For a better calculation of dynamic thresholds and seasonal patterns, it would be useful if the baselinewma() and baselinedev() trend functions had a possibility to determine from which trend values (avg, max or min) the moving average should be calculated from.

Assuming we had this possibility, we could, for example, make very good predictions about the range in which a metric has been in the last few days, weeks or months.

Parameters

avg: Default as it was before. Collect the avg trend values and calculate the baseline from them.
max: Collect the max trend values only and calculate the baseline from them.
min: Collect the min trend values only and calculate the baseline from them.

Examples

baselinewma(//vfs.fs.size[{#FSNAME},used], 1h:now/h+1h, "d", 10, "max")
baselinewma(//vfs.fs.size[{#FSNAME},used], 1h:now/h+1h, "w", 3, "max") 
baselinewma(//vfs.fs.size[{#FSNAME},used], 1h:now/h+1h, "M", 3, "max")

baselinewma(//vfs.fs.size[{#FSNAME},used], 1h:now/h+1h, "d", 10, "min")
baselinewma(//vfs.fs.size[{#FSNAME},used], 1h:now/h+1h, "w", 3, "min")
baselinewma(//vfs.fs.size[{#FSNAME},used], 1h:now/h+1h, "M", 3, "min")

With regard to dynamic thresholds, you could then take the baseline obtained and check whether a metric is within the valid range. A good use case for this would be scripts, backup jobs or other recurring patterns, which may write almost to the maximum of a hard disk, but no more. For such cases, even an utilization of 98% would be acceptable, which could now be predicted with such a feature successfully.