From 4b842ce4458b484d2870573f7c8d63a11df98f15 Mon Sep 17 00:00:00 2001
From: Yu Yang <yuyang0@users.noreply.github.com>
Date: Wed, 18 Oct 2023 10:55:53 +0800
Subject: [PATCH] better metrics for node status (#620)

---
 metrics/handler.go  | 13 +------------
 metrics/metrics.go  | 45 +++++++++++++++++----------------------------
 utils/utils.go      |  5 +++++
 utils/utils_test.go |  5 +++++
 4 files changed, 28 insertions(+), 40 deletions(-)

diff --git a/metrics/handler.go b/metrics/handler.go
index f00ad7acd..cca790f70 100644
--- a/metrics/handler.go
+++ b/metrics/handler.go
@@ -20,16 +20,8 @@ func (m *Metrics) ResourceMiddleware(cluster cluster.Cluster) func(http.Handler)
 			if err != nil {
 				logger.Error(ctx, err, "Get all nodes err")
 			}
-			podUpDownNodes := map[string][]int{}
 			for node := range nodes {
-				if podUpDownNodes[node.Podname] == nil {
-					podUpDownNodes[node.Podname] = []int{0, 0}
-				}
-				if node.IsDown() {
-					podUpDownNodes[node.Podname][1]++
-				} else {
-					podUpDownNodes[node.Podname][0]++
-				}
+				m.SendPodNodeStatus(ctx, node)
 				metrics, err := m.rmgr.GetNodeMetrics(ctx, node)
 				if err != nil {
 					logger.Error(ctx, err, "Get metrics failed")
@@ -37,9 +29,6 @@ func (m *Metrics) ResourceMiddleware(cluster cluster.Cluster) func(http.Handler)
 				}
 				m.SendMetrics(ctx, metrics...)
 			}
-			for podname, ud := range podUpDownNodes {
-				m.SendPodUpDownNodes(ctx, podname, ud[0], ud[1])
-			}
 
 			h.ServeHTTP(w, r)
 		})
diff --git a/metrics/metrics.go b/metrics/metrics.go
index 6b10d5122..f30bc393f 100644
--- a/metrics/metrics.go
+++ b/metrics/metrics.go
@@ -2,6 +2,7 @@ package metrics
 
 import (
 	"context"
+	"fmt"
 	"os"
 	"strconv"
 	"sync"
@@ -21,12 +22,10 @@ import (
 )
 
 const (
-	deployCountKey  = "core.%s.deploy.count"
-	deployCountName = "core_deploy"
-	upNodesKey      = "core.up.nodes"
-	upNodesName     = "core_up_nodes"
-	downNodesKey    = "core.down.nodes"
-	downNodesName   = "core_down_nodes"
+	deployCountKey    = "core.%s.deploy.count"
+	deployCountName   = "core_deploy"
+	podNodeStatusKey  = "pod.node.%s.up"
+	podNodeStatusName = "pod_node_up"
 
 	gaugeType   = "gauge"
 	counterType = "counter"
@@ -58,22 +57,17 @@ func (m *Metrics) SendDeployCount(ctx context.Context, n int) {
 	m.SendMetrics(ctx, metrics)
 }
 
-func (m *Metrics) SendPodUpDownNodes(ctx context.Context, podname string, up, down int) {
-	log.WithFunc("metrics.SendUpDownNodes").Info(ctx, "Update deploy counter")
-	m1 := &plugintypes.Metrics{
-		Name:   upNodesName,
-		Labels: []string{m.Hostname, podname},
-		Key:    upNodesKey,
-		Value:  strconv.Itoa(up),
+func (m *Metrics) SendPodNodeStatus(ctx context.Context, node *types.Node) {
+	log.WithFunc("metrics.SendPodNodeStatus").Info(ctx, "update pod node status")
+	up := !node.IsDown()
+	metrics := &plugintypes.Metrics{
+		Name:   podNodeStatusName,
+		Labels: []string{m.Hostname, node.Podname, node.Name},
+		Key:    fmt.Sprintf(podNodeStatusKey, node.Name),
+		Value:  strconv.Itoa(utils.Bool2Int(up)),
 	}
 
-	m2 := &plugintypes.Metrics{
-		Name:   downNodesName,
-		Labels: []string{m.Hostname, podname},
-		Key:    downNodesKey,
-		Value:  strconv.Itoa(down),
-	}
-	m.SendMetrics(ctx, m1, m2)
+	m.SendMetrics(ctx, metrics)
 }
 
 // SendMetrics update metrics
@@ -231,15 +225,10 @@ func InitMetrics(ctx context.Context, config types.Config, metricsDescriptions [
 		Help: "core deploy counter",
 	}, []string{"hostname"})
 
-	Client.Collectors[upNodesName] = prometheus.NewGaugeVec(prometheus.GaugeOpts{
-		Name: upNodesName,
+	Client.Collectors[podNodeStatusName] = prometheus.NewGaugeVec(prometheus.GaugeOpts{
+		Name: podNodeStatusName,
 		Help: "number of up nodes",
-	}, []string{"hostname", "podname"})
-
-	Client.Collectors[downNodesName] = prometheus.NewGaugeVec(prometheus.GaugeOpts{
-		Name: downNodesName,
-		Help: "number of down nodes",
-	}, []string{"hostname", "podname"})
+	}, []string{"hostname", "podname", "nodename"})
 
 	once.Do(func() {
 		prometheus.MustRegister(maps.Values(Client.Collectors)...)
diff --git a/utils/utils.go b/utils/utils.go
index 4b7ebc518..523df850c 100644
--- a/utils/utils.go
+++ b/utils/utils.go
@@ -14,6 +14,7 @@ import (
 	"os"
 	"strings"
 	"time"
+	"unsafe"
 
 	"github.com/cockroachdb/errors"
 	"github.com/projecteru2/core/cluster"
@@ -270,3 +271,7 @@ func safeSplit(s string) []string {
 
 	return result
 }
+
+func Bool2Int(a bool) int {
+	return *(*int)(unsafe.Pointer(&a)) & 1
+}
diff --git a/utils/utils_test.go b/utils/utils_test.go
index 0ed53734a..b8e7b1f62 100644
--- a/utils/utils_test.go
+++ b/utils/utils_test.go
@@ -221,3 +221,8 @@ func TestSHA256(t *testing.T) {
 	str := "hhh"
 	assert.Equal(t, "24d166cd6c8b826c779040b49d5b6708d649b236558e8744339dfee6afe11999", SHA256(str))
 }
+
+func TestBool2Int(t *testing.T) {
+	assert.Equal(t, 1, Bool2Int(true))
+	assert.Equal(t, 0, Bool2Int(false))
+}