/// <summary>
/// Function that gets fan speed requirements
/// from all blades. It also updates the blade states.
/// </summary>
private static void GetAllBladePwmRequirements()
{
// Rate is required to timestep over each individual Blade call
double rate = (double)getBladePwmReqtTimePeriodInMilliseconds / (double)MaxSledCount;
double timeDiff = 0;
for (byte blade = 1; blade <= MaxSledCount; blade++)
{
// Handle shutdown state
if (ChassisState.ShutDown)
{
return;
}
// default PWM setting
byte PWM = (byte)ConfigLoaded.MinPWM;
// Query blade type from IPMI layer
ChassisState.BladeTypeCache[blade - 1] = (byte)WcsBladeFacade.clients[blade].BladeClassification;
// wait for rate limiter which includes the previous time difference for sensor get, and then issue get fan requirement
double sleepTime = rate - timeDiff;
if (sleepTime > rate)
{
sleepTime = rate;
}
if (sleepTime > 0)
{
Thread.Sleep(TimeSpan.FromMilliseconds(sleepTime));
}
Tracer.WriteInfo("GetBladeRequirement called at {0} for BladeId {1} (state: {2})", DateTime.Now, blade,
ChassisState.GetStateName(blade));
// Check for the condition where known state is hardpoweroff, but someone plugged a new blade in
if (ChassisState.GetBladeState(blade) == (byte)BladeState.HardPowerOff)
{
CheckPowerEnableState(blade);
}
// Log Start time
DateTime startTime = DateTime.Now;
#region Check fail State -> Initialize
// If blade was in Fail state
if (ChassisState.GetBladeState(blade) == (byte)BladeState.Fail)
{
// If failed count is greater than a maximum value, we move it to Initialization state
if (ChassisState.FailCount[blade - 1] > ConfigLoaded.MaxFailCount)
{
// Move to Initialization state so that this blade could be reinitialized
Tracer.WriteInfo("GetAllBladePwmRequirements: State Transition for blade {0}: {1} -> Initialization", blade,
ChassisState.GetStateName(blade));
ChassisState.SetBladeState(blade, (byte)BladeState.Initialization);
}
else
{
// Moving out of Fail state - First we use a light-weight get GUID to check whether the blade is there.
// do not allow retries on Get System Guid
DeviceGuid guid = WcsBladeFacade.GetSystemGuid(blade, false);
if (guid.CompletionCode == (byte)CompletionCode.Success)
{
Tracer.WriteInfo("GetAllBladePwmRequirements: GUID present for blade {0}, GUID: {1}", blade, guid.Guid.ToString());
DeviceGuid cachedGuid = WcsBladeFacade.GetCachedGuid(blade);
if (guid.Guid == cachedGuid.Guid)
{
// Change state to Probation and assume the system was in fail due to timeout.
Tracer.WriteInfo("GetAllBladePwmRequirements: State Transition for blade {0}: {1} -> Probation", blade,
ChassisState.GetStateName(blade));
ChassisState.SetBladeState(blade, (byte)BladeState.Probation);
}
else
{
// Change state to Initialization as the device has changed.
Tracer.WriteInfo("GetAllBladePwmRequirements: State Transition for blade {0}: {1} -> Probation", blade,
ChassisState.GetStateName(blade));
ChassisState.SetBladeState(blade, (byte)BladeState.Initialization);
}
}
else
{
Tracer.WriteInfo("GetAllBladePwmRequirements: Get System GUID returns a bad completion status: {0}", guid.CompletionCode);
}
}
// Increase time spent in Fail state everytime we are in this state
ChassisState.FailCount[blade - 1]++;
}
#endregion
#region Move Initialize -> Probation
// Handles Initialization
if (ChassisState.GetBladeState(blade) == (byte)BladeState.Initialization)
{
BladePowerStatePacket powerstate = ChassisState.BladePower[blade - 1].GetCachedBladePowerState();
if (powerstate.CompletionCode == 0)
{
if (powerstate.DecompressionTime == 0)
{
// Will result in Hard Power off or Probation
ReInitialize(blade);
}
}
}
#endregion
// Normal operation - possible states are probation or healthy
if (ChassisState.GetBladeState(blade) == (byte)BladeState.Probation ||
ChassisState.GetBladeState(blade) == (byte)BladeState.Healthy)
{
#region Jbod (no sensor reading)
if (ChassisState.GetBladeType(blade) == (byte)BladeType.Jbod)
{
// Do not allow retries on system guid.
