feat(backend): optimize dbha's alert info close #13269

Author: xjxia

dbm-services/common/dbha/ha-module/agent/monitor_agent.go

@@ -216,7 +216,8 @@ func (a *MonitorAgent) RefreshGMCache() {
 	}
 
 	// delete not exists or expired ip from gm cache
-	log.Logger.Debugf("try to refresh gm cache instances, cache number:%d", len(a.ReportGMCache))
+	log.Logger.Debugf("[%s] try to refresh gm cache instances, cache number:%d",
+		a.DetectType, len(a.ReportGMCache))
 	now := time.Now()
 	for ip, cacheIns := range a.ReportGMCache {
 		if now.After(cacheIns.ReporterGMTime.Add(time.Second * 600)) {

dbm-services/common/dbha/ha-module/dbmodule/dbmysql/SpiderProxyLayer_switch.go

@@ -82,7 +82,8 @@ func (ins *SpiderProxyLayerSwitch) ElectPrimaryCandidate() (*TdbctlInfo, error)
 			continue
 		}
 		if ins.Ip == spider.IP && ins.Port == spider.Port {
-			ins.ReportLogs(constvar.InfoResult, fmt.Sprintf("skip self node[%s#%d]]", spider.IP, spider.Port))
+			ins.ReportLogs(constvar.InfoResult, fmt.Sprintf("skip itself [%s#%d] in dbmeta",
+				spider.IP, spider.Port))
 			continue
 		}
 
@@ -117,6 +118,11 @@ func (ins *SpiderProxyLayerSwitch) ElectPrimaryCandidate() (*TdbctlInfo, error)
 
 	ins.ReportLogs(constvar.InfoResult, "try to elect an appropriate node as primary")
 	for _, node := range nodes {
+		if node.Host == ins.Ip && ins.AdminPort == node.Port {
+			ins.ReportLogs(constvar.InfoResult, fmt.Sprintf("skip itself [%s#%d] in route table",
+				node.Host, node.Port))
+			continue
+		}
 		//1. clusterRole must be Secondary
 		//should not happen
 		if strings.EqualFold(node.ClusterRole, PrimaryRole) {

dbm-services/common/dbha/ha-module/dbutil/db_detect.go

@@ -356,6 +356,9 @@ func (b *BaseDetectDB) GetDBType() types.DBType {
 
 // GetDBRole return
 func (b *BaseDetectDB) GetDBRole() string {
+	if b.DBRole == "" {
+		return "N/A"
+	}
 	return b.DBRole
 }
 

dbm-services/common/dbha/ha-module/gm/connection.go

@@ -180,8 +180,8 @@ func (conn *AgentConnection) parse(readLen int) error {
 
 				// unpack success
 				// replay ok
-				log.Logger.Infof("process net package success. Type:%s, Body:%s",
-					conn.netPackage.DetectType, conn.netPackage.Body)
+				log.Logger.Infof("process agent[%s#%d]'s net package success. Type:%s, Body:%s",
+					conn.Ip, conn.Port, conn.netPackage.DetectType, conn.netPackage.Body)
 				n, err := conn.NetConnection.Write([]byte("OK"))
 				if err != nil {
 					log.Logger.Error("write failed. agent ip:", conn.Ip, " port:", conn.Port)

dbm-services/common/dbha/ha-module/gm/gdm.go

@@ -1,6 +1,7 @@
 package gm
 
 import (
+	"dbm-services/common/dbha/ha-module/util"
 	"net"
 	"strconv"
 	"sync"
@@ -105,7 +106,8 @@ func (gdm *GDM) listenAndDoAccept() {
 			log.Logger.Errorf("accept socket failed. err:%s", err.Error())
 			continue
 		} else {
-			log.Logger.Infof("gdm accept success, agent ip: %v\n", conn.RemoteAddr().String())
+			log.Logger.Infof("gdm accept success, Local: %v, Remote: %v\n",
+				conn.LocalAddr().String(), conn.RemoteAddr().String())
 		}
 		agentConn := AgentConnection{
 			NetConnection: conn,
@@ -132,6 +134,8 @@ func (gdm *GDM) isReporterRecently(ins *DoubleCheckInstanceInfo) bool {
 		log.Logger.Infof("instance[%s#%d] cached, skip report", ip, port)
 		return true
 	}
+	log.Logger.Infof("gdm cache instance:%s, %s",
+		util.GraceStructString(ins), util.GraceStructString(ins.db))
 	// 刷新缓存
 	gdm.ReporterCache[ip] = ins
 	return false

dbm-ui/backend/db_monitor/tpls/alarm/cloud/dbha二次探测失败电话告警.json

@@ -35,6 +35,7 @@
               "cluster_type",
               "appid",
               "cluster_domain",
+              "instance_role",
               "server_ip",
               "server_port",
               "machine_type"
@@ -253,4 +254,4 @@
   "alert_source": "event",
   "custom_conditions": [],
   "export_at": "2024-08-13T15:24:37+08:00"
-}
+}