I think we have a bug in Cursor.executemany() method. Given the code it runs Cursor.execute() in a cycle, while execute() is actually replacing result of every previous call with a result from the last one.

So, if we, for example, will call executemany() with several reads, we'll get only results of the last one:

But in fact both reads, with SingerId=12 and with SingerId=15, were executed and we got results.

@c24t, we already have a proposition of how to deal with executemany() in the design doc, PTAL at StreamedManyResultSets section.