DeviceGuid guid = WcsBladeFacade.GetSystemGuid(blade, false);
if (guid.CompletionCode == (byte)CompletionCode.Success)
{
Tracer.WriteInfo("GetAllBladePwmRequirements: GUID present for JBOD {0}, GUID: {1}",
blade, guid.Guid.ToString());
// Change state to Healthy
if (ChassisState.GetBladeState(blade) == (byte)BladeState.Probation)
{
Tracer.WriteInfo("GetAllBladePwmRequirements: State Transition for JBOD {0}: {1} -> Healthy",
blade, ChassisState.GetStateName(blade));
ChassisState.SetBladeState(blade, (byte)BladeState.Healthy);
}
}
else
{
Tracer.WriteInfo("GetAllBladePwmRequirements: Get System GUID for JBOD {0} failed with status {1}",
blade, guid.CompletionCode);
// Set it to failed state, where we will retry guids and reinitialize if needed
Tracer.WriteInfo("GetAllBladePwmRequirements: State Transition for JBOD {0}: {1} -> Fail",
blade, ChassisState.GetStateName(blade));
ChassisState.SetBladeState(blade, (byte)BladeState.Fail);
}
// No need to check for sensor reading, just continue
continue;
}
#endregion
#region Server -> Get PWM move to Healthy or move to Fail
// Call temperature reading list command
SensorReading Temps = WcsBladeFacade.GetSensorReading((byte)blade, (byte)ConfigLoaded.InputSensor, PriorityLevel.System);
if (Temps.CompletionCode != (byte)CompletionCode.Success)
{
Tracer.WriteWarning("GetAllBladePwmRequirements: BladeId: {0} - GetSensorReading for temperature failed with code {1:X}",
blade, Temps.CompletionCode);
// Move to Fail state if no readings were obtained
Tracer.WriteInfo("GetAllBladePwmRequirements: State Transition for blade {0}: {1} -> Fail", blade,
ChassisState.GetStateName(blade));
ChassisState.SetBladeState(blade, (byte)BladeState.Fail);
}
else
{
Tracer.WriteInfo("GetAllBladePwmRequirements: #### BladeId = " + blade + " Sensor id= " + ConfigLoaded.InputSensor +
" Sensor reading = " + Temps.Reading + " Raw = " + Temps.RawReading +
", LowerNonCritical= " + ConfigLoaded.SensorLowThreshold + ", UpperNonCritical= " + ConfigLoaded.SensorHighThreshold);
// Handle state logic if needed
// Probation state should be shifted to Healthy since there was no timeout, & sensorread succeeded
if (ChassisState.GetBladeState(blade) == (byte)BladeState.Probation)
{
// Change state to healthy
Tracer.WriteInfo("GetAllBladePwmRequirements: State Transition for blade {0}: {1} -> Healthy",
blade, ChassisState.GetStateName(blade));
ChassisState.SetBladeState(blade, (byte)BladeState.Healthy);
ChassisState.FailCount[blade - 1] = 0; // reset the fail count
// When a blade transitions to 'Healthy' state, enable/disable default blade operations
EnableDisableDefaultBladeOperations(blade);
}
if (ConfigLoaded.InputSensor != 1) // Non-PWM sensor.
