When you configure an alert, you can specify the number of samples required to calculate the alert status. The calculation appears to reduce the sample number if there is no data to add to the calculation. For example, if I have an alert configured on Request Wait Time Avg with 10 samples of 1 minute intervals, what happens if there was only 1 request in the last 30 minutes? It appears that Monitor is looking back over the last 10 1-minute intervals and removing every interval without a request. Then instead of dividing the total request wait time by 10, it divides the time by 1. If I have a service that once a day has a long wait time and gets normal wait times for the rest of the day, I will receive an alert for that one request when it might have been fine for the previous 2 hours.
Here is an example. The graph shows response times for the last 24 hours. One request at 8:25 PM crossed the warning threshold, but the previous request was at 7:35 PM. When Monitor calculates the average wait time, it only includes the single request at 8:25. I don't want to be alerted about this one request at 8:25. I have set the number of samples to be 10 so that Monitor will use multiple samples in the alert calculation.
