For example we are loading a fairly large dataset over a slow network
connection. When doing development, the data is essentially static, so
it would be nice to be able to spool the dataset to disk once, and
then read it from there, until some flag is reset in the configuration
to re-read the data from the real datasource.