aws-applicationautoscaling: Missing treatmissingdata attribute in `BasicStepScalingPolicyProps`

[EHadoux]

Describe the bug

When defining a step scaling policy for a fargate service, the BasicStepScalingPolicyProps doesn't allow for the specification of how to treat the missing data for the alarms that are created. I imagine that it's because it's potentially creating several alarms and we don't want them having the same behaviour. But I couldn't find a way to find the alarms created to then add that afterwards.
I'm scaling up and down depending on the ApproximateNumberOfMessagesVisible in an SQS, but the metric doesn't return a datapoint when it's empty (instead of 0).

Expected Behavior

We have a way to specify how the alarms treat the missing data.

Current Behavior

The attribute doesn't exist and we can't get the alarms created with scaleOnMetric. The default on the alarms is to treat the missing data as missing.

Reproduction Steps

1. Create a fargate service
2. call autoScaleTaskCount
3. call scaleOnMetric

Possible Solution

No response

Additional Information/Context

No response

CDK CLI Version

2.37.1

Framework Version

No response

Node.js Version

16.15.1

OS

Mac OS

Language

Typescript

Language Version

Typescript (4.7.4)

Other information

No response

[peterwoodworth]

We end up creating a CfnScalableTarget with these props - I can't find any options offered to us by CloudFormation, are you aware of any?

[EHadoux]

We end up creating a CfnScalableTarget with these props - I can't find any options offered to us by CloudFormation, are you aware of any?

Interesting. It looks like that Cfn defines the target but it doesn't contain the source (the alarms) so they are probably created somewhere else.
Best case scenario they are created explicitly so we can pass the treatmissingdata directly, worst case they are not created explicitly but we can get a reference to them and add that option afterwards.
I'll edit this comment if I find where that happens in the code.

[EHadoux]

Okay so after a bit of investigation:

aws-cdk/packages/@aws-cdk/aws-applicationautoscaling/lib/step-scaling-policy.ts

Lines 143 to 151 in fbed1e0

	this.lowerAlarm = new cloudwatch.Alarm(this, 'LowerAlarm', {
	// Recommended by AutoScaling
	metric: props.metric,
	alarmDescription: 'Lower threshold scaling alarm',
	comparisonOperator: cloudwatch.ComparisonOperator.LESS_THAN_OR_EQUAL_TO_THRESHOLD,
	evaluationPeriods: props.evaluationPeriods ?? 1,
	datapointsToAlarm: props.datapointsToAlarm,
	threshold,
	});

One of the alarms is defined there, we could pass the argument to that.

Alternatively, if

aws-cdk/packages/@aws-cdk/aws-applicationautoscaling/lib/base-scalable-attribute.ts

Lines 72 to 74 in fbed1e0

	protected doScaleOnMetric(id: string, props: BasicStepScalingPolicyProps) {
	this.target.scaleOnMetric(id, props);
	}

was return this.target.scaleOnMetric, the returned class contains the alarms as class attributes, we might be able to change the treatmissingdata attribute afterwards (although it might not be the most practical solution).

Maybe adding the option there

aws-cdk/packages/@aws-cdk/aws-applicationautoscaling/lib/step-scaling-policy.ts

Line 191 in fbed1e0

export interface ScalingInterval {