Consider that you apply a filter on the Contact object with the condition that the first names start with the letter A.

The following image shows the data subset filter criteria that you can configure:

Since the Account object is parent to the Contact object and the filter is applied on the Contact object, the task does not select a relationship for child record selection. To maintain referential integrity, the task traverses from Contact to Account through both the relationships Account and ParentAccount__r. The number of join operations to compute the subset with the default path selection is two.

The following image shows the graphical representation of the relationship between Account and Contact objects:

The task first loads the records from the Contact object on which the filter is applied. Then the task traverses through the paths from Contact to Account through the relationships Account and ParentAccount__r, and loads the records.

The following image shows the sequence in which the task selects the records:

Based on the filter applied, the Contact object has 12 rows. From Contact to Account, the Account object has 13 rows through the relationships Account and ParentAccount__r.

To view the number of subset rows, you estimate the subset. If there are common records from multiple join operations, the task updates the records. In this use case, though the Account object shows total 13 rows, the task loads 12 subset rows that are unique for the Account object from all the relationships. The task loads 12 subset rows for the Contact object.

The following image shows the subset statistics that you can estimate in a task: