...
For this, we defined two classes in the dataiku__metrics_checks.py module of the code base. This means that a check on an element (dataset or folder) can be called from any location be it a check python probe, a recipe or a notebook. For better clarity and error tracking, we still prefer to instantiate the classes from the element itself so the scenario can be stopped as soon as a check returns an error.
Common initialization
The following Metric and MetricsAndChecks classes initialization is mostly done through usage of the commonmetric_init_helper function.
Both classes are instantiated on a specific element of the project, be it a dataset or a folder. The function handles the discrepancies between folders and datasets to provide attributes in the same way for both.
Note that it currently only works for folders stored in GCS or another external location, as we do not use internal Dataiku storage (in order to not fill the DSS disk space).
Metric class
The class contains functions that are related to a specific metric, be it for example retrieving the last value or n last values of a given metric.
MetricsAndChecks class
This class is usually instantiated in the status . These classes are instanced in the metrics / checks tab of each dataset of folder that needs to be monitored and the . The right check functions are then called there. You will find more information on each of the class in their dedicated sub-pages.
We define the error level of the check during the instantiation of the class, the two possible values being :
- ERROR when we want a check failure to block the execution of the project
- WARNING otherwise
A global error level can also be defined in the project config by filling the project_error_level value of the metrics_dict dictionnary. When defined, it will override all of the checks error level.
Checks developed in this class are meant to be generic in order to be used by several datasets and folders of the project (even applicable to other projects for most of them). This has for consequence that the returned message will also be quite generic and will in general display the name of the metric that failed a check with the used threshold. It might be necessary to have in mind the element (dataset or folder) on which it is called to better understand the error.
You can You can also find the list of current checks used in the project here with their related element (dataset or folder) and error level (ERROR when we want a check failure to block the execution of the project, WARNING otherwise).Below is an example of usage for a check :.
Usage example
From the "All_features_prices_df_joined" dataset, we can access the metrics and checks by using the "Status" tab.
...