Hi,

While Lewis and I were discussing over NUTCH-1205, we identified
the Properties object as the major source of trouble/confusion when
configuring datastores. First and foremost, it's makes no sense that we
have 2 ways to configure a store, namely via Configuration and Properties.
Besides this there seems to be some trouble with the serialization and
initialization of the stores. (Sometimes runtime Properties settings are
not correctly used). We have a few ways to solve this problem:

-Stop supporting the adding of dynamic properties (runtime settings) and
only support the static gora.properties file. People wanting to use runtime
props shall use Configuration somehow.
-Completely remove the Properties object from Gora altogether. Migrate
existing properties to Configuration.
-Trying to make both Properties and Configuration work (that's what the
current direction seems to be). Difficult it seems.
-Something else?

I think the second option is best. The advantage of Configuration is that
it inheritently works with mapreduce because it is automatically
(de)serialized and available in the mappers and reducers. What do you think?

Ferdy.

Reply via email to