完善服务状态报表功能

pkg/report-form/node.go

@@ -3,6 +3,7 @@ package report_form
 import (
 	"fmt"
 	"git.quantgroup.cn/DevOps/enoch/pkg/glog"
+	"sort"
 	"time"
 )
 
@@ -12,8 +13,9 @@ type Node struct {
 	startTime   time.Time
 	endTime     time.Time
 	//cpu
-	averageCpu int
-	maxCpu     int
+	removeN100MaxCpu int
+	averageCpu       int
+	maxCpu           int
 	//mem
 	averageMem int
 	maxMem     int
@@ -37,10 +39,65 @@ func NewNode(serviceName string, address string, startTime time.Time, endTime ti
 	rtn.initMem()
 	rtn.initDisk()
 	rtn.initThread()
+	rtn.initRemoveN100MaxCpu()
 
 	return rtn
 }
 
+func (n *Node) initRemoveN100MaxCpu() {
+	const n100 = 3
+
+	//求count
+	countSqlFormat := `SELECT count(system_load_average) FROM machine_info ` +
+		`WHERE sys_name = '%s' AND host = '%s' AND time >= %d AND time < %d;`
+	sql := fmt.Sprintf(countSqlFormat, n.serviceName, n.address, n.startTime.UnixNano(), n.endTime.UnixNano())
+	count, ok := queryOneAndToInt(sql)
+	if !ok {
+		return
+	}
+
+	topSqlFormat := `SELECT top("x", %d) from (SELECT system_load_average/processors*100 as "x" FROM machine_info ` +
+		`WHERE sys_name = '%s' AND host = '%s' AND time >= %d AND time < %d)`
+	sql = fmt.Sprintf(topSqlFormat, count*n100/100+1, n.serviceName, n.address, n.startTime.UnixNano(), n.endTime.UnixNano())
+	resp, err := query(sql)
+	if err != nil {
+		glog.Error("query sql err :", err, sql)
+		return
+	}
+	if len(resp.Results) != 1 {
+		fmt.Println("??3", len(resp.Results))
+		return
+	}
+	res := resp.Results[0]
+	if len(res.Series) != 1 {
+		return
+	}
+	row := res.Series[0]
+	topIdx, ok := getModRowKeyIdx("top", row)
+	if !ok {
+		return
+	}
+	sort.Slice(row.Values, func(i, j int) bool {
+		x1, ok1 := jsonNumberToInt(row.Values[i][topIdx])
+		x2, ok2 := jsonNumberToInt(row.Values[j][topIdx])
+		if !ok1 || !ok2 {
+			return false
+		}
+		if x1 < x2 {
+			return true
+		}
+		return false
+	})
+	if len(row.Values) == 0 {
+		return
+	}
+	removeN100MaxCpu, ok := jsonNumberToInt(row.Values[0][topIdx])
+	if !ok {
+		return
+	}
+	n.removeN100MaxCpu = removeN100MaxCpu
+}
+
 //依次返回：平均、最大
 //如果获取失败返回0，0，0
 func (n *Node) queryKey(key string) (int, int) {
@@ -48,13 +105,14 @@ func (n *Node) queryKey(key string) (int, int) {
 	sqlFormat := `SELECT mean("x"), max("x") ` +
 		`FROM (` +
 		`SELECT %s AS "x" FROM machine_info ` +
-		`WHERE sys_name = '%s' AND host = '%s'AND time >= %d AND time < %d` +
+		`WHERE sys_name = '%s' AND host = '%s'AND time >= %d AND time < %d ` +
 		`);`
 
 	sql := fmt.Sprintf(sqlFormat, key, n.serviceName, n.address, n.startTime.UnixNano(), n.endTime.UnixNano())
 	resp, err := query(sql)
 	if err != nil {
 		glog.Error("node query key:", err)
+		return 0, 0
 	}
 	if len(resp.Results) != 1 {
 		return 0, 0
@@ -124,6 +182,10 @@ func (n *Node) GetMaxCpu() int {
 	return n.maxCpu
 }
 
