I have a CloudwatchAgent emitting High-Res metrics from my ASG. The config looks like this:

// AMAZON-CLOUDWATCH-AGENT.JSON
{
  "agent": {
    "metrics_collection_interval": 10,
    "run_as_user": "cwagent",
    "debug": true
  },
  "metrics": {
    "aggregation_dimensions": [
      [
        "AutoScalingGroupName"
      ]
    ],
    "append_dimensions": {
      "AutoScalingGroupName": "${aws:AutoScalingGroupName}"
    },
    "force_flush_interval": 10,
    "metrics_collected": {
      "cpu": {
        "drop_original_metrics": [
          "cpu_usage_active"
        ],
        "measurement": [
          {
            "name": "cpu_usage_active",
            "rename": "CPUUtilization",
            "unit": "Percent"
          }
        ],
        "metrics_collection_interval": 10,
        "resources": [
          "*"
        ],
        "totalcpu": true
      },
      "mem": {
        "measurement": [
          {
            "name": "mem_used_percent",
            "rename": "MemoryUtilization",
            "unit": "Percent"
          }
        ],
        "metrics_collection_interval": 10
      }
    },
    "namespace": "ASG_SLB_HighRes"
  }
}

And a corresponding Alarm with this config:

{
    "MetricAlarms": [
        {
            "AlarmName": "Some-Name",
            "AlarmArn": "Some-ARN",
            "AlarmConfigurationUpdatedTimestamp": "Some-Timestamp",
            "ActionsEnabled": true,
            "OKActions": [],
            "AlarmActions": [],
            "InsufficientDataActions": [],
            "StateValue": "OK",
            "StateReason": "Threshold Crossed: 1 datapoint...",
            "StateReasonData": "",
            "StateUpdateTimestamp": "Some-Timestamp",
            "MetricName": "CPUUtilization",
            "Namespace": "Some-Namespace",
            "Statistic": "Average",
            "Dimensions": [
                {
                    "Name": "AutoscalingGroupName",
                    "Value": "some-asg"
                }
            ],
            "Period": 10,
            "EvaluationPeriods": 1,
            "Threshold": 50.0,
            "ComparisonOperator": "GreaterThanThreshold",
            "TreatMissingData": "notBreaching",
            "StateTransitionedTimestamp": "Some-Timestamp"
        }
    ],
    "CompositeAlarms": []
}

Period: value is already the minimum of 1 as per AWS Doc

EvaluationPeriods: also set to 1 <- minimum possible

DatapointsToAlarm: Have not set it, since it just then defaults to EvaluationPeriods

However, the Cloudwatch Alarm goes to ALARM State about 25-26 seconds after the breaching data point. I have noticed that there is always a difference of about ~25 seconds in the queryDate and startDate in the StateReasonData of alarm. It is always around 25-26 seconds!

for example: This is the StateReasonData from the Alarm History:

"StateReasonData":{
    "version": "1.0",
    "queryDate": "2026-07-10T20:10:05.419+0000"
    "startDate": "2026-07-10T20:09:40.000+0000",
    "statistic": "Average",
    "period": 10,
    "recentDatapoints": [
        50.000
    ],
    "threshold": 50,
    "evaluatedDatapoints":[
        {
            "timestamp": "2026-07-10T20:09:40.000+0000",
            "sampleCount": 9,
            "value": 50.000
        }
    ]
}

Any actions that I have attached to this alarm, hence, trigger after this lag of ~25 seconds. But this defeats the entire purpose of having high-res metrics! I have searched across the internet, AWS Docs and LLMs -> and I can not explain this delay.

What is this delay? Is this expected? Where can I find more about this? Or if there is further room for optimization in the configs above?