Merge pull request #668 from WeBankPartners/dev

Dev
WeBankPartners · Oct 30, 2020 · 6633aec · 6633aec
2 parents a4483af + 61ba749
commit 6633aec
Show file tree

Hide file tree

Showing 52 changed files with 2,507 additions and 601 deletions.
diff --git a/build/conf/ping_exporter.json b/build/conf/ping_exporter.json
@@ -26,7 +26,7 @@
       "weight": 1
     },
     "file": {
-      "enabled" : true,
+      "enabled" : false,
       "path": "ip.txt",
       "weight": 2
     },

diff --git a/build/register.xml b/build/register.xml
@@ -91,7 +91,7 @@
 
     <!-- 6.运行资源 - 描述部署运行本插件包需要的基础资源(如主机、虚拟机、容器、数据库等) -->
     <resourceDependencies>
-        <docker imageName="open-monitor:{{PLUGIN_VERSION}}" containerName="open-monitor-{{PLUGIN_VERSION}}" portBindings="19091:19091,14241:14241,{{ALLOCATE_PORT}}:8080" volumeBindings="{{BASE_MOUNT_PATH}}/prometheus/logs:/app/monitor/prometheus/logs,{{BASE_MOUNT_PATH}}/prometheus/data:/app/monitor/prometheus/data,{{BASE_MOUNT_PATH}}/prometheus/rules:/app/monitor/prometheus/rules,{{BASE_MOUNT_PATH}}/alertmanager/logs:/app/monitor/alertmanager/logs,{{BASE_MOUNT_PATH}}/alertmanager/data:/app/monitor/alertmanager/data,{{BASE_MOUNT_PATH}}/consul/logs:/app/monitor/consul/logs,{{BASE_MOUNT_PATH}}/consul/data:/app/monitor/consul/data,{{BASE_MOUNT_PATH}}/monitor/logs:/app/monitor/monitor/logs,{{BASE_MOUNT_PATH}}/agent_deploy:/app/deploy,{{BASE_MOUNT_PATH}}/transgateway/logs:/app/monitor/transgateway/logs,{{BASE_MOUNT_PATH}}/transgateway/data:/app/monitor/transgateway/data,/etc/localtime:/etc/localtime" envVariables="MONITOR_DB_HOST={{DB_HOST}},MONITOR_DB_PORT={{DB_PORT}},MONITOR_DB_SCHEMA={{DB_SCHEMA}},MONITOR_DB_USER={{DB_USER}},MONITOR_DB_PWD={{DB_PWD}},CORE_ADDR={{CORE_ADDR}},GATEWAY_URL={{GATEWAY_URL}},MONITOR_HOST_IP={{ALLOCATE_HOST}},MONITOR_CHECK_EVENT_KEY={{MONITOR_CHECK_EVENT_KEY}},MONITOR_CHECK_EVENT_TO_MAIL={{MONITOR_CHECK_EVENT_TO_MAIL}},MONITOR_CHECK_EVENT_INTERVAL_MIN={{MONITOR_CHECK_EVENT_INTERVAL_MIN}},MONITOR_ARCHIVE_ENABLE={{MONITOR_ARCHIVE_ENABLE}},MONITOR_ARCHIVE_MYSQL_HOST={{MONITOR_ARCHIVE_MYSQL_HOST}},MONITOR_ARCHIVE_MYSQL_PORT={{MONITOR_ARCHIVE_MYSQL_PORT}},MONITOR_ARCHIVE_MYSQL_USER={{MONITOR_ARCHIVE_MYSQL_USER}},MONITOR_ARCHIVE_MYSQL_PWD={{MONITOR_ARCHIVE_MYSQL_PWD}},MONITOR_LOG_LEVEL={{MONITOR_LOG_LEVEL}},JWT_SIGNING_KEY={{JWT_SIGNING_KEY}},ALARM_FIRING_CALLBACK={{MONITOR_ALARM_FIRING_CALLBACK}},ALARM_RECOVER_CALLBACK={{MONITOR_ALARM_RECOVER_CALLBACK}}"/>
+        <docker imageName="open-monitor:{{PLUGIN_VERSION}}" containerName="open-monitor-{{PLUGIN_VERSION}}" portBindings="19091:19091,14241:14241,{{ALLOCATE_PORT}}:8080" volumeBindings="{{BASE_MOUNT_PATH}}/prometheus/logs:/app/monitor/prometheus/logs,{{BASE_MOUNT_PATH}}/prometheus/data:/app/monitor/prometheus/data,{{BASE_MOUNT_PATH}}/prometheus/rules:/app/monitor/prometheus/rules,{{BASE_MOUNT_PATH}}/alertmanager/logs:/app/monitor/alertmanager/logs,{{BASE_MOUNT_PATH}}/alertmanager/data:/app/monitor/alertmanager/data,{{BASE_MOUNT_PATH}}/consul/logs:/app/monitor/consul/logs,{{BASE_MOUNT_PATH}}/consul/data:/app/monitor/consul/data,{{BASE_MOUNT_PATH}}/monitor/logs:/app/monitor/monitor/logs,{{BASE_MOUNT_PATH}}/agent_deploy:/app/deploy,{{BASE_MOUNT_PATH}}/transgateway/logs:/app/monitor/transgateway/logs,{{BASE_MOUNT_PATH}}/transgateway/data:/app/monitor/transgateway/data,/etc/localtime:/etc/localtime,{{BASE_MOUNT_PATH}}/certs:/data/certs" envVariables="MONITOR_DB_HOST={{DB_HOST}},MONITOR_DB_PORT={{DB_PORT}},MONITOR_DB_SCHEMA={{DB_SCHEMA}},MONITOR_DB_USER={{DB_USER}},MONITOR_DB_PWD={{DB_PWD}},CORE_ADDR={{CORE_ADDR}},GATEWAY_URL={{GATEWAY_URL}},MONITOR_HOST_IP={{ALLOCATE_HOST}},MONITOR_CHECK_EVENT_KEY={{MONITOR_CHECK_EVENT_KEY}},MONITOR_CHECK_EVENT_TO_MAIL={{MONITOR_CHECK_EVENT_TO_MAIL}},MONITOR_CHECK_EVENT_INTERVAL_MIN={{MONITOR_CHECK_EVENT_INTERVAL_MIN}},MONITOR_ARCHIVE_ENABLE={{MONITOR_ARCHIVE_ENABLE}},MONITOR_ARCHIVE_MYSQL_HOST={{MONITOR_ARCHIVE_MYSQL_HOST}},MONITOR_ARCHIVE_MYSQL_PORT={{MONITOR_ARCHIVE_MYSQL_PORT}},MONITOR_ARCHIVE_MYSQL_USER={{MONITOR_ARCHIVE_MYSQL_USER}},MONITOR_ARCHIVE_MYSQL_PWD={{MONITOR_ARCHIVE_MYSQL_PWD}},MONITOR_LOG_LEVEL={{MONITOR_LOG_LEVEL}},JWT_SIGNING_KEY={{JWT_SIGNING_KEY}},ALARM_FIRING_CALLBACK={{MONITOR_ALARM_FIRING_CALLBACK}},ALARM_RECOVER_CALLBACK={{MONITOR_ALARM_RECOVER_CALLBACK}}"/>
         <mysql schema="monitor" initFileName="init.sql" upgradeFileName="upgrade.sql"/>
         <s3 bucketName="wecube-agent"/>
     </resourceDependencies>
@@ -541,7 +541,8 @@
                 <inputParameters>
                     <parameter datatype="string" required="N" sensitiveData="N" mappingType="entity" mappingEntityExpression="">guid</parameter>
                     <parameter datatype="string" required="Y" sensitiveData="N" mappingType="entity" mappingEntityExpression="">host_ip</parameter>
-                    <parameter datatype="string" required="Y" sensitiveData="N" mappingType="entity" mappingEntityExpression="">process_tag</parameter>
+                    <parameter datatype="string" required="Y" sensitiveData="N" mappingType="entity" mappingEntityExpression="">process_name</parameter>
+                    <parameter datatype="string" required="N" sensitiveData="N" mappingType="entity" mappingEntityExpression="">process_tag</parameter>
                     <parameter datatype="string" required="N" sensitiveData="N" mappingType="entity" mappingEntityExpression="">display_name</parameter>
                 </inputParameters>
                 <outputParameters>
@@ -554,7 +555,8 @@
                 <inputParameters>
                     <parameter datatype="string" required="N" sensitiveData="N" mappingType="entity" mappingEntityExpression="">guid</parameter>
                     <parameter datatype="string" required="Y" sensitiveData="N" mappingType="entity" mappingEntityExpression="">host_ip</parameter>
-                    <parameter datatype="string" required="Y" sensitiveData="N" mappingType="entity" mappingEntityExpression="">process_tag</parameter>
+                    <parameter datatype="string" required="Y" sensitiveData="N" mappingType="entity" mappingEntityExpression="">process_name</parameter>
+                    <parameter datatype="string" required="N" sensitiveData="N" mappingType="entity" mappingEntityExpression="">process_tag</parameter>
                     <parameter datatype="string" required="N" sensitiveData="N" mappingType="entity" mappingEntityExpression="">display_name</parameter>
                 </inputParameters>
                 <outputParameters>
@@ -564,5 +566,35 @@
                 </outputParameters>
             </interface>
         </plugin>
+        <plugin name="log_monitor" targetPackage="" targetEntity="" registerName="" targetEntityFilterRule="">
+            <interface action="add" path="/wecube-monitor/api/v1/agent/export/log_monitor/add" filterRule="">
+                <inputParameters>
+                    <parameter datatype="string" required="N" sensitiveData="N" mappingType="entity" mappingEntityExpression="">guid</parameter>
