fix: 更新告警参数描述,修正CPU使用率计算逻辑, license检查天数错误

This commit is contained in:
TsMask
2025-05-12 19:07:42 +08:00
parent 872567b28a
commit 73958f4333
4 changed files with 25 additions and 9 deletions

View File

@@ -873,7 +873,7 @@ Minor=3
Warning=4
AddInfo: 告警补充信息
cpuUseGt: CPU使用率大于 范围0~100%
cpuUseGt: CPU使用率大于 范围0~100*CPU核心数
memUseGt: 内存使用率大于, 范围0~100%
diskUseGt: 磁盘使用率大于, 范围0~100%', 'Checks the memory/CPU/disk check health of the network element and sends alerts when thresholds are crossed.
@@ -891,7 +891,7 @@ Minor=3
Warning=4
AddInfo: Additional information on alarms
cpuUseGt: CPU utilization greater than, range 0 to 100%
cpuUseGt: CPU utilization is greater than, range 0~100*number of CPU cores
memUseGt: Memory utilization greater than, range 0 to 100%
diskUseGt: Disk utilization greater than, range 0 to 100%');
INSERT INTO "sys_i18n" VALUES (754, 'job.ne_alarm_state_check_license', '网元告警-License到期检查', 'NE Alarm-License Expire Check');

View File

@@ -767,7 +767,7 @@ INSERT INTO `sys_i18n` VALUES (749, 'dictType.cdr_sip_code_cause', 'IMS-Voice-SI
INSERT INTO `sys_i18n` VALUES (750, 'job.backup_export_log', '备份-日志数据定期导出', 'Backup-Periodic export of Log Data');
INSERT INTO `sys_i18n` VALUES (751, 'job.backup_export_log_remark', 'dataType: 类型支持 operate/login\nfileType: 文件类型 csv/xlsx\nhour: 数据时间从任务执行时间前的小时数', 'Backup-Periodic export of dataType: type support operate/login\nfileType: file type csv/xlsx\nhour: data time from the hour before the task execution time');
INSERT INTO `sys_i18n` VALUES (752, 'job.ne_alarm_state_check_cmd', '网元告警-内存/CPU/磁盘检查', 'NE Alarm-Memory/CPU/Disk Checks');
INSERT INTO `sys_i18n` VALUES (753, 'job.ne_alarm_state_check_cmd_remark', '检查网元的内存/CPU/磁盘检查健康状况,在出现过阈值时发出警报。\r\n\r\nAlarm type:\r\nCommunicationAlarm=1\r\nEquipmentAlarm=2\r\nProcessingFailure=3\r\nEnvironmentalAlarm=4\r\nQualityOfServiceAlarm=5\r\n\r\nSeverity:\r\nCritical=1\r\nMajor=2\r\nMinor=3\r\nWarning=4\r\n\r\nAddInfo: 告警补充信息\r\ncpuUseGt: CPU使用率大于 范围0~100%\r\nmemUseGt: 内存使用率大于, 范围0~100%\r\ndiskUseGt: 磁盘使用率大于, 范围0~100%', 'Checks the memory/CPU/disk check health of the network element and sends alerts when thresholds are crossed.\n\nAlarm type:\nCommunicationAlarm=1\nEquipmentAlarm=2\nProcessingFailure=3\nEnvironmentalAlarm=4\nQualityOfServiceAlarm=5\n\nSeverity:\nCritical=1\nMajor=2\nMinor=3\nWarning=4\r\n\r\nAddInfo: Additional information on alarms\r\ncpuUseGt: CPU utilization greater than, range 0 to 100%\r\nmemUseGt: Memory utilization greater than, range 0 to 100%\r\ndiskUseGt: Disk utilization greater than, range 0 to 100%');
INSERT INTO `sys_i18n` VALUES (753, 'job.ne_alarm_state_check_cmd_remark', '检查网元的内存/CPU/磁盘检查健康状况,在出现过阈值时发出警报。\r\n\r\nAlarm type:\r\nCommunicationAlarm=1\r\nEquipmentAlarm=2\r\nProcessingFailure=3\r\nEnvironmentalAlarm=4\r\nQualityOfServiceAlarm=5\r\n\r\nSeverity:\r\nCritical=1\r\nMajor=2\r\nMinor=3\r\nWarning=4\r\n\r\nAddInfo: 告警补充信息\r\ncpuUseGt: CPU使用率大于 范围0~100*CPU核心数\r\nmemUseGt: 内存使用率大于, 范围0~100%\r\ndiskUseGt: 磁盘使用率大于, 范围0~100%', 'Checks the memory/CPU/disk check health of the network element and sends alerts when thresholds are crossed.\n\nAlarm type:\nCommunicationAlarm=1\nEquipmentAlarm=2\nProcessingFailure=3\nEnvironmentalAlarm=4\nQualityOfServiceAlarm=5\n\nSeverity:\nCritical=1\nMajor=2\nMinor=3\nWarning=4\r\n\r\nAddInfo: Additional information on alarms\r\ncpuUseGt: CPU utilization is greater than, range 0~100*number of CPU cores\r\nmemUseGt: Memory utilization greater than, range 0 to 100%\r\ndiskUseGt: Disk utilization greater than, range 0 to 100%');
INSERT INTO `sys_i18n` VALUES (754, 'job.ne_alarm_state_check_license', '网元告警-License到期检查', 'NE Alarm-License Expire Check');
INSERT INTO `sys_i18n` VALUES (755, 'job.ne_alarm_state_check_license_remark', '检查网元的License是否即将到期在出现过阈值时发出警报。\r\n\r\nAlarm type:\r\nCommunicationAlarm=1\r\nEquipmentAlarm=2\r\nProcessingFailure=3\r\nEnvironmentalAlarm=4\r\nQualityOfServiceAlarm=5\r\n\r\nSeverity:\r\nCritical=1\r\nMajor=2\r\nMinor=3\r\nWarning=4\r\n\r\nAddInfo: 告警补充信息\r\ndayLt: 天数小于默认30天', 'Checks if the network element is License is about to expire and sends an alert if a threshold is crossed.\n\nAlarm type:\nCommunicationAlarm=1\nEquipmentAlarm=2\nProcessingFailure=3\nEnvironmentalAlarm=4\nQualityOfServiceAlarm=5\n\nSeverity:\nCritical=1\nMajor=2\nMinor=3\nWarning=4\r\n\r\nAddInfo: Additional information on alarms\r\ndayLt: Days less than, default 30 days');

