Aggregates can be formed either over the current bucket (hour, day) or the entire lifetime of the underlying timeseries. In some cases, it would also be useful to aggregate over a limited window, e.g., the last 6 hours. That could look something like
The current ways to aggregate would be expressed as window: 1 (default) and window: "all" (lifetime/cumulative)
There's a bunch of things that aren't entirely clear:
Is it enough if the window is always an integer multiple of the bucket? Or should we offer expressing arbitrary times there, like window: "90 minutes"
How would users express that a windowed aggregate should be done for one interval but not the others, e.g., if the above aggregation is done both hourly and daily, how do you express "only add this to the hourly aggregates"
Description
Aggregates can be formed either over the current bucket (hour, day) or the entire lifetime of the underlying timeseries. In some cases, it would also be useful to aggregate over a limited window, e.g., the last 6 hours. That could look something like
The current ways to aggregate would be expressed as
window: 1
(default) andwindow: "all"
(lifetime/cumulative)There's a bunch of things that aren't entirely clear:
window: "90 minutes"