+                    <parameter datatype="string" required="Y" sensitiveData="N" mappingType="entity" mappingEntityExpression="">host_ip</parameter>
+                    <parameter datatype="string" required="Y" sensitiveData="N" mappingType="entity" mappingEntityExpression="">path</parameter>
+                    <parameter datatype="string" required="Y" sensitiveData="N" mappingType="entity" mappingEntityExpression="">keyword</parameter>
+                    <parameter datatype="string" required="N" sensitiveData="N" mappingType="entity" mappingEntityExpression="">priority</parameter>
+                </inputParameters>
+                <outputParameters>
+                    <parameter datatype="string" sensitiveData="N" mappingType="context">errorCode</parameter>
+                    <parameter datatype="string" sensitiveData="N" mappingType="context">errorMessage</parameter>
+                    <parameter datatype="string" sensitiveData="N" mappingType="context">guid</parameter>
+                </outputParameters>
+            </interface>
+            <interface action="delete" path="/wecube-monitor/api/v1/agent/export/log_monitor/delete" filterRule="">
+                <inputParameters>
+                    <parameter datatype="string" required="N" sensitiveData="N" mappingType="entity" mappingEntityExpression="">guid</parameter>
+                    <parameter datatype="string" required="Y" sensitiveData="N" mappingType="entity" mappingEntityExpression="">host_ip</parameter>
+                    <parameter datatype="string" required="Y" sensitiveData="N" mappingType="entity" mappingEntityExpression="">path</parameter>
+                    <parameter datatype="string" required="Y" sensitiveData="N" mappingType="entity" mappingEntityExpression="">keyword</parameter>
+                    <parameter datatype="string" required="N" sensitiveData="N" mappingType="entity" mappingEntityExpression="">priority</parameter>
+                </inputParameters>
+                <outputParameters>
+                    <parameter datatype="string" sensitiveData="N" mappingType="context">errorCode</parameter>
+                    <parameter datatype="string" sensitiveData="N" mappingType="context">errorMessage</parameter>
+                    <parameter datatype="string" sensitiveData="N" mappingType="context">guid</parameter>
+                </outputParameters>
+            </interface>
+        </plugin>
     </plugins>
 </package>
diff --git a/monitor-agent/node_exporter/VERSION b/monitor-agent/node_exporter/VERSION
@@ -1 +1 @@
-1.7.1.16
+1.9.0
diff --git a/monitor-agent/node_exporter/collector/process_monitor_linux.go b/monitor-agent/node_exporter/collector/process_monitor_linux.go
@@ -31,13 +31,13 @@ func (c *processMonitorCollector) Update(ch chan<- prometheus.Metric) error {
 	for _,v := range ProcessCacheObj.get() {
 		ch <- prometheus.MustNewConstMetric(c.processMonitor,
 			prometheus.GaugeValue,
-			v.Value, v.Name, v.Command)
+			v.Value, v.DisplayName, v.Command)
 		ch <- prometheus.MustNewConstMetric(c.processCpuMonitor,
 			prometheus.GaugeValue,
-			v.CpuUsedPercent, v.Name, v.Command)
+			v.CpuUsedPercent, v.DisplayName, v.Command)
 		ch <- prometheus.MustNewConstMetric(c.processMemMonitor,
 			prometheus.GaugeValue,
-			v.MemUsedByte, v.Name, v.Command)
+			v.MemUsedByte, v.DisplayName, v.Command)
 	}
 	return nil
 }
@@ -69,6 +69,7 @@ func NewProcessMonitorCollector() (Collector, error) {
 
