About the Forecaster app
The Forecaster app creates forecasts based on historical usage on a metric of interest and then generates either of these signals:
- Forecast signal — if the threshold is predicted to be breached within a defined duration.
- Static signal — if the threshold has currently been breached.
You can create and view Forecaster configurations in the Obcervant page on the Web Console.
App components Copied
The app consists of the following components to cover a vast amount of time series for forecasting at run-time:
Note
As of the 2.2.0 version, the Forecaster app is capable of handling 20k entities / time series.
Preprocessor Copied
The preprocessor processes live metrics and selectively passes metrics of interest to be forecasted. An entity can be excluded for forecasting for the following reasons:
- data is too noisy to produce any meaningful forecasts
- there is no clear periodic pattern detected
- historical usage has not changed significantly
- historical usage has always been far away from reaching the defined threshold
- entity is inactive as the last updated time is too old
Forecaster Copied
The Forecaster app currently supports two forecast models. The app selects a most suitable model and makes predictions based on historical data.
Signal generator Copied
The signal generator processes forecasts and live metric data to create signals if the metric of interest is predicted to breach the threshold (forecast signal) or has already breached the threshold (static signal). For more information, see Signals.
Together, these app components form a funnel-like workflow.
