Raking ratio estimation - choice of counts?

I am trying to use a raking ratio estimation in some survey analysis.

I have three variables that I'll be using - variable 1 and variable 2 (both demographic type information) which have no missing data and on which I'll complete the raking procedure. And variable 3 which I'll then apply the weights to for my estimation, and which DOES have missing data.

My question is: when completing the raking procedure, do I use the survey counts based on all units in the sample or just those units for which I have complete data for variable 3 as these are the values what I'll be applying the weights to? (even though the information for variables 1 and 2 is available for the item non-response also).

Thank you in advance!