+func (n *Node) GetRemoveN100MaxCpu() int {
+	return n.removeN100MaxCpu
+}
+
 func (n *Node) GetAverageMem() int {
 	return n.averageMem
 }

pkg/report-form/report_form.go

@@ -150,8 +150,17 @@ func Run(n int) string {
 	rtn.WriteString(fmt.Sprintf("总访问量：%v  总响应时间：%v  平均响应时间：%v\n\n",
 		sm.GetCount(), sm.GetSumDuration(), sm.GetAverageDuration()))
 
+	//服务平均响应时间排行
+	serviceTable := NewTable("服务平均响应时间排行", "duration", "sys_name", "traffic_volume")
+	serviceList := sm.GetAverageDurationServiceList()
+	for _, s := range serviceList {
+		_ = serviceTable.AddRecord(s.GetAverageDuration(), s.Name(), s.GetCount())
+	}
+	rtn.WriteString(serviceTable.ToString())
+	rtn.WriteString("\n")
+
 	//服务响应时间中位值排行
-	medianTable := NewTable("服务响应时间中位值排行", "duration", "sys_name", "path", "count")
+	medianTable := NewTable("服务接口响应时间中位值排行", "duration", "sys_name", "path", "traffic_volume")
 	medianList := sm.GetMedianDurationPathList()
 	for _, m := range medianList {
 		_ = medianTable.AddRecord(m.GetMedianDuration(), m.GetServiceName(), m.GetPath(), m.GetCount())
@@ -160,7 +169,7 @@ func Run(n int) string {
 	rtn.WriteString("\n")
 
 	//服务响应时间平均值排行
-	averageTable := NewTable("服务响应时间平均值排行", "duration", "sys_name", "path", "count")
+	averageTable := NewTable("服务接口响应时间平均值排行", "duration", "sys_name", "path", "traffic_volume")
 	averageList := sm.GetAverageDurationPathList()
 	for _, m := range averageList {
 		_ = averageTable.AddRecord(m.GetAverageDuration(), m.GetServiceName(), m.GetPath(), m.GetCount())
@@ -169,7 +178,7 @@ func Run(n int) string {
 	rtn.WriteString("\n")
 
 	//响应时间最长的请求排行
-	maxTable := NewTable("响应时间最长的请求排行", "duration", "sys_name", "path", "timestamp", "trace_id")
+	maxTable := NewTable("服务接口响应时间最高值排行", "duration", "sys_name", "path", "timestamp", "trace_id")
 	maxList := sm.GetMaxDurationTracePointList()
 	for _, m := range maxList {
 		_ = maxTable.AddRecord(m.Duration, m.ServiceName, m.Path, m.Timestamp, m.TraceId)
@@ -177,20 +186,20 @@ func Run(n int) string {
 	rtn.WriteString(maxTable.ToString())
 	rtn.WriteString("\n")
 
-	//平均cpu使用率排行
-	cpuTable := NewTable("平均cpu使用率排行", "cpu", "sys_name", "count", "average_duration")
-	cpuList := sm.GetAverageCpuServiceList()
-	for _, s := range cpuList {
-		if s.GetAverageCpu() < cpuMin {
+	//去topN100后的峰值cpu使用率排行
+	maxRemoveN100CpuTable := NewTable("常规cpu使用率排行", "cpu", "sys_name", "traffic_volume", "average_duration")
+	maxRemoveN100CpuList := sm.GetRemoveN100MaxCpuServiceList()
+	for _, s := range maxRemoveN100CpuList {
+		if s.GetRemoveN100MaxCpu() < cpuMin {
 			break
 		}
-		_ = cpuTable.AddRecord(s.GetAverageCpu(), s.Name(), s.GetCount(), s.GetAverageDuration())
+		_ = maxRemoveN100CpuTable.AddRecord(s.GetRemoveN100MaxCpu(), s.Name(), s.GetCount(), s.GetAverageDuration())
 	}
-	rtn.WriteString(cpuTable.ToString())
+	rtn.WriteString(maxRemoveN100CpuTable.ToString())
 	rtn.WriteString("\n")
 
