And what are the benefits of the insert only approach?
Well, since the main store is compressed (which is good for reading performance, detailes in the next blog)
changes would be expensive. Just imagine you want to make a change in a compressed text file. You need
to uncompress, make the change and compress again. For this reason data is only appended to the
uncompressed delta store where change operations are much cheaper. The merge moves the data from
delta to main and thus compresses only from time to time and not with every change. Setting the deletion
flag is based on rowids and works for compressed and uncompressed data.
In the next blog we will dive a little bit deeper in the column store and discuss the main and the delta in more