Nice twist on that concept! I've also seen a parameter and calculated field used where flipping the parameter value after initially generating the empty extract allows the full extract to occur on the server.
Having encountered this Out of Memory issue, I tried both the date exclusion approach and the USERNAME() approach, and neither is working. This is rather vexing as I've already successfully published the data source to Server once. Now I want to modify it, but I cannot successfully create the extract again on my desktop. The extract includes a fairly complex custom SQL statement on an Oracle database (which is what I'm trying to modify), but pulls only a few thousand rows.
Now that it has been published as a data source, it is also causing trouble on the server -- it refuses to refresh. The Oracle error it's throwing is ORA-12170 - TNS:Connect timeout occurred.
I'm at a loss.