You can create a monitoring rule in the
Monitoring page by clicking on the
New button, on the top-right of your screen. Compared to auto-coverage, you have access to a wider range of monitoring logics for your data assets.
The first step is selecting the asset you want to monitor. Then, you will have to chose the monitoring category to be applied. Depending on your data, your business case and the criticality of that data, you have various possibilities. Here is the exhaustive list of all available templates.
You have the possibility to implement a monitoring rule either based on static or dynamic thresholds computed with machine learning models. What we advise is to use:
- Static: for more known business cases. For instance, when you want to monitor that the price must remain within a given price interval.
- Dynamic: when you don't have any specific thresholds but still want to detect any anomaly.
In case you use a ML-based rule, an important parameter will be the historic: the number of minutes, hours or days of recorded data on which ML models will train. You should select this parameter according to the business periodicity you want to be detected by the model. By default, the model will train on one year of data.
For all rules, it is possible to add a WHERE clause, using the appropriate SQL syntax, to filter records with conditions.
One can also add a GROUP BY statement, in order to distinguish the rule on one field. The rule becomes multi-dimensional, running separately on each group that can have different success or failure statuses.
Updated 17 days ago