 type processMonitorObj struct {
 	Name  string
+	DisplayName string
 	Value   float64
 	Command  string
 	CpuUsedPercent  float64
@@ -113,29 +114,41 @@ func (c *processCache) start()  {
 		c.Lock.Lock()
 		processUsedList := getProcessUsedResource()
 		if len(processUsedList) > 0 {
-			for _,v := range c.ProcessMonitor {
-				tmpName := v.Name
-				//tmpTag := ""
+			for _,tmpProcessMonitorObj := range c.ProcessMonitor {
+				tmpNameList := strings.Split(tmpProcessMonitorObj.Name, "^")
+				tmpName := tmpNameList[0]
+				tmpTag := ""
+				if len(tmpNameList) > 1 {
+					tmpTag = tmpNameList[1]
+				}
+				nameSplit := strings.Split(tmpName, ",")
+				if tmpTag != "" {
+					tmpProcessMonitorObj.DisplayName = fmt.Sprintf("%s(%s)", tmpName, tmpTag)
+				}else{
+					tmpProcessMonitorObj.DisplayName = tmpName
+				}
 				var tmpCount float64 = 0
-				//if strings.Contains(v.Name, "(") {
-				//	tmpNameList := strings.Split(v.Name, "(")
-				//	tmpName = tmpNameList[0]
-				//	tmpTag = strings.Replace(tmpNameList[1], ")", "", -1)
-				//}
 				for _,vv := range processUsedList {
-					//if vv.Name == tmpName && strings.Contains(vv.Cmd, tmpTag) {
-					if strings.Contains(vv.Name, tmpName) || strings.Contains(vv.Cmd, tmpName) {
+					nameMatch := ""
+					for _,nameSplitObj := range nameSplit {
+						if vv.Name == strings.ToLower(nameSplitObj) {
+							nameMatch = nameSplitObj
+							break
+						}
+					}
+					if nameMatch != "" && strings.Contains(vv.Cmd, tmpTag) {
+						tmpProcessMonitorObj.DisplayName = fmt.Sprintf("%s(%s)", nameMatch, tmpTag)
 						tmpCount = tmpCount + 1
 						if len(vv.Cmd) > 100 {
-							v.Command = vv.Cmd[:100]
+							tmpProcessMonitorObj.Command = vv.Cmd[:100]
 						}else{
-							v.Command = vv.Cmd
+							tmpProcessMonitorObj.Command = vv.Cmd
 						}
-						v.CpuUsedPercent = vv.Cpu
-						v.MemUsedByte = vv.Mem
+						tmpProcessMonitorObj.CpuUsedPercent = vv.Cpu
+						tmpProcessMonitorObj.MemUsedByte = vv.Mem
 					}
 				}
-				v.Value = tmpCount
+				tmpProcessMonitorObj.Value = tmpCount
 			}
 		}
 		c.Lock.Unlock()
@@ -209,8 +222,22 @@ func (c *processCache) checkNum(names []string) []int {
 	var result []int
 	for _,v := range names {
 		count := 0
+		tmpNameList := strings.Split(v, "^")
+		tmpName := tmpNameList[0]
+		tmpTag := ""
+		if len(tmpNameList) > 1 {
+			tmpTag = tmpNameList[1]
+		}
+		nameSplit := strings.Split(tmpName, ",")
 		for _,vv := range processUseList {
-			if strings.Contains(vv.Name, v) || strings.Contains(vv.Cmd, v) {
+			nameMatch := ""
+			for _,nameSplitObj := range nameSplit {
+				if vv.Name == strings.ToLower(nameSplitObj) {
+					nameMatch = nameSplitObj
+					break
+				}
+			}
+			if nameMatch != "" && strings.Contains(vv.Cmd, tmpTag) {
 				count = count + 1
 			}
 		}
@@ -254,7 +281,13 @@ func ProcessMonitorHttpHandle(w http.ResponseWriter, r *http.Request)  {
 		for i,v := range checkNumResult {
 			if v != 1 {
 				w.WriteHeader(http.StatusBadRequest)
-				w.Write([]byte(fmt.Sprintf("Process %s num = %d", param.Process[i], v)))
+				tmpProcessName := param.Process[i]
+				if strings.HasSuffix(tmpProcessName, "^") {
+					tmpProcessName = tmpProcessName[:len(tmpProcessName)-1]
+				}else{
+					tmpProcessName = strings.Replace(tmpProcessName, "^", "(", -1) + ")"
+				}
+				w.Write([]byte(fmt.Sprintf("Process %s num = %d", tmpProcessName, v)))
 				illegalFlag = true
 				break
 			}
@@ -292,7 +325,7 @@ func getProcessUsedResource() []processUsedResource {
 							tmpProcessObj.Pid = tmpPid
 						}
 					}else if tmpIndex == 2 {
-						tmpProcessObj.Name = vv
+						tmpProcessObj.Name = strings.ToLower(vv)
 					}else if tmpIndex == 3 {
 						tmpCpu,_ := strconv.ParseFloat(vv, 64)
 						tmpProcessObj.Cpu = tmpCpu

