
@@ -42,12 +48,13 @@ func (ReplicaSetAnalyzer) Analyze(a common.Analyzer) ([]common.Result, error) {
 				ReplicaSet:     rs,
 				FailureDetails: failures,
 			}
+			AnalyzerErrorsMetric.WithLabelValues(kind, rs.Name, rs.Namespace).Set(float64(len(failures)))
 		}
 	}
 
 	for key, value := range preAnalysis {
 		var currentAnalysis = common.Result{
-			Kind:  "ReplicaSet",
+			Kind:  kind,
 			Name:  key,
 			Error: value.FailureDetails,
 		}

diff --git a/pkg/analyzer/service.go b/pkg/analyzer/service.go
@@ -13,6 +13,12 @@ type ServiceAnalyzer struct{}
 
 func (ServiceAnalyzer) Analyze(a common.Analyzer) ([]common.Result, error) {
 
+	kind := "Service"
+
+	AnalyzerErrorsMetric.DeletePartialMatch(map[string]string{
+		"analyzer_name": kind,
+	})
+
 	// search all namespaces for pods that are not running
 	list, err := a.Client.GetClient().CoreV1().Endpoints(a.Namespace).List(a.Context, metav1.ListOptions{})
 	if err != nil {
@@ -71,12 +77,13 @@ func (ServiceAnalyzer) Analyze(a common.Analyzer) ([]common.Result, error) {
 				Endpoint:       ep,
 				FailureDetails: failures,
 			}
+			AnalyzerErrorsMetric.WithLabelValues(kind, ep.Name, ep.Namespace).Set(float64(len(failures)))
 		}
 	}
 
 	for key, value := range preAnalysis {
 		var currentAnalysis = common.Result{
-			Kind:  "Service",
+			Kind:  kind,
 			Name:  key,
 			Error: value.FailureDetails,
 		}

diff --git a/pkg/analyzer/statefulset.go b/pkg/analyzer/statefulset.go
@@ -11,6 +11,13 @@ import (
 type StatefulSetAnalyzer struct{}
 
 func (StatefulSetAnalyzer) Analyze(a common.Analyzer) ([]common.Result, error) {
+
+	kind := "StatefulSet"
+
+	AnalyzerErrorsMetric.DeletePartialMatch(map[string]string{
+		"analyzer_name": kind,
+	})
+
 	list, err := a.Client.GetClient().AppsV1().StatefulSets(a.Namespace).List(a.Context, metav1.ListOptions{})
 	if err != nil {
 		return nil, err
@@ -61,12 +68,13 @@ func (StatefulSetAnalyzer) Analyze(a common.Analyzer) ([]common.Result, error) {
 				StatefulSet:    sts,
 				FailureDetails: failures,
 			}
+			AnalyzerErrorsMetric.WithLabelValues(kind, sts.Name, sts.Namespace).Set(float64(len(failures)))
 		}
 	}
 
 	for key, value := range preAnalysis {
 		var currentAnalysis = common.Result{
-			Kind:  "StatefulSet",
+			Kind:  kind,
 			Name:  key,
 			Error: value.FailureDetails,
 		}