{
PWM = GetPwmFromTemperature(Temps.Reading,
ConfigLoaded.SensorLowThreshold,
ConfigLoaded.SensorHighThreshold);
}
else
{
// PWM should never be higher or lower than the threshold.
if (Temps.Reading < ConfigLoaded.MinPWM || Temps.Reading > ConfigLoaded.MaxPWM)
{
Tracer.WriteWarning("PWM value " + Temps.Reading + " on blade " + blade +
" is out of range (lowThreshold: " + ConfigLoaded.MinPWM +
" - highThreshold: " + ConfigLoaded.MaxPWM);
PWM = (byte)ConfigLoaded.MinPWM;
}
else
{
PWM = (byte)Temps.Reading;
}
}
Tracer.WriteInfo("PWM value on blade {0} for Sensor {1} = {2}", blade, InputSensor, PWM);
}
#endregion
}
// write value into requirements table
BladeRequirementTable[blade - 1] = PWM;
// Log end time and capture time of execution for sensor get command
DateTime endTime = DateTime.Now;
timeDiff = endTime.Subtract(startTime).TotalMilliseconds; // convert time difference into milliseconds
}
}
/// <summary>
/// Function that gets all the fan speed requirements
/// from the Blade. It also updates the balde state
/// </summary>
private void GetAllBladePwmRequirements()
{
// Rate is required to timestep over each individual Blade call
double rate = (double)GetTimePeriod / (double)MaxSledCount;
double timeDiff = 0;
for (byte blade = 1; blade <= MaxSledCount; blade++)
{
// Handle shutdown state
if (ChassisState.ShutDown)
{
return;
}
// default PWM setting
byte PWM = (byte)ConfigLoaded.MinPWM;
// Query blade type from IPMI layer
ChassisState.BladeTypeCache[blade - 1] = (byte)WcsBladeFacade.clients[blade].BladeClassification;
// wait for rate limiter which includes the previous time difference for sensor get, and then issue get fan requirement
double sleepTime = rate - timeDiff;
if (sleepTime > rate)
{
sleepTime = rate;
}
if (sleepTime > 0)
{
Thread.Sleep(TimeSpan.FromMilliseconds(sleepTime));
}
if (CommunicationDevice.IsSafeMode())
{
// Do not perform any sensor reading - continue in the for loop
Tracer.WriteInfo("Monitoring thread: Safe Mode, Skipping sensor read");
continue;
}
Tracer.WriteInfo("GetBladeRequirement called at {0} for sledId {1} (state: {2})", DateTime.Now, blade,
ChassisState.GetStateName(blade));
// Check for the condition where known state is hardpoweroff, but someone plugged a new blade in
if (ChassisState.GetBladeState(blade) == (byte)BladeState.HardPowerOff)
{
ChassisState.PowerFailCount[blade - 1]++;
// TODO: identify if this period is sufficient to do this check
if (ChassisState.PowerFailCount[blade - 1] > (ConfigLoaded.MaxRetries * ConfigLoaded.Population))
{
CheckPowerEnableState(blade);
ChassisState.PowerFailCount[blade - 1] = 0;
}
}
// Log Start time
DateTime startTime = DateTime.Now;
// If blade was in Fail state
if (ChassisState.GetBladeState(blade) == (byte)BladeState.Fail)
{
// If failed count is greater than a maximum value, we move it to Initialization state
if (ChassisState.FailCount[blade - 1] > ConfigLoaded.MaxFailCount)
{
// Move to Initialization state so that this sled could be reinitialized
Tracer.WriteInfo("State Transition for Sled {0}: {1} -> Initialization", blade,
ChassisState.GetStateName(blade));
ChassisState.SetBladeState(blade, (byte)BladeState.Initialization);
}
else
{
// Moving out of Fail state - First we use a light-weight get GUID to check whether the blade is there
DeviceGuid guid = WcsBladeFacade.GetSystemGuid(blade);
if (guid.CompletionCode == (byte)CompletionCode.Success)
{
Tracer.WriteInfo("GUID present for sled {0}, GUID: {1}", blade, guid.Guid.ToString());
