refactor of remove & clean invalid nodes in cache & metrics

cluster/calcium/node.go

@@ -90,11 +90,11 @@ func (c *Calcium) RemoveNode(ctx context.Context, nodename string) error {
 			},
 			// then: remove node resource metadata
 			func(ctx context.Context) error {
-				if err = c.rmgr.RemoveNode(ctx, nodename); err != nil {
+				if err := c.rmgr.RemoveNode(ctx, nodename); err != nil {
 					return err
 				}
 				enginefactory.RemoveEngineFromCache(ctx, node.Endpoint, node.Ca, node.Cert, node.Key)
-				metrics.Client.RemoveInvalidNodes([]string{nodename})
+				metrics.Client.RemoveInvalidNodes(nodename)
 				return nil
 			},
 			// rollback: do nothing

go.mod

@@ -26,6 +26,7 @@ require (
 	github.com/pkg/errors v0.9.1
 	github.com/projecteru2/libyavirt v0.0.0-20230921032447-a617cf0c746c
 	github.com/prometheus/client_golang v1.15.0
+	github.com/prometheus/client_model v0.3.0
 	github.com/rs/zerolog v1.29.1
 	github.com/sanity-io/litter v1.5.5
 	github.com/stretchr/testify v1.8.2
@@ -105,7 +106,6 @@ require (
 	github.com/opencontainers/runc v1.1.6 // indirect
 	github.com/pjbgf/sha1cd v0.3.0 // indirect
 	github.com/pmezard/go-difflib v1.0.0 // indirect
-	github.com/prometheus/client_model v0.3.0 // indirect
 	github.com/prometheus/common v0.42.0 // indirect
 	github.com/prometheus/procfs v0.9.0 // indirect
 	github.com/rogpeppe/go-internal v1.10.0 // indirect

metrics/handler.go

@@ -5,8 +5,12 @@ import (
 	"net/http"
 
 	"github.com/projecteru2/core/cluster"
+	enginefactory "github.com/projecteru2/core/engine/factory"
 	"github.com/projecteru2/core/log"
 	"github.com/projecteru2/core/types"
+	"github.com/prometheus/client_golang/prometheus"
+	promClient "github.com/prometheus/client_model/go"
+	"golang.org/x/exp/slices"
 )
 
 // ResourceMiddleware to make sure update resource correct
@@ -20,7 +24,8 @@ func (m *Metrics) ResourceMiddleware(cluster cluster.Cluster) func(http.Handler)
 			if err != nil {
 				logger.Error(ctx, err, "Get all nodes err")
 			}
-			activeNodes := make(map[string]*types.Node, 0)
+			existNodenames := getExistNodenames()
+			activeNodes := map[string]*types.Node{}
 			for node := range nodes {
 				metrics, err := m.rmgr.GetNodeMetrics(ctx, node)
 				if err != nil {
@@ -30,8 +35,31 @@ func (m *Metrics) ResourceMiddleware(cluster cluster.Cluster) func(http.Handler)
 				activeNodes[node.Name] = node
 				m.SendMetrics(ctx, metrics...)
 			}
-			m.DeleteInactiveNodesWithCache(ctx, activeNodes)
+			// refresh nodes
+			invalidNodenames := []string{}
+			for _, nodename := range existNodenames {
+				if node := activeNodes[nodename]; node != nil {
+					invalidNodenames = append(invalidNodenames, nodename)
+					enginefactory.RemoveEngineFromCache(ctx, node.Endpoint, node.Ca, node.Cert, node.Key)
+				}
+			}
+			m.RemoveInvalidNodes(invalidNodenames...)
 			h.ServeHTTP(w, r)
 		})
 	}
 }
+
+func getExistNodenames() []string {
+	metrics, _ := prometheus.DefaultGatherer.Gather()
+	nodenames := []string{}
+	for _, metric := range metrics {
+		for _, mf := range metric.GetMetric() {
+			if i := slices.IndexFunc(mf.Label, func(label *promClient.LabelPair) bool {
+				return label.GetName() == "nodename"
+			}); i != -1 {
+				nodenames = append(nodenames, mf.Label[i].GetValue())
+			}
+		}
+	}
+	return nodenames
+}

metrics/metrics.go

@@ -6,14 +6,13 @@ import (
 	"strconv"
 	"sync"
 
-	enginefactory "github.com/projecteru2/core/engine/factory"
 	"github.com/projecteru2/core/log"
 	"github.com/projecteru2/core/resource"
 	"github.com/projecteru2/core/resource/cobalt"
 	plugintypes "github.com/projecteru2/core/resource/plugins/types"
 	"github.com/projecteru2/core/types"
 	"github.com/projecteru2/core/utils"
-	io_prometheus_client "github.com/prometheus/client_model/go"
+	promClient "github.com/prometheus/client_model/go"
 	"golang.org/x/exp/slices"
 
 	statsdlib "github.com/CMGS/statsd"
@@ -88,55 +87,16 @@ func (m *Metrics) SendMetrics(ctx context.Context, metrics ...*plugintypes.Metri
 	}
 }
 
-func (m *Metrics) DeleteInactiveNodesWithCache(ctx context.Context, activeNodesMap map[string]*types.Node) {
-	metricNodeNameMap := m.getNodeNameMapFromMetrics()
-	// 计算差集
-	invalidNodes := make([]string, 0)
-	for nodeName := range metricNodeNameMap {
-		if node, exists := activeNodesMap[nodeName]; !exists {
-			invalidNodes = append(invalidNodes, nodeName)
-			enginefactory.RemoveEngineFromCache(ctx, node.Endpoint, node.Ca, node.Cert, node.Key)
-		}
-	}
-	m.RemoveInvalidNodes(invalidNodes)
-}
-
-func (m *Metrics) getNodeNameMapFromMetrics() map[string]bool {
-	metrics, _ := prometheus.DefaultGatherer.Gather()
-	nodeNameMap := make(map[string]bool, 0)
-	for _, metric := range metrics {
-		for _, mf := range metric.GetMetric() {
-			if len(mf.Label) == 0 {
-				continue
-			}
-			for _, label := range mf.Label {
-				if label.GetName() == "nodename" {
-					nodeNameMap[label.GetValue()] = true
-					break
-				}
-			}
-		}
-	}
-	return nodeNameMap
-}
-
 // RemoveInvalidNodes 清除多余的metric标签值
-func (m *Metrics) RemoveInvalidNodes(invalidNodes []string) {
+func (m *Metrics) RemoveInvalidNodes(invalidNodes ...string) {
 	if len(invalidNodes) == 0 {
 		return
 	}
 	for _, collector := range m.Collectors {
-		if collector == nil {
-			return
-		}
 		metrics, _ := prometheus.DefaultGatherer.Gather()
 		for _, metric := range metrics {
 			for _, mf := range metric.GetMetric() {
-				if len(mf.Label) == 0 {
-					continue
-				}
-
-				if !slices.ContainsFunc(mf.Label, func(label *io_prometheus_client.LabelPair) bool {
+				if !slices.ContainsFunc(mf.Label, func(label *promClient.LabelPair) bool {
 					return label.GetName() == "nodename" && slices.ContainsFunc(invalidNodes, func(nodename string) bool {
 						return label.GetValue() == nodename
 					})