def getDisk(self): mountPoints, diskTotal = [], 0 #diskTotal累加 parts = psutil.disk_partitions() #分区信息 for p in parts: mp = str(p).split(",")[1].split("=")[1] mountPoints.append(mp) for x in mountPoints: pTotal = psutil.disk_usage(x.strip("'")).total #单个分区容量 diskTotal += pTotal diskTotal = changeUnit(int(diskTotal)) rootSize, rootRate = psutil.disk_usage('/').free, psutil.disk_usage('/').percent #获取/分区的剩余空间和使用率 writeToFile("totalDisk", diskTotal) writeToFile("rootSize", changeUnit(int(rootSize))) writeToFile("rootRate", rootRate) if float(rootRate) > float(85): """ON开关,防止多次重复报警;报警一次之后设为关闭""" if self.sDb.read(self.telIP, "rootAlarm") == "ON": ## 状态为"NO"且开关为"ON"时,报故障 sub = "DBTX ROOT partition Alarm:%s" % self.ps1 msg = "DateTime: %s\n根分区已使用%s%%, 超过85%%, 请警惕!!!" % (getTimeNow(), rootRate) saveLog.warning(msg) sendMail(sub, msg) self.sDb.update(self.telIP, "rootAlarm", "OFF") ## 报警后,将开关置为"OFF" else: if self.sDb.read(self.telIP, "rootAlarm") == "OFF": ## 状态OK且开关为"OFF"时,报恢复 sub = "DBTX ROOT partition is OK:%s" % self.ps1 msg = "DateTime: %s\n根分区可用空间充足%s." % (getTimeNow(), changeUnit(int(rootSize))) saveLog.info(msg) sendMail(sub, msg) self.sDb.update(self.telIP, "rootAlarm", "ON") ## 使用率不足85%时,将开关置为"ON"
redisData["connects_rate"] = (int(redisData["total_connects"]) - int(last_connects)) / float(60) redisData["command_rate"] = (int(redisData["total_commands"]) - int(last_commands)) / float(60) fZbx = str(redisData).strip("{}").replace(",","\n").replace("'","") ## 转换成zabbix需要的格式 fObj = open(rZbx,'w') fObj.write(fZbx) fObj.close() except Exception,err: saveLog.error("getRedisBaseInfo(): %s" % str(err)) ## 分析数据,必要时报警: 1, 使用内存大于1G; 2, 命中率低于20% if int(redisData["used_memory"]) > 1073741824: sDb = opRedis() ## 实例化 Redis库操作类 if sDb.read(telIP, "rRateAlarm") == "ON": sub = "DBTX Redis Memory Alarm: %s" % ps1 msg = "DateTime: %s\nRedis使用内存为: %s, 超过1G,请警惕 !!!" % (getTimeNow(), changeUnit(int(redisData["used_memory"]))) saveLog.warning(msg) sendMail(sub, msg) sDb.update(telIP, "rRateAlarm", "OFF") ## 报警后,将开关置为"OFF" else: sDb = opRedis() ## 实例化 Redis库操作类 if sDb.read(telIP, "rRateAlarm") == "OFF": sub = "DBTX Redis Memory is OK: %s" % ps1 msg = "DateTime: %s\nRedis使用内存情况正常: %s." % (getTimeNow(), changeUnit(int(redisData["used_memory"]))) saveLog.info(msg) sendMail(sub, msg) sDb.update(telIP, "rRateAlarm", "ON") ## if totalHits != 0 and redisData["hit_rate"] < 20: ## sDb = opShelveDb() ## 实例化 shelve库操作类 ## if sDb.readDb(pubDb, telIP, "rRateAlarm") == "ON":
def appCpuAnalyze(srvPidStatus, srvFlag, sDb): from dbtx.Merge.etc import ps1 corMuser = "******" tNowCpu = topNow() if not sDb.exists(srvFlag, "ct"): ## Redis库中不存在该服区标识,初始化游戏程序占用CPU达到95%的次数 sDb.intAppCpuCt(srvFlag) for app in srvPidStatus: appName = app.split("_")[0] if srvPidStatus[app][app+"_status"] == 0 or "Snapshot" in app: ## 如果app状态异常,不检测CPU,MEM信息,忽略掉Snapshot continue appPid = srvPidStatus[app][app+"_pid"].split() for pid in appPid: ## CellApp,GateApp,CopyApp会有多个进程ID nCpu, appCount = tNowCpu[pid], int(sDb.read(appName, srvFlag)) if float(nCpu) > 95.5: ## CPU使用率都大于95%时,count+1 appCount += 1 sDb.update(appName, srvFlag, appCount) ## 将新的count值更新到Redis库 else: sDb.update(appName, srvFlag, 0) ## 如果没有超过95%,则置为0 if int(appCount) >= 8: if sDb.read(srvFlag, "isFixing") == "OFF": ## 如果维护标识为"OFF" if sDb.read(srvFlag, "appCpuAlarm") == "ON": ## 状态为"NO"且开关为"ON"时,报故障 sub = "DBTX App-CPU Alarm: %s" % ps1 msg = "DateTime: %s\n%s CPU使用率连续五次达到95%%, 进程ID为: %s !!!" % (getTimeNow(), app, pid) saveLog.warning(msg) sendMail(sub, msg, corMuser) ## sDb.updateDb(mulDb, srvFlag, "appCpuAlarm", "OFF") ## 报警后,将开关置为"OFF" sDb.update(appName, srvFlag, 0) ## 报警后,重置count值 if "DbCacheApp_"+srvFlag == app: proc = psutil.Process(int(pid)) shr = proc.get_ext_memory_info()[2] if float(shr) > 1932735283.2: ## DbCacheApp共享内存大于1.8G时,报警 if sDb.read(srvFlag, "isFixing") == "OFF": ## 如果维护标识为"OFF" if sDb.read(srvFlag, "appShrAlarm") == "ON": ## 状态为"NO"且开关为"ON"时,报故障 sub = "DBTX DbCacheApp-SHR Alarm: %s" % ps1 msg = "DateTime: %s\n%s DbCacheApp 共享内存达到红色警戒,当前占用共享内存: %s !!!" % (getTimeNow(), srvFlag, changeUnit(shr)) saveLog.warning(msg) sendMail(sub, msg, corMuser) sDb.update(srvFlag, "appShrAlarm", "OFF") ## 报警后,将开关置为"OFF" else: if sDb.read(srvFlag, "isFixing") == "OFF": ## 如果维护标识为"OFF" if sDb.read(srvFlag, "appShrAlarm") == "OFF": ## 状态OK且开关为"OFF"时,报恢复 sub = "DBTX DbCacheApp-SHR Alarm: %s" % ps1 msg = "DateTime: %s\n%s DbCacheApp 占用共享内存恢复到正常水平: %s ." % (getTimeNow(), srvFlag, changeUnit(shr)) saveLog.warning(msg) sendMail(sub, msg, corMuser) sDb.update(srvFlag, "appShrAlarm", "ON")
def getMem(self): totMem = changeUnit(int(psutil.TOTAL_PHYMEM)) #获取物理内存总数,diskTotal()函数用来单位转换 writeToFile("totalMem", totMem)