...
As mentioned above, the Parallel Lookup stage does not accept user-specified expressions for key lookups. S2PX therefore injects a pre-lookup Prepare Transformer stage which pre-calculates any required Key Expressions and supplies them as one or more ephemeral column(s). These columns are only used to support the Lookup functionality and aren't propagated downstream once they have been used. S2PX also generates a post-lookup Logic Transformer stage which calculates other Transformer Stage Variables and output link derivations as required. These Transformers also perform pre- and post-processing required to support the S2PX translation of Hashed Files - See Hashed File Lookups for more details.
Another difference between Server and Parallel Jobs is that Server Jobs support multiple sparse lookups per Transformer stage, whereas the Parallel Lookup stage doesn’t. When a Server lookup has multiple reference sources then S2PX translates these into individual Parallel Lookup stages, each of which supports a single sparse lookup source. This also provides the option for users to easily toggle each generated Lookup between in-memory and sparse lookup modes as part of any post-conversion tuning activity.
...
Server Hashed Files are converted to Parallel DRS Stages
The Hashed File data are loaded into an underlying relational database which is subsequently accessed by Parallel Jobs using the DRS stage
Those data are stored using a special S2PX-specific format which is necessary to align with Server’s somewhat flexible approach to Hashed File metadata!
An important issue this introduces is that the specially-formatted data retrieved from the Hashed File-derived database tables will need to be pre-converted before they can be used in a Parallel Lookup (which does not permit Key Expressions). The requirement to introduce leading ! characters (used as a prefix for non-null fields) and @ characters (indicating Nulls) in order to ensure lookup keys match one another is another responsibility of the Prepare Transformer, introduced above. These temporary changes are then removed in the Logic Transformer, restoring the input data to its original format. See Using Hashed-File database tables in lookups for more information.
Sparse vs. in-memory
...