Issue with dimension and partition processing

Let me start from the Cube structure:

1. DIMENSION:
   • D1 with attributes:

D1_id (primary key)

D2_id (foreign key)

• D2 with attributes:

D2_id (primary key)

1. MEASURE GROUP connection made to :
   • D1, relationship  type :regular  (D1_id)
   • D2 relationship  type: referenced  (D2_id)

Measure group include 12 partitions (monthly ones)

CUBE PROCESSING METHOD:

1) ProcessUpdate (dimension)

2) ProcessFull (last partition)

3) ProcessIndexe (all partitions except the last one)

Dimension D1 has been updated (D2_id was modified) and after processing we can see:

• Correct D2_id  if we use attribute from D1 dimension
• Correct D2_id for last partition ale old one D2_id for historical partitions if we use attribute from D2 dimension

Can anyone explain to me why historical partitions show wrong D2_id if D1_id hasn’t changed?