-	//平均cpu使用率排行
-	maxCpuTable := NewTable("峰值cpu使用率排行", "cpu", "sys_name", "count", "average_duration")
+	//峰值cpu使用率排行
+	maxCpuTable := NewTable("峰值cpu使用率排行", "cpu", "sys_name", "traffic_volume", "average_duration")
 	maxCpuList := sm.GetMaxCpuServiceList()
 	for _, s := range maxCpuList {
 		if s.GetMaxCpu() < cpuMin {
@@ -202,7 +211,7 @@ func Run(n int) string {
 	rtn.WriteString("\n")
 
 	//内存峰值使用率排行
-	memTable := NewTable("峰值内存使用率排行", "mem", "sys_name", "count", "average_duration")
+	memTable := NewTable("峰值内存使用率排行", "mem", "sys_name", "traffic_volume", "average_duration")
 	memList := sm.GetMaxMemServiceList()
 	for _, s := range memList {
 		if s.GetMaxMem() < memMin {
@@ -214,7 +223,7 @@ func Run(n int) string {
 	rtn.WriteString("\n")
 
 	//硬盘峰值使用率排行
-	diskTable := NewTable("峰值硬盘使用率排行", "disk", "sys_name", "count", "average_duration")
+	diskTable := NewTable("峰值硬盘使用率排行", "disk", "sys_name", "traffic_volume", "average_duration")
 	diskList := sm.GetMaxDiskServiceList()
 	for _, s := range diskList {
 		if s.GetMaxDisk() < diskMin {

pkg/report-form/service.go

@@ -58,7 +58,7 @@ func NewService(name string, startTime time.Time, endTime time.Time) *Service {
 		if _, ok := IgnorePathMap[strings.ToLower(pathSplit[1])]; ok {
 			continue
 		}
-		glog.Info("init path:", rtn.name, path)
+		glog.Info("init path: ", rtn.name, " ",  path)
 		pathObj, ok := NewPath(rtn.name, path, rtn.startTime, rtn.endTime)
 		if !ok {
 			continue
@@ -244,6 +244,12 @@ func (s *Service) GetAverageCpu() int {
 	})
 }
 
+func (s *Service) GetRemoveN100MaxCpu() int {
+	return s.getNodeMapValue(func(node *Node) int {
+		return node.GetRemoveN100MaxCpu()
+	})
+}
+
 func (s *Service) GetMaxCpu() int {
 	return s.getNodeMapValue(func(node *Node) int {
 		return node.GetMaxCpu()

pkg/report-form/service_map.go

@@ -75,7 +75,24 @@ func (sm *ServiceMap) GetAverageCpuServiceList() []*Service {
 	return rtn
 }
 
-//峰值平均cpu使用率列表
+//去N100后的峰值cpu使用率列表
+func (sm *ServiceMap) GetRemoveN100MaxCpuServiceList() []*Service {
+	rtn := make([]*Service, 0)
+	for _, s := range sm.serviceMap {
+		rtn = append(rtn, s)
+	}
+
+	sort.Slice(rtn, func(i, j int) bool {
+		if rtn[i].GetRemoveN100MaxCpu() > rtn[j].GetRemoveN100MaxCpu() {
+			return true
+		}
+		return false
+	})
+
+	return rtn
+}
+
+//峰值cpu使用率列表
 func (sm *ServiceMap) GetMaxCpuServiceList() []*Service {
 	rtn := make([]*Service, 0)
 	for _, s := range sm.serviceMap {
@@ -126,6 +143,28 @@ func (sm *ServiceMap) GetMaxDiskServiceList() []*Service {
 	return rtn
 }
 
+//服务平均响应时间排行
+func (sm *ServiceMap) GetAverageDurationServiceList() []*Service {
+	rtn := make([]*Service, 0)
+
+	for _, s := range sm.serviceMap {
+		//屏蔽掉访问量为0的服务
+		if s.GetCount() == 0 {
+			continue
+		}
+		rtn = append(rtn, s)
+	}
+
+	sort.Slice(rtn, func(i, j int) bool {
+		if rtn[i].GetAverageDuration() > rtn[j].GetAverageDuration() {
+			return true
+		}
+		return false
+	})
+
+	return rtn
+}
+
 //获取最大中位响应时间列表
 func (sm *ServiceMap) GetMedianDurationPathList() []*Path {
 	rtn := make([]*Path, 0)