Denormalized rows are read using the Denormalized-View and then sorted by the flattening keys.

The sorted data is used to build IDT rows. Data is moved from denormalized rows into an IDT row. An IDT row is written out either

The latter will occur when a repeating field is full. If more data arrives with the same key value, additional IDT rows are built with the same key value.

The construction phase also verifies that non-repeating fields have the same value in all denormalized rows because it is not possible to store more than one value. An incorrect design or selection of the flattening keys can result in this situation. If this occurs the load will issue warnings similar to this:

Warning: Illegal break in denormalized record n, field f: ’xxx’