View File

@@ -3,6 +3,7 @@ package ne_alarm_state_check_cmd
import (
"encoding/json"
"fmt"
"runtime"
"sort"
"strconv"
"strings"
@@ -26,6 +27,8 @@ var NewProcessor = &NeAlarmStateCheckCMDProcessor{
alarmService: neDataService.NewAlarm,
wsSendService: wsService.NewWSSend,
count: 0,
triggerMax: 4,
triggerCount: 0,
}
// NeAlarmStateCheckCMDProcessor 网元告警内存/CPU/磁盘检查
@@ -36,7 +39,8 @@ type NeAlarmStateCheckCMDProcessor struct {
alarmService *neDataService.Alarm // 告警信息服务
wsSendService *wsService.WSSend // ws发送服务
count int // 执行次数
triggerMax int // 阈值连续触发次数大于才会产生告警
triggerCount int // 阈值连续触发次数
}
// alarmParams 告警参数
@@ -47,7 +51,7 @@ type alarmParams struct {
SpecificProblem string `json:"specificProblem"` // Alarm Cause: CPU/Menory/Disk status received from target NE reaches the threshold
SpecificProblemID string `json:"specificProblemId"` // AC10100
AddInfo string `json:"addInfo"` // 告警补充信息
CPUUseGt int64 `json:"cpuUseGt"` // CPU使用率大于 范围0~100%
CPUUseGt int64 `json:"cpuUseGt"` // CPU使用率大于 范围0~100*CPU核心数
MemUseGt int64 `json:"memUseGt"` // 内存使用率大于, 范围0~100%
DiskUseGt int64 `json:"diskUseGt"` // 磁盘使用率大于, 范围0~100%
@@ -72,8 +76,9 @@ func (s *NeAlarmStateCheckCMDProcessor) Execute(data any) (any, error) {
return nil, fmt.Errorf("json params err: %v", err)
}
// 检查使用率
if params.CPUUseGt > 100 || params.CPUUseGt < 0 {
return nil, fmt.Errorf("cpuUseGt must be between 0 and 100")
numCPU := runtime.NumCPU()
if params.CPUUseGt > int64(numCPU*100) || params.CPUUseGt < 0 {
return nil, fmt.Errorf("cpuUseGt must be between 0 and 100 * NumCPU")
}
if params.MemUseGt > 100 || params.MemUseGt < 0 {
return nil, fmt.Errorf("memUseGt must be between 0 and 100")
@@ -137,6 +142,7 @@ func (s *NeAlarmStateCheckCMDProcessor) Execute(data any) (any, error) {
// 进行新增
newAlarm, err := s.alarmNew(neInfo, params)
params.AddInfo = addInfo // 恢复附加信息
s.triggerCount = 0 // 重置连续触发次数
if err != nil {
result[neTypeAndId] = err.Error()
continue
@@ -227,7 +233,10 @@ func (s NeAlarmStateCheckCMDProcessor) serverState(state map[string]any, cpuUseG
warnMsg = append(warnMsg, fmt.Sprintf("disk usage %.2f%%", sysDiskUsage))
}
if len(warnMsg) > 0 {
return fmt.Errorf("greater than %s", strings.Join(warnMsg, ", "))
s.triggerCount++
if s.triggerCount > s.triggerMax {
return fmt.Errorf("greater than %s", strings.Join(warnMsg, ", "))
}
}
return nil
}

View File

@@ -120,8 +120,15 @@ func (s *NeAlarmStateCheckLicenseProcessor) Execute(data any) (any, error) {
}
// 未记录
if alarmStatus == "" {
addInfo := params.AddInfo
if params.AddInfo != "" {
params.AddInfo = params.AddInfo + ", " + err.Error()
} else {
params.AddInfo = err.Error()
}
// 进行新增
newAlarm, err := s.alarmNew(neInfo, params)
params.AddInfo = addInfo // 恢复附加信息
if err != nil {
result[neTypeAndId] = err.Error()
continue
@@ -150,7 +157,7 @@ func (s NeAlarmStateCheckLicenseProcessor) serverState(state map[string]any, day
}
// 计算距离天数
daysLeft := int64(time.Since(expireTime).Hours() / 24)
daysLeft := int64(time.Until(expireTime).Hours() / 24)
if daysLeft < dayLt {
return fmt.Errorf("license will expire in %d days", daysLeft)
}