Returns the moving average of each time series over the specified time window.
|timeWindow||Amount of time in the moving time window. You can specify a time measurement based on the clock or calendar (1s, 1m, 1h, 1d, 1w), the window length (1vw) of the chart, or the bucket size (1bw) of the chart. Default is minutes if the unit is not specified.|
|expression||Expression describing the time series you want moving averages for.|
mavg() function computes the moving average of each time series over a shifting time window. For example,
mavg(60m, ts(my.metric)) returns, at each point, the average of the data values over the previous 60 minutes for each specified time series.
Here’s how to select your averaging function:
mavg()to see the moving average in a specified time window.
avg()to see the average (the mean).
median()to see the median. Using
median()is preferred if there are a lot of outliers.
mpercentile()with a percentile of 50 to see the moving median.
Let’s say you have the following expression:
ts(my.metric) reports data values once per hour, but only started reporting data 24 hours ago.
mavg() will add up all reported points within the last 48 hours and divide by the number of reported points (not number of hours). So in that example
mavg() divides by 24 not by 48, because technically, we have only 24 reported points within the last 48 hours.
The following example shows the requests latency for a single app server (app-1):
And here’s what we see when we apply
At times, using
msum() can get you the information you want. In that case, use
msum() because performance is better.