HeadCountAlgo/mascommon/include/PerformanceMonitor.h

#ifndef __MAS_PFM_MONITOR_H
#define __MAS_PFM_MONITOR_H

#include "DBMgr.h"
#include "FileHelper.h"

#include "Locks.h"
#include "Thread.h"
#include "SysUtils.h"

#include "nvml.h"

NAMESPACE_MAS_BEGIN

#define MASPFMMONITOR PerformanceMonitor::GetInstance()

typedef nvmlReturn_t (DECLDIR * GPUPERF_NVMLINIT)(void);
typedef nvmlReturn_t (DECLDIR * GPUPERF_NVMLSHUTDOWN)(void);
typedef nvmlReturn_t (DECLDIR * GPUPERF_NVMLDEVICEGETCOUNT)(TZ_Uint32 *);
typedef nvmlReturn_t (DECLDIR * GPUPERF_NVMLDEVICEGETHANDLEBYINDEX)(TZ_Uint32 index, 
		nvmlDevice_t *device);
typedef nvmlReturn_t (DECLDIR * GPUPERF_NVMLDEVICEGETMEMORYINFO)(nvmlDevice_t device, 
		nvmlMemory_t *memory);
typedef nvmlReturn_t (DECLDIR * GPUPERF_NVMLDEVICEGETUTILIZATIONRATE)(nvmlDevice_t device,
		nvmlUtilization_t *utilization);
typedef nvmlReturn_t (DECLDIR * GPUPERF_NVMLDEVICEGETTEMPERATURE)(nvmlDevice_t device, 
		nvmlTemperatureSensors_t sensorType, unsigned int *temp);

// 系统GPU情况
struct SysGPUPerf
{
	TZ_Uint32 gpuIndex;
	TZ_Uint32 gpuUseRate;
	TZ_Uint64 memTotal;
	TZ_Uint64 memFree;
	TZ_Uint64 memUsed;
};

class PerformanceMonitor : tzc::OSThread {

public:
	static PerformanceMonitor * GetInstance();
	static void DestoryInstance();

	TZ_INT Initialize();
	TZ_INT Dispose();

	/* 设置采集时间间隔 */
	void SetIntervalTime(TZ_INT time);

private:
	PerformanceMonitor();
	~PerformanceMonitor();
	
	void initData();

	/* 功能：按设定间隔采集系统性能数据，并写入数据库 */
	virtual void Entry();

	/* CPU占用率 */
	void collectCPUData();

	/* 内存占用率 */
	void collectMemData();

	/* 磁盘IO */
	void collectDiskIOData();

	/* 网络带宽占用率 */
	void collectBandWidthRate();

	/* CPU信息 */
	void collectCPUInfo();

	/* 网络连接信息 */
	void collectNetworkInfo();

	/* 磁盘使用率 */
	void collectDiskInfo();

	/* GPU信息 */
	void collectGPUInfo();
	void collectGPUData();

	void clearGPUperf();

	template<typename T>
	TZ_INT loadNvmlFunc(T & funcPoint, const std::string & funcName);

private:
	TZ_BOOL									m_inited;

	TZ_INT									m_timeCnt;
	TZ_INT									m_intervalTime;		/* 自定义采集频率 */

	GPUPERF_NVMLINIT						m_fNvmlInit;
	GPUPERF_NVMLSHUTDOWN					m_fNvmlShutdown;
	GPUPERF_NVMLDEVICEGETCOUNT				m_fNvmlDeviceGetCount;
	GPUPERF_NVMLDEVICEGETHANDLEBYINDEX		m_fNvmlDeviceGetHandleByIndex;
	GPUPERF_NVMLDEVICEGETMEMORYINFO			m_fNvmlDeviceGetMemoryInfo;
	GPUPERF_NVMLDEVICEGETUTILIZATIONRATE	m_fNvmlDeviceGetUtilizationRate;
	GPUPERF_NVMLDEVICEGETTEMPERATURE		m_fNvmlDeviceGetTemperature;
	
	TZ_DLLHD								m_hdLib;

	TblPerformanceRecordItem				m_tbl;

	/* CPU占用率参数 */
	SysStatInfo								m_cpuDataInfo;
	TZ_DOUBLE								m_cpuUse;

	/* 内存占用率参数 */
	SysMemInfo								m_memDataInfo;

	/* 磁盘IO参数 */
	std::list<SysDiskStatInfo>				m_diskDataList;
	TZ_DOUBLE								m_readSpeed;
	TZ_DOUBLE								m_writeSpeed;

	/* 网络带宽占用率参数 */
	std::list<SysNetBandInfo>				m_netDataList;
	std::map<std::string, TZ_DOUBLE>		m_bandWidthInfoMap;

	/* 系统信息参数 */
	SystemInfo								m_sysInfo;
	std::list<InterfaceInfo>				m_interFaces;

	/* CPU信息参数 */
	CPUInfo									m_cpuInfo;

	/* 网络接口信息参数(端口使用情况) */
	std::list<NetstatInfo>					m_netInfos;

	/* 磁盘信息参数 */
	std::list<DiskstatInfo>					m_diskInfos;

	/* GPU信息参数 */
	std::string								m_GPUName;
	std::vector<SysGPUPerf>					m_vGPUPerf;
	tzc::Mutex								m_lockGPUPerf;

	static PerformanceMonitor				* _ins;
	static tzc::Mutex						_insLock;
};

template<typename T>
TZ_INT PerformanceMonitor::loadNvmlFunc(T & funcPoint, const std::string & funcName)
{
	if (m_hdLib == nullptr)
	{
		TZLogWarn("The hd Lib is nullptr");
		return MEC_FAILED;
	}

	TZ_DLLHD hdFunc = tzc::SysUtils::GetSymbol(m_hdLib, funcName.c_str());
	if (hdFunc == nullptr)
	{
		TZLogWarn("The handle of function is nullptr:%s", funcName.c_str());
		return MEC_FAILED;
	}

	funcPoint = (T)(hdFunc);
	TZLogInfo("load function success!!!:%s", funcName.c_str());

	return MEC_OK;
}

NAMESPACE_MAS_END

#endif