// Change state to Probation
Tracer.WriteInfo("State Transition for Sled {0}: {1} -> Probation", blade,
ChassisState.GetStateName(blade));
ChassisState.SetBladeState(blade, (byte)BladeState.Probation);
}
else
{
Tracer.WriteInfo("Get System GUID returns a bad completion status: {0}", guid.CompletionCode);
}
}
// Increase time spent in Fail state everytime we are in this state
ChassisState.FailCount[blade - 1]++;
}
// Handles Initialization
if (ChassisState.GetBladeState(blade) == (byte)BladeState.Initialization)
{
this.ReInitialize(blade);
}
// Normal operation - possible states are probation or healthy
if (ChassisState.GetBladeState(blade) == (byte)BladeState.Probation ||
ChassisState.GetBladeState(blade) == (byte)BladeState.Healthy)
{
if (ChassisState.GetBladeType(blade) == (byte)BladeType.Jbod)
{
DeviceGuid guid = WcsBladeFacade.GetSystemGuid(blade);
if (guid.CompletionCode == (byte)CompletionCode.Success)
{
Tracer.WriteInfo("GUID present for jbod {0}, GUID: {1}", blade, guid.Guid.ToString());
// Change state to Probation
Tracer.WriteInfo("State Transition for jbod {0}: {1} -> Healthy", blade, ChassisState.GetStateName(blade));
ChassisState.SetBladeState(blade, (byte)BladeState.Healthy);
}
else
{
Tracer.WriteInfo("Get System GUID for jbod {0} failed with status {1}", blade, guid.CompletionCode);
// Set it to failed state, where we will retry guids and reinitialize if needed
Tracer.WriteInfo("State Transition for jbod {0}: {1} -> Fail", blade, ChassisState.GetStateName(blade));
ChassisState.SetBladeState(blade, (byte)BladeState.Fail);
}
// No need to check for sensor reading, just continue
continue;
}
// Call temperature reading list command
SensorReading Temps = WcsBladeFacade.GetSensorReading((byte)blade, (byte)ConfigLoaded.InputSensor, PriorityLevel.System);
if (Temps.CompletionCode != (byte)CompletionCode.Success)
{
Tracer.WriteWarning("SledId: {0} - getTempSensorReading failed with code {1:X}", blade, Temps.CompletionCode);
// Move to Fail state if no readings were obtained
Tracer.WriteInfo("State Transition for Sled {0}: {1} -> Fail", blade,
ChassisState.GetStateName(blade));
ChassisState.SetBladeState(blade, (byte)BladeState.Fail);
}
else
{
Tracer.WriteInfo("#### Sledid= " + blade + " Sensor id= " + ConfigLoaded.InputSensor + " Sensor reading= " +
Temps.Reading + " Raw= " + Temps.RawReading + ", LowerNonCritical= " +
ConfigLoaded.SensorLowThreshold + ", UpperNonCritical= " + ConfigLoaded.SensorHighThreshold);
// Handle state logic if needed
// Probation state should be shifted to Healthy since there was no timeout, & sensorread succeeded
if (ChassisState.GetBladeState(blade) == (byte)BladeState.Probation)
{
// Change state to healthy
Tracer.WriteInfo("State Transition for Sled {0}: {1} -> Healthy", blade,
ChassisState.GetStateName(blade));
ChassisState.SetBladeState(blade, (byte)BladeState.Healthy);
ChassisState.FailCount[blade - 1] = 0; // reset the fail count
}
PWM = GetPwmFromTemperature(Temps.Reading,
ConfigLoaded.SensorLowThreshold,
ConfigLoaded.SensorHighThreshold);
Tracer.WriteInfo("PWM value for Sensor {0} = {1}", InputSensor, PWM);
}
}
// write value into requirements table
BladeRequirementTable[blade - 1] = PWM;
// Log end time and capture time of execution for sensor get command
DateTime endTime = DateTime.Now;
timeDiff = endTime.Subtract(startTime).TotalMilliseconds; // convert time difference into milliseconds
}
}