private static void CheckDriverLife(object objectInfo)
{
using var tryLock = new TryLock(_lock);
if (!NVIDIA_MON.nhm_nvidia_is_nvapi_alive() || !NVIDIA_MON.nhm_nvidia_is_nvml_alive())
{
FailCounter++;
RestartDrivers();
if (FailCounter == 20)
{
DriverAliveCheckTimer.Change(0, 60000);
CurrentTimeout = 60000;
}
else if (FailCounter == 30)
{
DriverAliveCheckTimer.Change(0, 3600000);
CurrentTimeout = 3600000;
}
}
else
{
FailCounter = 0;
if (CurrentTimeout != 10000)
{
DriverAliveCheckTimer.Change(0, 10000);
}
}
}
private static void RestartDrivers()
{
NVIDIA_MON.nhm_nvidia_deinit();
NVIDIA_MON.nhm_nvidia_init();
}
public static Task <List <DeviceMonitor> > GetDeviceMonitors(IEnumerable <BaseDevice> devices)
{
return(Task.Run(() =>
{
var ret = new List <DeviceMonitor>();
void addCPUs()
{
var cpus = devices.GetDeviceTypes <CPUDevice>();
foreach (var cpu in cpus)
{
ret.Add(new DeviceMonitorCPU(cpu.UUID));
}
}
void addAMDs()
{
var amds = devices.GetDeviceTypes <AMDDevice>();
if (!amds.Any())
{
return;
}
AMD_ODN.nhm_amd_set_debug_log_level(_amdDebugLogLevel);
AMD_ODN.nhm_amd_reg_log_cb(_amdLog);
var amdInit = AMD_ODN.nhm_amd_init();
if (0 != amdInit)
{
Logger.Info("DeviceMonitorManager", $"AMD nhm_amd_init {amdInit}");
return;
}
foreach (var amd in amds)
{
var hasRet = AMD_ODN.nhm_amd_has_adapter(amd.PCIeBusID);
if (0 == hasRet)
{
ret.Add(new DeviceMonitorAMD(amd.UUID, amd.PCIeBusID));
}
else
{
Logger.Info("DeviceMonitorManager", $"AMD nhm_amd_has_adapter {hasRet} for BusID {amd.PCIeBusID}");
}
}
}
void addNVIDIAs()
{
var nvidias = devices.GetDeviceTypes <CUDADevice>();
if (!nvidias.Any())
{
return;
}
NVIDIA_MON.nhm_nvidia_set_debug_log_level(_nvidiaDebugLogLevel);
NVIDIA_MON.nhm_nvidia_reg_log_cb(_nvidiaLog);
var initialNvmlRestartTimeWait = Math.Min(500 * nvidias.Length, 5000); // 500ms per GPU or initial MAX of 5seconds
var nvidiaUUIDAndBusIds = nvidias.ToDictionary(nvidia => nvidia.UUID, nvidia => nvidia.PCIeBusID);
var nvidiaInit = NVIDIA_MON.nhm_nvidia_init();
NVIDIA_MON.nhm_nvidia_reg_log_cb(_nvidiaLog);
DeviceMonitorNVIDIA.Init();
if (nvidiaInit != 0)
{
Logger.Info("DeviceMonitorManager", $"AMD nhm_nvidia_init {nvidiaInit}");
return;
}
foreach (var nvidia in nvidias)
{
ret.Add(new DeviceMonitorNVIDIA(nvidia.UUID, nvidia.PCIeBusID));
}
}
addCPUs();
addAMDs();
addNVIDIAs();
return ret;
}));
}