I’m recording a custom metric in cloudwatch where once every minute we send a ping from several connected clients. We’re trying to graph these in a Number chart over a period of 5 minutes and are using the SUM statistic. What I expect to see in the output is the number of times in the past minute the metric has received a successful ping, i.e. I expect each of the outputs on the graph to show 5. We have had this graph for several months, when you look at it you usually see a 4, 5 or 6 there, which is fine.
However, it appears as though AWS has changed something recently where instead of looking at the past rolling 5 minutes, it looks at 5 minute blocks. In other words, if it’s 1:39 PM it will show the number 5 (1:35 - 1:39), but when it rolls over to 1:40 PM, it just shows the number 1. It looks like it’s looking for the SUM of all pings from 1:40 - 1:44, because at 1:41 PM it will show 2, 1:42 it shows 3 and so on up to 1:44 PM. This is a problem for us because if our ping mechanism goes down the way we’ll know something is wrong is by seeing that number be < 5. So 4 is concerning, and 3 or 2 probably indicate a problem. With the way it’s currently SUMming the metrics, we don’t have any forewarning like that.
Does anyone know how we can tell Cloudwatch graphing to apply its statistic to the previous 5-minute period on a rolling basis instead? For example at 1:42 PM the statistic would look at all metrics during the 1:38 - 1:42 5-minute time period as it used to do?