diff --git a/monitor-agent/ping_exporter/funcs/model.go b/monitor-agent/ping_exporter/funcs/model.go
@@ -22,6 +22,7 @@ func DebugLog(msg string, v ...interface{}){
 }
 
 type PingResultObj struct {
+	Ip  string
 	UpDown  int
 	UseTime float64
 }
diff --git a/monitor-agent/ping_exporter/funcs/source.go b/monitor-agent/ping_exporter/funcs/source.go
@@ -55,7 +55,14 @@ func InitSourceList()  {
 			log.Printf("read file %s error: %v \n", Config().Source.File.Path, err)
 		}else{
 			for _,v := range strings.Split(string(ips), "\n") {
-				sourceMap[strings.TrimSpace(v)] = weight
+				tmpMessage := strings.TrimSpace(v)
+				if strings.Contains(tmpMessage, "^") {
+					tmpSplit := strings.Split(tmpMessage, "^")
+					sourceMap[tmpSplit[0]] = weight
+					sourceRemoteData = append(sourceRemoteData, &PingExportSourceObj{Ip:tmpSplit[0], Guid:tmpSplit[1]})
+				}else {
+					sourceMap[tmpMessage] = weight
+				}
 			}
 		}
 	}

diff --git a/monitor-agent/ping_exporter/http_check/task.go b/monitor-agent/ping_exporter/http_check/task.go
@@ -31,15 +31,17 @@ func httpCheckTask()  {
 	clearHttpCheckResult()
 	startTime := time.Now()
 	httpCheckList := funcs.GetHttpCheckList()
+	http.DefaultClient.CloseIdleConnections()
 	wg := sync.WaitGroup{}
 	//var successCounter int
 	for _,v := range httpCheckList {
 		wg.Add(1)
 		go func(method string,url string) {
-			defer wg.Done()
-			b := doHttpCheck(method, url)
+			//b := doHttpCheck(method, url)
+			b := doHttpCheckNew(method, url)
 			writeHttpCheckResult(method, url, b)
 			funcs.DebugLog("http check %s:%s result %d ", method, url, b)
+			wg.Done()
 		}(v.Method,v.Url)
 	}
 	wg.Wait()
@@ -68,6 +70,24 @@ func doHttpCheck(method,url string) int  {
 	return resp.StatusCode
 }
 
