Sorry so this is talking only about the delta storage
Let's say a table has 1000 rows and 10 columns.
3 columns are loaded in memory
Now some one insert 5 rows into this table
Now the delta storage will have 5 rows, but all 10 columns for this rows are in the delta storage. (so table in memory is partially loaded, but delta is fully loaded)
Similarly lets say a table has 1000 rows and only 3 of 10 columns are loaded in memory
And some one modifies 1 column of all 1000 rows, then the whole table is fully loaded in delta storage
And lets say if HANA DB is restarted and again the 3 columns are required for a query, since all of them is in delta, the table in memory will be partially loaded with the 3 columns and delta will be fully loaded with 10 columns
This is my understanding so far