Posted On: Mar 29, 2017
Today, Amazon CloudWatch is excited to announce that CloudWatch Alarms now has two new settings to configure alarms on metrics with sparse data or with low sample counts. With the first setting, you have the option to treat missing metric data as good (alarm threshold not breached), bad (alarm threshold breached), maintain the alarm state or use the current default treatment. For example, you can use the treat missing data as good setting for alarms on HTTPCode_ELB_5XX_Count metric. This will ensure that you get alerted only when there are consecutive ELB server errors and not when the errors are sporadic.
The second setting is for alarms based on metric percentiles that might have small number of data samples. By default, CloudWatch Alarms evaluate percentile values irrespective of the sample count, but now you have the option to ignore percentiles with low data samples and maintain the alarm state. The new alarm settings can be applied on any CloudWatch Alarms through the alarms console or PutMetricAlarm API.
To learn more about these new CloudWatch Alarms settings, please see the documentation here.