+func doHttpCheckNew(method,url string) int {
+	var resp *http.Response
+	var err error
+	if method == "post" {
+		resp,err = http.Post(url, "application/json", strings.NewReader(""))
+	}else{
+		resp,err = http.Get(url)
+	}
+	if err != nil {
+		log.Printf("do http check -> method:%s url:%s response error: %v \n", method, url, err)
+		return 2
+	}
+	if resp.Body != nil {
+		resp.Body.Close()
+	}
+	return resp.StatusCode
+}
+
 func writeHttpCheckResult(method,url string,statusCode int)  {
 	resultLock.Lock()
 	httpCheckResultList = append(httpCheckResultList, &funcs.HttpCheckObj{Method:method, Url:url, StatusCode:statusCode})

diff --git a/monitor-agent/ping_exporter/icmpping/icmp.go b/monitor-agent/ping_exporter/icmpping/icmp.go
@@ -60,13 +60,14 @@ func InitIcmpBytes() {
 	icmpBytes = buffer.Bytes()
 }
 
-func StartPing(distIp string, timeout int) (int,float64) {
+func StartPing(distIp string, timeout int) (int,float64,bool) {
 	var raddr net.IPAddr = net.IPAddr{IP: net.ParseIP(distIp)}
+	isConfused := false
 	//如果你要使用网络层的其他协议还可以设置成 ip:ospf、ip:arp 等
 	conn, err := net.DialIP("ip4:icmp", &localAddress, &raddr)
 	if err != nil {
 		fmt.Println(err.Error())
-		return 3,0
+		return 3,0,isConfused
 	}
 	defer conn.Close()
 	re := 0
@@ -83,33 +84,36 @@ func StartPing(distIp string, timeout int) (int,float64) {
 	useTime := float64(time.Now().Sub(startTime).Nanoseconds()) / 1e6
 	if re >= 3 {  // 发4个ICMP包,如果有3个回复成功则算ping通
 		addSuccessIp(distIp)
-		return 0,useTime/float64(re)
+		return 0,useTime/float64(re),isConfused
 	}else{
+		isConfused = true
 		if tq == 4 {
-			addSuccessIp(distIp)
-			return 0,useTime
+			//addSuccessIp(distIp)
+			return 0,useTime,isConfused
 		}
 		if re == 1 && tq == 3 {
-			addSuccessIp(distIp)
-			return 0,useTime
+			//addSuccessIp(distIp)
+			return 0,useTime,isConfused
 		}
-		if re == 2 && tq == 2{  // 如果有2个回复成功和2个太快回复(下面把这当做了一种异常,有时候主机不通也会出现这种情况),也算主机是通的
-			addSuccessIp(distIp)
-			return 0,useTime/2
-		}
-		if useTime < 6100 {  // 如果4个包不是全部2秒超时,则算异常需要重试
-			funcs.DebugLog("%s ping retry,%.3f ms, renum : %d ## ", distIp, useTime, re)
-			t := GetRetryMap(distIp, re)
-			if t>=4{  // 如果这几次检测中有总数超过4个成功的包返回,也算是成功,经测试在网络流量高的时候会有大概5%-10%的测试IP会只返回2个响应成功的包
-				addSuccessIp(distIp)
-				return 0,useTime/float64(re)
-			}
-			addRetryIp(distIp)
-			return 2,0
-		}else {
-			funcs.DebugLog("%s ping fail,%.3f ms, renum : %d ## ", distIp, useTime, re)
-			return 1,useTime
+		if re == 2 && tq == 2 {  // 如果有2个回复成功和2个太快回复(下面把这当做了一种异常,有时候主机不通也会出现这种情况),也算主机是通的
+			//addSuccessIp(distIp)
+			return 0,useTime/2,isConfused
 		}
+		funcs.DebugLog("%s ping fail,%.3f ms, renum : %d ## ", distIp, useTime, re)
+		return 1,useTime,false
+		//if useTime < 6100 {  // 如果4个包不是全部2秒超时,则算异常需要重试
+		//	funcs.DebugLog("%s ping retry,%.3f ms, renum : %d ## ", distIp, useTime, re)
+		//	t := GetRetryMap(distIp, re)
+		//	if t>=4{  // 如果这几次检测中有总数超过4个成功的包返回,也算是成功,经测试在网络流量高的时候会有大概5%-10%的测试IP会只返回2个响应成功的包
+		//		addSuccessIp(distIp)
+		//		return 0,useTime/float64(re)
+		//	}
+		//	addRetryIp(distIp)
+		//	return 2,0
+		//}else {
+		//	funcs.DebugLog("%s ping fail,%.3f ms, renum : %d ## ", distIp, useTime, re)
+		//	return 1,useTime
+		//}
 	}
 }