Target scaler is not working (.NET)

[pierre-sk]

Hello everyone,

I create this issue to link to this fix proposal: #451

The KafkaTargetScaler is using the obsolete ScaleMonitorDescriptor constructor without the functionId parameter.

[Obsolete("This constructor is obsolete. Use the version that takes function id instead.")]
public ScaleMonitorDescriptor(string id)
{
	Id = id;
}

public ScaleMonitorDescriptor(string id, string functionId) : this(id)
{
	FunctionId = functionId;
}

When target scaling is enabled (by default) the ScaleManager compares the FunctionId to evaluate if a ScaleMonitor is used in case a TargetScaler already exists. But for the same function, the FunctionId of the TargetScaler and ScaleMonitor are not equal.

// Check if TBS enabled on app level
if (scaleOptions.Value.IsTargetScalingEnabled)
{
	HashSet<string> targetScalerFunctions = new HashSet<string>();
	foreach (var scaler in targetScalers)
	{
		string scalerUniqueId = GetTargetScalerFunctionUniqueId(scaler);
		if (!_targetScalersInError.Contains(scalerUniqueId))
		{
			string assemblyName = GetAssemblyName(scaler.GetType());
			bool featureDisabled = configuration.GetValue<string>(assemblyName) == "0";
			if (!featureDisabled)
			{
				targetScalersToSample.Add(scaler);
				targetScalerFunctions.Add(scalerUniqueId);
			}
		}
	}

	foreach (var monitor in scaleMonitors)
	{
		string monitorUniqueId = GetScaleMonitorFunctionUniqueId(monitor);
		// Check if target based scaler exists for the function
		if (!targetScalerFunctions.Contains(monitorUniqueId))
		{
			scaleMonitorsToSample.Add(monitor);
		}
	}
}

Both ScaleMonitor and TargetScaler are retrieved for sampling. It leads to a bad behavior as the ScaleManager doesn't use the TargetScaler to override the vote result when there is any ScaleMonitor status.

// Set correct vote if all the triggers are target
if (!scaleStatuses.Any() && aggregateScaleStatus.TargetWorkerCount.HasValue)
{
    aggregateScaleStatus.Vote = (ScaleVote)aggregateScaleStatus.TargetWorkerCount.Value.CompareTo(context.WorkerCount);
}

Example of a bad result on /admin/host/scale/status function runtime endpoint:

{
    "vote": 0,
    "targetWorkerCount": 4,
    "functionScaleStatuses": {
        "my.functions.myfirstfunction.runasync-kafkatrigger-topic.myconsumer": {
            "vote": 0
        },
        "my.functions.mysecondfunction.runasync-kafkatrigger-topic.myconsumer": {
            "vote": 0
        }
    },
    "functionTargetScalerResults": {
        "My.Functions.MyFirstFunction.RunAsync": {
            "targetWorkerCount": 4
        },
        "My.Functions.MySecondFunction.RunAsync": {
            "targetWorkerCount": 1
        }
    }
}

[pierre-sk]

Hello @jainharsh98,

Any chance to take a look to #451?

Thanks

Regards